This double-blind, placebo-controlled study (n=30) investigates the learning effects of psilocybin (up to 20 mg) in a probabilistic cue-reward task with emotional cues. It finds that psilocybin preserves learning effects, is non-inferior to placebo, and suggests higher exploratory behaviour. The 20 mg group showed significantly better learning rates than placebo.
Serotonergic psychedelics hold promise as a treatment modality for various psychiatric disorders and are currently applied in psychedelic-assisted psychotherapy. We investigated the learning effects of the serotonin receptor agonist psilocybin in a probabilistic cue-reward task with emotional cues in the form of neutral or fearful faces, presented either consciously or subconsciously. This study represents the first investigation into reinforcement learning with psilocybin. Across different dosages, psilocybin preserved learning effects and was statistically noninferior compared to placebo, while suggesting a higher exploratory behavior. Notably, the 20 mg group exhibited significantly better learning rates against the placebo group. Psilocybin induced inferior results with subconscious cues compared to placebo, and better results with conscious neutral cues in some conditions. These findings suggest that modulating serotonin signaling in the brain with psilocybin sufficiently preservers reinforcement learning.
Casanova and colleagues situate their study in the context of a renewed interest in serotonergic psychedelics as potential therapeutics for psychiatric disorders. The introduction summarises that psilocybin, a 5-HT–acting tryptamine found in “magic mushrooms,” alters perception, mood and self-referential processing and has been linked to neuroplastic changes in animals and symptomatic improvement in conditions such as major depressive disorder. The authors highlight gaps in knowledge about how psilocybin affects reinforcement learning in humans, and specifically whether its effects differ when emotional cues are processed consciously versus subconsciously. The study set out to test whether psilocybin alters reinforcement learning in a probabilistic cue-reward task that used fearful and neutral facial cues presented at durations intended to favour either subconscious (33 ms) or conscious (47 ms) processing. The investigators had three preregistered hypotheses: that psilocybin would impair learning from negative (fearful) stimuli by decreasing their salience, that psilocybin would enhance learning with neutral and conscious stimuli via neuroplastic effects, and that reduced top-down processing under psychedelics would make subconscious cues more disruptive to learning under psilocybin than under placebo. The authors frame the question as clinically relevant because reinforcement learning processes are implicated in psychiatric disorders and in therapeutic change.
Papers cited by this study that are also in Blossom
Nichols, D. E. · Pharmacological Reviews (2016)
This was a randomized, double-blind, placebo-controlled crossover study in 30 healthy volunteers (23 male, 7 female; mean age 29), all white and right-handed. Participants underwent two test sessions 14 days apart and received either psilocybin or placebo (mannitol) in counterbalanced order. Psilocybin doses were weight-adapted into three categories: 10 mg (<50 kg), 15 mg (<80 kg), and 20 mg (≥80 kg). Key exclusion criteria included personal or family history of major psychiatric illness, major medical conditions, seizure history, current psychotropic medication, and pregnancy or breastfeeding. Approximately 240 minutes after substance administration participants completed EmotLearn, a probabilistic cue-reward task programmed in Presentation®. On each measurement day participants completed four tasks (eight tasks total across the study), each of 60 trials. Each task paired one presentation-duration condition (subconscious 33 ms or conscious 47 ms) with one emotional cue type (fearful or neutral faces from a standard face database). On each trial a face cue was shown, then a pair of symbols; participants selected a symbol and received +10 CHF or −10 CHF feedback. One symbol in each pair had a 70% chance of reward (30% loss); the other symbol had the inverse probabilities. Monetary outcomes were virtual and not paid out. The design aimed to probe how emotional valence and awareness modulate probabilistic reinforcement learning under psilocybin versus placebo. Raw trial-level accuracy (proportion correct per trial) was extracted and smoothed with a Gaussian filter. The Shapiro–Wilk test indicated non-normality of the learning data. The primary inferential approach was a hierarchical Bayesian mixed-effects regression using a beta distribution for the bounded accuracy outcome. Models included fixed and random effects for substance, cue category (fearful vs neutral, subconscious vs conscious), dosage and interactions; inference used Hamiltonian Monte Carlo (No-U-Turn Sampler) implemented in Stan with four chains of 2000 iterations each, 1000 warm-up iterations discarded, yielding 4000 post-warmup draws. Pairwise marginal means and highest posterior density (HPD) intervals were used to assess credible differences (95% HPD). Nonparametric Mann–Whitney tests were used for specific between-condition comparisons (e.g. reaction times, psychometric scales) and α = 0.05 was the nominal significance level.
