Psychedelics relax predictive processing in the post-acute period by remodeling cortico-cortical feedback circuits

Serotonergic psychedelics such as psilocybin, LSD and 5-MeO-DMT are of growing interest because they may produce therapeutic effects that last days to weeks after a single dose. Earlier research had already shown that these drugs can induce structural plasticity in medial prefrontal cortex, including dendritic spine growth, but it remained unclear what those changes mean for circuit function after the acute drug experience has passed. In particular, the authors note that existing work had not directly linked psychedelic-induced plasticity to a lasting change in hierarchical predictive processing, especially in a way that could be studied in both humans and animals. West and colleagues set out to test whether recent psychedelic exposure relaxes predictive processing by weakening top-down cortical influence on sensory processing. Using a translational visual predictive-processing paradigm in humans and mice, they asked whether recent use of serotonergic psychedelics would reduce context-dependent suppression of responses to predictable stimuli, and whether this would be accompanied by changes in cortico-cortical feedback circuits and spine plasticity in mice. The study was designed to examine the post-acute period, rather than the acute psychedelic state, so that any observed effects would be more relevant to longer-lasting neurobiological change.

The study combined human behavioural/EEG experiments with mouse electrophysiology and histology. In humans, the researchers recruited adults from university advertising, social media and local psychedelic community groups. Eligible participants were 18 years or older and did not have autism spectrum disorder, bipolar disorder, schizophrenia, current major depressive episodes or substance abuse other than nicotine. The main human sample comprised 47 individuals: 16 controls with no recent psychedelic use, and 31 recent users. Within the psychedelic-exposed participants, 16 had recently taken psilocybin or LSD within 21 days, and 12 had recently inhaled Incilius alvarius toad toxin, which the authors treated as a 5-MeO-DMT source. Dosage could not be directly verified, though participants were screened to exclude only sub-perceptual microdoses. Human participants completed a saccadic prediction task in a dark room while eye movements were recorded with electrooculography and cortical activity with 32-channel EEG. The task involved repeated predictable stimulus locations, unexpected deviant locations, and deviant trials that either did or did not require updating expectations. Participants completed two runs, and the researchers also analysed a passive visual oddball EEG task in the same cohort. EEG data were processed in the time-frequency domain, with analyses focused on early and late post-stimulus windows and on conventional frequency bands, especially delta and theta. The authors also examined inter-trial phase locking, which indexes how consistently neural activity aligns to stimulus onset across trials. Group comparisons used mixed ANOVAs, with follow-up tests for significant interactions. In mice, the authors administered psilocybin (1 mg/kg, intraperitoneally) or saline to awake head-fixed animals and examined visual predictive processing 24 hours later, in the post-acute period. Local field potentials were recorded from primary visual cortex (V1), and in some animals also from anterior cingulate area (ACa), using standard oddball and global-local oddball paradigms. These paradigms were designed to separate local deviance detection from broader context-based predictive suppression. The researchers focused on early post-stimulus activity, especially delta-theta bands, and used non-parametric Granger causality analyses to assess directionality of connectivity between ACa and V1. They also used retrograde labelling to identify ACa neurons projecting to V1 and quantified dendritic spine density and volume with confocal imaging and automated analysis. Mouse statistics were modelled with linear mixed-effects approaches, with mouse as a random effect and sex included as a covariate or fixed effect as appropriate.

