Psilocybin alters visual contextual computations

Psilocybin is a serotonergic classical psychedelic known to alter perception and large-scale brain dynamics, yet the underlying computational changes remain unclear. Contextual computations—how local sensory inputs are integrated with surrounding information—are ubiquitous across sensory and cognitive systems and have been mechanistically linked to phenomena such as divisive normalization and surround suppression in visual cortex. Earlier work has related contextual visual illusions, such as the Ebbinghaus illusion, to properties of primary visual cortex and to circuit-level mechanisms that could be relevant to broader cognitive effects of psychedelics. Aqil and colleagues set out to test whether psilocybin alters visual-contextual computations at both behavioural and neural levels, and to capture those changes with an explicit computational model. They used a randomised, double-blind, placebo-controlled crossover design administering placebo, 5mg and 10mg psilocybin, and combined psychophysical testing of the Ebbinghaus illusion with ultra-high-field (7T) fMRI mapping and population receptive field (pRF) modelling. The study aimed to determine whether psilocybin modifies contextual perception and contextual modulation of cortical responses, and whether a single-timecourse computational model could link these effects.

Participants were recruited via public online adverts and screened by questionnaire followed by in-person assessment. Exclusion criteria included age <21 or >55 years, no prior hallucinogen experience, prior adverse reactions to hallucinogens or MRI, personal or family history of psychiatric or neurological conditions, and ongoing medication or recreational drug use. Prospective participants also completed a preliminary scanning session to confirm normal fixation and eye movement behaviour; two candidates were excluded at this stage for abnormal eye movements. Eighteen participants (aged 22–45 years, eight female) were enrolled. Data from five participants were excluded because of scanner artefacts and one for excessive head motion, leaving twelve participants in the final fMRI analyses. The extracted text does not clearly report the final behavioural sample size for the psychophysics tasks. The experimental design was a randomised, double-blind, placebo-controlled crossover with three sessions per participant, each separated by at least two weeks. Doses tested were placebo, 5mg and 10mg of psilocybin. Behavioural testing used the Ebbinghaus illusion: participants fixated a central cross while two white circles were presented briefly (0.4 s) on an isoluminant background and indicated which appeared larger. Control trials presented circles in isolation; test trials presented one circle surrounded by larger circles to induce the illusion. A small dot-colour change task at fixation was performed during fMRI to maintain attention; eye movements were recorded throughout. Psychophysical data were modelled using a standard signal-detection framework: perceived radii of reference and embedded stimuli were treated as samples from normal distributions, with parameters for sensory noise (σ^2), contextual bias (δ) and lapse rate (p) to account for failures to process a stimulus. Parameters were estimated in a hierarchical Bayesian framework using the No-U-Turn Sampler (NUTS), a Hamiltonian Monte Carlo algorithm, implemented in pymc. Group- and participant-level parameter structure used an offset-based hierarchical parameterisation to improve sampling efficiency. Parameters were estimated separately for the three dose conditions and for with/without surround trials via a regression approach. fMRI acquisition used a 7T scanner with a standard drifting checkerboard bar pRF mapping stimulus: a high-contrast checkerboard bar sweeping across the central 10 degrees of visual field in eight directions. Raw scanner data were converted to NIfTI and stored in BIDS format; thermal denoising used the NORDIC algorithm and preprocessing (susceptibility distortion correction, coregistration, resampling to fsnative surfaces) was performed with fMRIprep v20.7. Responses were mapped to individual cortical surfaces and visual field maps were derived from polar angle and eccentricity estimates. Population receptive field modelling was performed with prfpy and prfpytools. Importantly, the haemodynamic response function (HRF)—the BOLD signal’s temporal shape linking neural activity to observed blood-oxygen-level-dependent (BOLD) responses—was allowed to vary by cortical location, participant and dose to separate putative neural changes from haemodynamic effects. The investigators also report cross-validated model comparisons that tested models incorporating neural changes only, haemodynamic changes only, both, or neither.

Behavioural results showed that psilocybin increased the magnitude of the Ebbinghaus illusion. Relative to true stimulus size, the measured illusion was 6% with placebo, 8.3% with 5mg psilocybin (p = 0.012), and 9.6% with 10mg psilocybin (p = 0.005). The authors additionally report percentage increases relative to placebo of 39% for 5mg and 59% for 10mg. Perceptual changes were specific to contextual trials: performance on control trials (stimuli presented without context) was not altered by psilocybin. The psychophysical modelling, which included lapse and noise parameters, indicated that the observed increase in the illusion could not be explained by changes in sensory noise or lapse rate alone. In the neuroimaging data, the central positive activation peak in the single-voxel/vertex timecourses was reported as near-identical between placebo and 10mg conditions, indicating preserved core stimulus-driven activation. However, surround suppression—negative deflections in the timecourse flanking the central positive peak—was systematically reduced in the 10mg condition. Representative reductions in surround suppression were documented in visual-field maps corresponding to V1, V2 and V3, while the amplitude of the central positive activation was not significantly altered. Timepoints with statistically significant differences versus placebo were identified using a two-sided Fisher permutation test with 1,000,000 permutations (p<0.01); consecutive significant timepoints were indicated. The investigators further report that intermediate doses (5mg and 10mg) produced systematic perceptual and cortical changes without significantly impairing fixation ability or simple task performance. Computational modelling using a pRF-based framework was able to capture the observed changes in cortical responses and link them to the behavioural alteration of contextual perception. Cross-validated model comparisons and haemodynamic fitting suggested that the alterations in cortical responses elicited by psilocybin could not be fully explained by concurrent changes in noise or haemodynamics, supporting an interpretation of altered contextual computations at the neural population level.

Aqil and colleagues interpret their findings as evidence that psilocybin alters contextual computations in human visual cortex and perception. They propose that the increased Ebbinghaus illusion and reduced surround suppression reflect a shift in how local sensory signals are combined with contextual information, and that a pRF-based computational model can parsimoniously account for both neural and behavioural changes. The authors place these results in a broader framework: visual spatial computations and mechanisms such as divisive normalization and surround suppression are ubiquitous across sensory and higher-order cortical areas, so alterations at early visual stages may have widespread effects on brain dynamics and subjective experience. The discussion situates the results relative to prior literature by noting links between contextual processing and psychiatric conditions, and by referencing preclinical work showing that psychedelics can reopen critical periods for contextual learning. The authors suggest that altered contextual computations could be a general mechanism contributing to the diverse effects of psychedelics and their therapeutic potential, for example in depression, which has been characterised as involving an impaired ability to integrate current input with context. Several limitations are acknowledged. First, the BOLD signal depends on both neuronal and haemodynamic factors, which psilocybin could affect; to address this, the investigators fitted HRF shapes at each cortical location and performed cross-validated comparisons of models that included neural and/or haemodynamic changes, reporting that haemodynamic explanations alone did not account for the results. Second, only intermediate doses (5mg and 10mg) were tested, lower than some prior neuroimaging studies that used doses up to 25mg; higher doses may produce task performance or fixation confounds. Third, psilocybin and its active metabolite psilocin bind multiple receptors (including 5-HT2A and 5-HT1A), and the study was not designed to dissociate receptor-level contributions; the authors note that future work could combine their paradigm with receptor-specific blockers. Finally, the study focused on spatial vision to provide a clear demonstration of altered contextual computations; the authors recommend extending this approach to other sensory and cognitive domains to test whether the observed computational change generalises.