Overall learning performance did not differ reliably between psilocybin and placebo: mean accuracy across all trials was 0.729 under psilocybin (N=30, n=3600, σ=0.332) versus 0.745 under placebo (N=30, n=3600, σ=0.324). The Bayesian model found the overall group difference included zero in the 95% HPD interval (0 ∈ 95% HPD), interpreted as noninferiority of psilocybin to placebo for the primary learning measure. Trial-by-trial curves showed comparable learning trajectories from ~50% at trial 1 towards higher accuracy by trial 30; only trial 20 showed a statistically significant advantage for placebo (p<0.005, Mann–Whitney, multiple-comparisons corrected). Levene’s test suggested borderline greater variance under psilocybin (p=0.05002), interpreted as more switching/exploratory behaviour. When analyses were stratified by cue awareness, subconscious cues under psilocybin produced worse performance than placebo: mean accuracy for subconscious cues was 0.714 with psilocybin versus 0.739 with placebo (N=30, n=1800 each, σ≈0.33), and pairwise comparisons showed psilocybin–subconscious was credibly worse than placebo–subconscious and worse than the three other conditions (0 ∉ 95% HPD). Emotional valence analyses showed that fearful versus neutral faces yielded similar learning curves under placebo until ~trial 22, after which they diverged; psilocybin fearful performance did not differ credibly from placebo fearful (0 ∈ 95% HPD) but was credibly worse than placebo neutral (0 ∉ 95% HPD). Subgroup contrasts highlighted that subconscious neutral cues particularly hindered learning under psilocybin: psilocybin–subconscious–neutral mean accuracy 0.717 (N=30, n=900, σ=0.342) versus placebo–subconscious–neutral 0.757 (N=30, n=900, σ=0.327), a credible difference (0 ∉ 95% HPD). Within the psilocybin condition, conscious–neutral cues performed better than both subconscious–neutral and subconscious–fearful (0 ∉ 95% HPD for both comparisons); the placebo group did not show these intra-condition differences. Order effects in the crossover were notable: participants who received psilocybin on the first day had lower mean accuracy (μ=0.689, n=16) than those who received psilocybin after placebo (μ=0.776, n=14) and those who received placebo first (μ=0.781, n=14); these differences were credibly significant (0 ∉ 95% HPD). Dose-stratified analyses showed the 20 mg psilocybin subgroup (μ=0.74, N=15, n=540) outperformed the same participants on placebo (μ=0.725, 0 ∉ 99% HPD), whereas the 15 mg subgroup showed the reverse (psilocybin μ=0.717 versus placebo μ=0.747, 0 ∉ 99% HPD). The 10 mg group included only two participants, limiting inference. Intrapersonal comparisons across the crossover showed heterogeneity: 12 participants (40%) performed significantly better with placebo, 7 participants (23.3%) performed significantly better with psilocybin, and 11 participants (36.7%) showed no significant within-person difference (Mann–Whitney, p<0.05 threshold). Monetary earnings did not differ significantly (placebo 133.9 ± 9.9 CHF vs psilocybin 103.5 ± 12 CHF, p=0.09), though the mean was 29.3% higher under placebo. Reaction times were slower on psilocybin (mean 1.73 s, σ=2.11 s) than placebo (mean 1.34 s, σ=1.62 s; p<0.001, Mann–Whitney); within psilocybin participants the 20 mg group had faster reaction times than the 15 mg group (1.49 s vs 2.01 s, p<0.001). Psychometric scales showed marked subjective effects: the 5D-ASC Vigilance Reduction score averaged 44 under psilocybin versus 9.75 under placebo (p<0.001, Cohen’s d 1.97), and the 11D-ASC Impaired Control and Cognition subscale averaged 29.48 under psilocybin versus 0.74 under placebo (p<0.001, Cohen’s d 1.52).