In the human psilocybin/LSD cohort, recent psychedelic users differed from controls on the saccadic prediction task in ways consistent with reduced predictive modulation. Although both groups generally made faster saccades to predictable than to unexpected targets, controls showed a marked enrichment of very fast express saccades to predictable locations, whereas recent psychedelic users did not. The proportion of express saccades to predictable targets was lower in the psychedelic group than in controls (10.5% versus 4.3%; t(30)=4.29, p=0.0002), while no group difference was seen for deviant locations. The difference in express saccades to predictable targets correlated with time since last psychedelic use (r=0.532, p=0.033), suggesting recovery towards typical predictive control over time. By contrast, measures of slower updating after deviant trials were broadly similar between groups, and both groups showed the expected slowing after the “no update” deviant, indicating intact behavioural updating. EEG results matched the behavioural findings. In controls, predictable visual targets elicited stronger early delta and theta responses and greater phase locking differences than deviant targets, consistent with predictive suppression of early sensory processing. Recent psychedelic users did not show this early location-dependent suppression. The key group-by-location interaction was present in the early window for delta power and theta power, and also for early delta phase locking, but not in later time windows. Late delta and theta responses to deviant stimuli were present in both groups, suggesting that later deviance detection or prediction-error-related activity remained intact. The magnitude of early phase locking to predictable stimuli was inversely correlated with days since last psychedelic use (r=-0.670, p=0.0045). No reliable effects of update type were found in the EEG analyses. A second human cohort who had recently taken 5-MeO-DMT showed a similar pattern. Compared with controls, they also showed fewer express saccades to predictable locations and altered early delta power and phase locking, again without evidence that late processing of deviance or behavioural updating was disrupted. The authors report that these effects were present regardless of sex or age. In mice, psilocybin produced a similar post-acute effect in V1. Twenty-four hours after dosing, saline-treated mice showed the expected context dependence: responses to a deviant orientation were enhanced in the standard oddball, but suppressed in the global-local paradigm when the same stimulus was globally predictable. After psilocybin, mice still showed deviance detection, but this was no longer modulated by global context in the same way. In other words, the suppressive effect of predictive context was lost, while basic response to unexpected stimuli remained. The authors also report that a subset of human participants analysed with the passive oddball task showed intact deviance detection despite reduced predictive suppression in the saccadic task. The connectivity analysis in mice suggested a mechanistic correlate. Granger causality indicated an altered balance between top-down ACa-to-V1 and bottom-up V1-to-ACa influences in the delta band 24 hours after psilocybin, consistent with weakened top-down modulation. Finally, structural analyses showed that ACa neurons projecting to V1 underwent spine growth after psilocybin. Spine density increased in females and spine volume increased in males, indicating sex-dependent effects on the measured structural metrics. The authors state that functional effects themselves did not show clear sex-by-group interactions.

The authors interpret the results as evidence that serotonergic psychedelics relax predictive processing in the post-acute period by weakening top-down cortical influence on sensory processing. They argue that, in both humans and mice, predictable stimuli normally evoke suppressed responses relative to unexpected stimuli, and that this context-dependent suppression is reduced after psychedelic exposure. In humans, this was seen as a loss of the usual enhancement of express saccades and early EEG suppression for predictable visual targets. In mice, a comparable loss of predictive suppression in V1 appeared 24 hours after psilocybin. The authors view the accompanying reduction in ACa-to-V1 directional influence as consistent with a circuit-level mechanism for this effect. West and colleagues place these findings within predictive processing and the ReBUS framework, which proposes that psychedelics may loosen entrenched priors and bias top-down inference, thereby enabling a more flexible reassessment of incoming information. They suggest that the spine growth observed in ACa neurons projecting to V1 may be the structural substrate of reduced precision in higher-level priors or weakened effective connectivity, even though the precise relationship between increased spine growth and decreased long-range influence remains uncertain. They also note that similar post-acute weakening in connectivity has been reported in other cortico-subcortical networks after psychedelic treatment, which they say is compatible with a broader reorganisation of network function. The authors acknowledge several limitations. The human component was retrospective and between-subjects rather than randomised, so causality cannot be established from the human data alone. The psychedelic group differed somewhat from controls in health/safety risk-taking and cannabis use, although excluding frequent cannabis users did not change the main findings. The second psychedelic cohort using 5-MeO-DMT included some participants who had also used psilocybin recently, which may have blurred time-since-use relationships. They also note that the human and mouse paradigms were not perfectly matched, and that the study focused on visual processing as a model system, so the extent to which these findings generalise to mood, anxiety or addiction outcomes is still unknown. In terms of implications, the authors suggest that reduced top-down influence after psychedelics may help explain the increased cognitive flexibility reported in the post-acute period. They propose that future clinical studies should test whether visual predictive-processing measures correlate with treatment response and whether manipulating ACa-to-V1-related plasticity changes both circuit function and behaviour. They also note that their paradigm is relatively rapid and non-invasive, which may make it useful for translational work in clinical settings.