The investigators interpret their results as indicating that psilocybin preserves reinforcement learning capacity in a probabilistic cue-reward task relative to placebo, while selectively altering how awareness of emotional cues affects performance. Both groups showed upward learning curves and comparable overall accuracy, supporting the claim that strategy finding and belief updating in an uncertain environment remain largely intact under the tested psilocybin conditions. However, subconscious cues—particularly neutral faces presented briefly—produced a credible decrement in performance under psilocybin compared with placebo, suggesting that altered top-down suppression of task-irrelevant input or disrupted attentional tracking may underlie the effect. Casanova and colleagues situate these findings within mechanistic accounts of serotonergic modulation and predictive coding. They discuss how psilocybin’s partial agonism at 5-HT2A (and modulatory effects at other 5-HT subtypes) may shift the balance of top-down versus bottom-up signalling, relax overly precise priors and increase bottom-up prediction error flow. Such changes could increase exploratory switching (consistent with the borderline higher variance under psilocybin) while also attenuating aversive processing in ways analogous to chronic SSRI effects. The authors compare their findings to prior work with LSD and ketamine, noting differences in receptor profiles and downstream dopaminergic or glutamatergic effects that may explain discrepancies in learning-rate findings across studies. The discussion acknowledges several limitations reported by the authors. First, the complexity of the dataset and subgroup hierarchies prevented fitting a formal reinforcement-learning computational model, so inference relied on a Bayesian mixed-effects regression. Second, the learning task was administered ~240 minutes after dosing, a time when acute subjective effects may be partially waning (the authors state acute physiological changes were present at approximately 50%). Third, the probabilistic learning paradigm may not capture the broader notion of therapeutic cognitive flexibility that psilocybin might facilitate in clinical populations. The authors also note order effects—poorer performance when psilocybin was administered first—interpreting this as practical evidence that beginning an unfamiliar task in a sober state may aid subsequent performance under psilocybin. Finally, they comment that subjective psychometric reports of impaired vigilance and cognition under psilocybin contrasted with largely preserved objective performance on the task, and that monetary incentives may have been less salient under psilocybin which could have affected earnings outcomes.
The authors conclude that, across the dosages tested, psilocybin preserved the ability to find strategies and make decisions in an uncertain probabilistic learning task and was statistically noninferior to placebo, with the 20 mg dose showing superior learning compared to placebo in their sample. They emphasise that subconscious emotional cues diminished learning under psilocybin, a finding they attribute to shifts in top-down versus bottom-up cortical–subcortical signalling. The study is presented as the first test of psilocybin in a probabilistic reinforcement-learning paradigm; the authors call for further research to clarify the mechanisms underlying therapeutic learning effects and to optimise dosing and administration protocols for clinical use.
Psilocybin (4-phosphoryloxy-dimethyltryptamine) is a serotonergic tryptamine hallucinogen structurally related to serotonin (5-HT; 5-hydroxytryptamine). It is found in "magic mushrooms" (e.g., Psilocybe cubensis) and has a rich history of ritual use in various cultures. When ingested, psilocybin is converted into its active form psilocin (4-hydroxy-dimethyltryptamine), which produces its mind-altering effects through various serotonin receptors and downstream effects on the GABAergic 4 , dopaminergic, and glutamatergic systems. Healthy humans ingesting psilocybin can experience profound changes in the sense of self, perception, mood, and cognition, including alterations in self-referential processing. In recent years, psychedelics have reemerged as promising treatments for various psychiatric disorders. Psilocybin and lysergic acid diethylamide (LSD) have been shown to significantly reduce symptoms of depression and anxiety in patients with major depressive disorder (MDD), anxiety, and substance use disorder. Interestingly, psychedelics have been shown to enhance synaptic neuroplasticity in animals, which might drive cognitive adaptations while facilitating therapeutic action through effects on learning and offer an original mechanism for the lasting beneficial outcomes in depressed patients. A further factor potentially contributing to therapeutic effects was that psilocybin decreases connectivity between brain regions involved in emotional processing, such as when recognizing angry faces. This shift in processing emotional information biases positive and neutral information over negative information, correlating with an increase in positive mood.
The psychedelic compound LSD was found to increase both reward and punishment learning rate in a learning task. By promoting a state of heightened plasticity and increased exploratory behavior, LSD led to enhanced rate of belief updating. These effects are thought to facilitate revision of maladaptive associations. While psychedelics have been demonstrated to acutely alter cognitive processing at different levels, such as attention, working memory, and goal-directedand prosocial behavior, little is known about the effects of psilocybin on learning, particularly concerning the role of conscious versus subconscious emotional processing. While LSD has a broader serotonin receptor binding profile, including activity at dopamine receptors, psilocybin primarily acts via partial agonism of 5-HT2A receptors in cortical and subcortical brain regions. Furthermore, it exerts modulatory effects on the 5-HT1A-receptorand others, including the 5-HT2B/2C/6/7-receptors. Although having no direct affinity for dopamine receptors, psilocybin has been found to elevate striatal dopamine (DA) concentration, with the 5-HT2A receptor hypothesized to activate the DA system. The mesocorticolimbic DA system plays a pivotal role in learning by capturing novelty through reward prediction errors (PE), measured by characteristic dopaminergic firing patterns. Increased DA firing is behavior-promoting whenever a reward is higher than expected (high striatal D1-receptor activation on direct pathway neurons leading to action promotion), while decreased firing rate is behavior-discouraging if a reward is less than expected (low striatal D2-receptor activation on indirect pathway neurons leading to action suppression). Serotonin is believed to maintain a delicate balance with DA in learning and behavior reinforcement. Studies indicate that depleting 5-HT levels can dysregulate the activity of DA neurons, leading to heightened responsiveness to immediate rewards and decreased inclination toward long-term goals. Conversely, increasing 5-HT levels with prolonged selective serotonin reuptake inhibitor (SSRI) administration can inhibit the activity of DA neurons, leading to a decreased emphasis on immediate rewards, promoting patience for future rewards and amplifying motivation toward long-term goals. To test the effects of a substance on learning, reinforcement learning can be applieda feedback-driven learning paradigm aimed at maximizing long-term rewards by learning optimal behavior through trial-and-error. Probabilistic learning tasks model behavioral dynamics in decision-making fraught with uncertainty, employing Bayesian inference to adapt priors continuously in uncertain environments, steering behavior toward desired outcomes. Agents simultaneously juggle uncertainty while associating stimuli with affective values, thereby enforcing associations between an unconditioned stimulus (a reward) and a conditioned stimulus (a cue). Emotional faces serve as cues in learning tasks, influencing the emotional valence of stimuli and the processing and retention of information. Emotional cues typically enhance learning, with negative cues (e.g., fearful faces) yielding additional improvement compared to neutral cues. However, the duration of presentation of cues also plays a role. Some studies observed subconscious cues to impede learning performance compared to conscious cues, demanding more cognitive resources for processing. Dysfunctions in reinforcement learning have been implicated in a range of psychiatric disorders, including addiction and major depressive disorder (MDD), whereas prolonged administration of SSRIs in MDD was associated with reduced reinforcement learning by blunting emotional responses, thereby impairing learning from negative stimuli while showing negligible effects on learning with positive stimuli. As emotional blunting is not observed with psilocybin in contrast to SSRIs, J o u r n a l P r e -p r o o f we sought to compare the extent of valence-specific learning effects against placebo. In particular, in this study, we investigate the potential therapeutic implication of psilocybin on learning by comparing it to placebo in a probabilistic cue-reward task with subconscious and conscious emotional cues. We had three hypotheses. First, we hypothesized that by decreasing the salience of learning-enhancing fearful faces, psilocybin would impede learning with negative stimuli. Second, based on the observed neuroplastic effect in animals but not yet studied in humans, we anticipated that psilocybin enhances learning with neutral and conscious stimuli. Lastly, we speculated that due to a reduction in top-down processing with psychedelics, subconscious cues, which demand greater attention, disrupt learning to a higher degree with psilocybin than with placebo.
Our measure of learning consisted of the positive change in accuracy (from 0 to 1) when determining the more lucrative symbol out of two in a learning task (Fig.). We calculated the accuracy in each task for each substance and plotted its change as a learning curve (Fig.). At the beginning of each task, both symbols held equal uncertainty, leading the learning curve to commence at approximately 50% accuracy at trial 1 and progressing until trial 30, reflecting a learning effect. Trial-to-trial comparisons showed no significant difference between psilocybin and placebo, except trial number 20 favoring placebo (p<0.005, Mann-Whitney-test, corrected for multiple comparisons). As a main result, the learning effect between psilocybin and placebo across all conditions did not reach significance (0 ∈ 95% HPD) (Fig.). Psilocybin led to a mean accuracy score over all trials of 0.729 (N=30, n=3600, σ=0.332), while the overall placebo accuracy score was 0.745 (N=30, n=3600, σ=0.324). The Hodges-Lehmann estimator indicating the median of all pairwise differences was 0, while the effect size (Cliff's delta) was 0.023, faintly favoring placebo. Moreover, in the Levene's Test for Homogeneity of Variance, the variance of psilocybin showed borderline significance (p=0.05002, N=30, n=7200) compared to placebo, suggesting a higher amount of switching and a greater flexibility in responses in the psilocybin group.
We calculated the pairwise marginal means from fitted models of fearful vs. neutral faces and subconscious vs. conscious presentation of cues. The mean accuracy score over all trials for subconscious cues with psilocybin was 0.714 (N=30, n=1800, σ=0.337), whereas for placebo it was 0.739 (N=30, n=1800, σ=0.327). The accuracy improvement of cues with different presentation times were plotted in a learning curve (Fig.). Regarding presentation times, psilocybin's performance with subconscious cues was significantly worse not only against placebo but also against the other three conditions (0 ∉ 95% HPD for all three) (Fig.). Similarly, we plotted the learning curve of emotional cues. With placebo, fearful and neutral cues led to a similar learning curve up to trial number 22, where they start to diverge (Fig.). The mean accuracy score for psilocybinneutral was 0.737 (N=30, n=1800, σ=0.332), whereas for placebo-fearful it was 0.754 (N=30, n=1800, σ=0.333). With emotional J o u r n a l P r e -p r o o f cues, psilocybin's performance with fearful faces was not significantly different against placebo (0 ∈ 95% HPD), while being significantly worse to placebo-neutral (0 ∉ 95% HPD) (Fig.).
When comparing the performance of psilocybin and placebo, subconscious neutral cues with psilocybin led to a mean accuracy of 0.717 (N=30, n=900, σ=0.342), while for placebo 0.757 (N=30, n=900, σ=0.327) with a statistically significant higher performance of placebo (0 ∉ 95% HPD). The other combinations of emotions and duration of presentation did not reach statistical significance (0 ∈ 95% HPD). Performance comparison among the psilocybin group with different cue-pairs led to consciousneutral cues faring better than both subconscious-neutral and subconscious-fearful group (0 ∉ 95% HPD for both), while the placebo group did not show statistical difference (0 ∈ 95% HPD). Moreover, the comparison of mixed cues led to the following results: placebo-subconscious-fearful fared worse than psilocybin-conscious-neutral (0 ∉ 99% HPD). Psilocybin-subconsciousfearful was inferior to placebo-conscious-neutral, placebo-conscious-fearful, placebo-subconscious-neutral, and psilocybinconscious-neutral (0 ∉ 95% HPD for all four) (Fig.&).
Participants receiving psilocybin on the first day (μ=0.689, n=16, N=3840) achieved significantly worse results than both those receiving psilocybin after placebo (μ=0.776, n=14, N=3360) and placebo on the first day (μ=0.781, n=14, N=3360) (0 ∉ 95% HPD for both) (Fig.&).
We compared the dose-dependent performance with a Bayesian mixed-effects regression model with a beta distribution. We split participants into three dosage levels and compared their scores with psilocybin and placebo. Although dosages were weightadapted into three categories, the 20 mg psilocybin group (μ=0.74, N=15, n=540) significantly outperformed the same group receiving placebo (μ=0.725, N=15, n=540) (0 ∉ 99% HPD). The 15 mg psilocybin group (μ=0.717, N=13, n=390) had the opposite result, with significantly inferior scores compared to the same group receiving placebo (μ=0.747, N=13, n=390) (0 ∉ 99% HPD). While the 10 mg group counted only two participants (N=2, n=60) and the analysis of that dosage level is limited, the higher dosage of psilocybin fared better against their placebo counterpart (Fig.).
Comparing the randomized crossover intrapersonal results of each participant's performance against themselves of psilocybin vs. placebo, 12 participants (40%) achieved a significantly higher accuracy with placebo (p<0.05, Mann-Whitney test), while 7 participants (23.3%) performed significantly better with psilocybin (p<0.05, Mann-Whitney test), and 11 participants (36.7%) did not show a significant difference between the two substances (p>0.05, Mann-Whitney test).
The monetary reward between the psilocybin and placebo groups did not reach significance (p=0.09, Mann-Whitney test). The placebo group managed to earn 133.9 ± 9.9 CHF (mean ± standard error), while the psilocybin group reached 103.5 ± 12 CHF. Although the monetary gain was fraught with probabilistic undulations, with placebo participants gained 29.3% more money than with psilocybin (Fig.).
Regarding the reaction times, on psilocybin, we observed a mean of 1.73 s per trial (σ = 2.11 s). On placebo, we observed a mean of 1.34 s (σ = 1.62 s). Placebo led to significantly faster reaction times (p<0.001, Mann-Whitney test). The Hodges-Lehmann estimator found that on average, psilocybin led to 0.36 s longer reaction times. Interestingly, the reaction times of the 20-mg group were statistically faster than those of the 15-mg group (1.49 s vs 2.01 s, p<0.001, Mann-Whitney test).
In psychometric scales, the dimension Vigilance Reduction of the 5D-ASC scale placebo reached a mean score of 9.75 compared to 44 under psilocybin (p<0.001, Mann-Whitney-test), a Cohen's d of 1.97, and a 0.32 Pearson correlation coefficient. In the 11D-ASC questionnaire, the dimension Impaired Control and Cognition, placebo had a mean score of 0.74 compared to 29.48 in psilocybin with a significant difference (p < 0.001, Mann-Whitney-test), a Cohen's d of 1.52, and a 0.1 Pearson correlation.
In this study, we investigated the effects of psilocybin on strategy finding in a probabilistic learning task next to the impact of various cues on awareness. We showed that psilocybin did not impede reinforcement learning by modulating serotonin signaling in the brain against placebo and that subconscious cues disturbed performance with psilocybin. As participants traversed trials, selecting the symbol associated with monetary gain resulted in an upward trajectory of the learning curve. Both the psilocybin and placebo groups exhibited a learning curve at a comparable rate, without reaching statistical difference. Of the thirty trials only one was significant in favor of placebo, which we attributed to random chance. Despite being a powerful hallucinogen that alters thought patterns and emotions, the capacity to navigate an uncertain environment and to update beliefs effectively prevailed in different tested dosages of psilocybin. Emotional cues from fearful faces signal unconditioned threat-related stimuli and have led to a performance boost in other studies by serving as a conditioned predictive stimulus in cognitive processing. Contrary to previous findings, in our task, fearful faces did not contribute to an enhancement in performance of placebo or psilocybin, and along neutral faces, exhibited similar learning rates when comparing placebo vs. psilocybin. In other studies, acute effects of psilocybin were found to diminish perception of negative emotions due to reduced activity in right amygdala activity, while simultaneously correlating with increased positive mood, and leading to 5-HT-mediated DA enhancement, facilitating fear extinction learning. In depression, there is a bias of processing negative over positive stimuli by deficient 5-HT activity. Similarly, this reduced perception and J o u r n a l P r e -p r o o f weakened amygdala blood-oxygen-level-dependent (BOLD) response of fearful faces was observed with both acuteand repeated SSRI administration. The weakened recognition of fearful faces was blocked by the 5-HT2A/2C-receptor antagonist ketanserin, indicating the role these receptors play in mediating these effects. An increase in 5-HT transmission (e.g., prolonged SSRI administration) results in attenuated aversive processing and a reduced BOLD response of the amygdala to fearful faces, while reduced 5-HT levels (e.g., acute tryptophan depletion, acute SSRI administration) augment aversive processing, heighten fear recognition, and diminish reward learning. Consequently, 5-HT depletion leads to heightened reactivity to aversive signals, prioritizing punishment over reward accompanied by increased impulsivity. While the acute effects on the amygdala of psilocybin resemble those of prolonged SSRI administration, the latter have been found to impair learning by blunting emotions, whereas psilocybin enhances the perception of positive emotions. Psilocybin resulted in 5% higher variance difference than placebo, which was borderline significant, suggesting a slightly higher amount of switching under psilocybin and pointing at a higher exploratory behavior. Similarly, LSD was shown to decrease stimulus stickiness in a learning task, hinting to an increased flexibility in decision-making of serotonergic psychedelics. This resembles acute single SSRI administration, which reduced 5-HT transmission in a probabilistic learning task and led to a higher tendency to inappropriately switch a strategy after punishment . Thus, in our study, we observed effects of psilocybin in line with both acute (inappropriate switching) and chronic (reduced aversive processing) SSRI administration, pointing at mixed effects possibly due to its partial agonism at different 5-HT receptors. Another role of serotonergic transmission in learning is that 5-HT is thought to interact with DA neurons by encoding the beneficialnessof a given action or outcome through promoting goaldirected behavior, assessing the level of satiation, and preventing impulsive and appetitive actions while anticipating a reward. Thus, 5-HT may be indicative of temporal discountingand signal an individual's motivation to either maintain or switch to a different behavior based on the perceived benefits of each option. Higher 5-HT levels led to less impulsive actions and increased acceptance of a delayed reward, compared to a bias toward smaller immediate rewards with lower 5-HT levels. Regarding the learning dynamics of varying presentation times, conscious cues corresponding to the cortical pathway notably engendered better results under psilocybin than cues presented subconsciously via the subcortical pathway. This result was reflected in other studies speculating that subconscious cues may evoke a feeling of uncertainty or anxiety, yielding less ideal choices. Subgroup analysis expanded on this result. Although subconscious cues overall hindered learning with psilocybin, it was only the neutral cues presented subconsciously under psilocybinnot subconscious fearful onesthat led to worse performance compared to placebo. Moreover, despite fearful and neutral cues equating performance in the psilocybin group, only conscious neutral cues were superior to both subconscious emotional modalities with psilocybin, pointing at a destabilizing performance of fearful faces with psilocybin in the conscious condition and a favorable in the subconscious condition. The remaining results, which compare mixed pairs (e.g placebo subconscious neutral against psilocybin conscious fearful) are speculative, since comparison might not be straightforwardly applicable. In informational processing, task-irrelevant visual information is initially processed in visual areas and subsequently suppressed through top-down attentional processing. Such a process demands more attention to accommodate subconscious J o u r n a l P r e -p r o o f cues accordingly. Reduced top-down processing under psilocybinmight attenuate the suppression of task-irrelevant visual information, leading to a larger learning impairment of subconscious cues. 5-HT activity at the subcortical level could induce motivational processes opposed to those mediated by DA by increasing punishment PE . However, in our study, only participants under psilocybin exhibited a significantly superior learning effect with conscious cues compared to subconscious cues and were more adversely affected by subconscious cues than those receiving placebo, hinting at a possible serotonergic distinction in these pathways. For example, the 5-HT1A-receptor is reputed to be involved in disturbing attentional tracking, while the 5-HT1A and 5-HT2Areceptors exert opposite actions in the mPFC. These contrasting findings point at the multifaceted pharmacological actions of psilocybin leading to substantial alterations in awareness and self-related cognitive processes in cortical-subcortical functional networks, such as within the prefrontal cortex (PFC), posterior cingulate cortex (PCC) and reducing other regions across the default mode network (DMN). Moreover, the 5-HT2A receptor is the main receptor activated by psilocybin and the predominant 5-HT subtype in the cortex and the DMN, especially in the PFC. Predictive coding indicates that top-down signaling is linked with prior expectations of lower-level neural activities, while bottomup signaling conveys PE. Psychedelics are believed to relax potentially pathologically overweighted priors by decreasing the top-down inhibitory control of the PFC and PCC, while simultaneously enhancing bottom-up information flow. These alterations correspondingly manifest a richer phenomenological experience and facilitate therapeutic insights by enabling new perspectives. A discrepancy of either top-down or bottom-up mediated signaling could result in aberrant behavior leading to learning impairments. Dysfunctional top-down signaling could distort sensory experience, up to possibly inducing hallucinations in the absence of a stimulus. Meanwhile, deviations in bottom-up signaling may indicate imprecise prior beliefs, requiring adaptation to accurately model the current experience of reality. The NMDA receptor antagonist ketamine was found to disrupt top-down control of prior expectation signaling in corticohippocampal circuits, leading to dysfunctional and overly precise bottom-up prediction error signaling, manifesting a failure in sensory attenuation and false inference cycles. In studies examining the effect of psychedelics on sensory stimuli, ketamine was shown to disrupt the generation of auditory event-related potential (ERP) mismatch negativity (MMN) and impair performance in a continuous performance task. The MMN reflects changes in PE and is a measure of learning. In the same task, psilocybin exhibited similar performance deficits while preserving MMN and lowering ERP in N100. Additionally, another study reported analogous disruptions in N100 and P300 due to psilocybin. This points to specific roles of the NMDA receptor in impairments of MMNand the 5-HT2 receptor in ERP alterations. The observed performance deficits in the continuous performance task for both substances were hypothesized to originate from shared downstream dopamineand excessive glutamate release. With tactile stimuli, psilocybin was found to weaken both the MMN response to surprising stimuli and the integration sensory inputs via aberrant PE processing, associated with alterations in subjective body and self-experience. Since the P300 amplitude correlates with learning, and psilocybin was found to dose-dependently reduce P300, it would be expected to see reduced learning effects with increasing dosages of psilocybin. However, we observed no significant reductions in learning against J o u r n a l P r e -p r o o f placebo, and notably, we even observed enhanced learning with a 20 mg dosage of psilocybin compared to placebo. In summary, psilocybin-induced 5-HT2-receptor signaling generally maintains MMN for most tested stimuli, except for a decrease observed with tactile inputs. This reflects a sufficiently functional learning system in line with our findings of preserved learning in a probabilistic learning task. Moreover, alterations in cortico-subcortical signaling might reflect our findings of altered learning effects of conscious and subconscious cues. Similar to previous findings, psilocybin led to slower reaction times than placebo. This effect was more pronounced for negative and neutral stimuli but not for positive stimuli. Surprisingly, when comparing the learning rate of the 20 mg group of psilocybin not only did it perform significantly better than the placebo counterpart but also exhibited faster reaction times than the 15 mg group of psilocybin. In a previous study, LSD exhibited both an increased punishment and reward learning rate compared to placebo in a learning task. While we observed learning improvement only with 20 mg dosages of psilocybin against placebo, the lack of difference in other variables might be due to our task specificity, the broader pharmacological profile of LSD(additional action at dopaminergic and adrenergic receptors), or the disparity in the used model. When considering the order of substance administration, starting the task with psilocybin yielded less favorable results compared to both commencing with placebo or receiving psilocybin in the second session. This could imply higher initial uncertainty with psilocybin that diminished as the task progressed and suggests the importance of starting an unfamiliar task in a sober state. Thus, participants who received placebo first might have benefited from an initial ordinary cognitive experience, better preparing them for the subsequent session with psilocybin. In contrast, participants who received psilocybin first had to adapt to performing the task while sober. The monetary reward did not reach statistical difference between psilocybin and placebo. Despite used as a main incentive for determining the correct choice, the total sum of money earned in the task did not pose a reliable measure of learning due to participants being able to earn money with a 30% chance even when selecting the incorrect symbol, respectively facing a 30% chance of loss with the correct symbol. Another reason for the difference could be that some participants noted psilocybin diminished the allure for monetary gain as money lost its importance and the pleasure derived from monetary gain was diminished compared to an ordinary waking state. In a future study, a more suitable incentive for psilocybin might be used to assess learning effects. Although the psychometric evaluation of the psilocybin group did report a subjective impairment of vigilance and cognition, their objective performance did not reflect this, as performance did significantly differ from placebo. This suggests that the underlying mechanism of decision-making and strategy finding sufficiently prevails in a probabilistic learning task, contrary to psychometric self-assessments. In contrast, an opposite pattern is seen with nootropics or neuroenhancers, where individuals report subjective cognitive improvement without objective enhancement. Of our hypotheses, we showed that first, fearful faces might impede learning with psilocybin only when consciously perceived. Second, conscious and not neutral stimuli led to better results with psilocybin. Lastly, subconscious cues led to a worse performance with psilocybin than with placebo, due to a reputed higher disruption of attention with psilocybin.
The manifold potential applications for psilocybin in treating psychiatric conditions involves a complex interaction between neurobiological changes and psychological insightsamong others through the interplay of 5HT2A-agonism, neuroplasticity, relaxed rigid beliefs, and decreased DMN activityleading to a revision of maladaptive thought patterns. Our study was the first to test psilocybin in a probabilistic learning task and to explore some of these dynamics. We found that the awarenessaltering effects of psilocybin with all tested dosages adequately preserved the capacity for strategy finding and decision-making against placebo, especially with a higher dosage. This suggests that learning in an uncertain environment under therapeutic sessions of psilocybin can sufficiently take place. Moreover, subconscious cues diminished learning with psilocybin, likely stemming from a hypothesized shift in top-down and bottom-up mediated signaling with psychedelics. Further research is needed to better understand the mechanisms underlying therapeutic learning effects and to determine the optimal dosing and administration protocols for psilocybin and other psychedelics for this purpose.
The limitations of the study were first, the complexity of the design and the amount of data measuring different hierarchical subgroups did not allow for the implementation of a proper reinforcement learning model, leading to a Bayesian mixed-effects regression model instead. Second, performance was measured approximately three to four hours post-substance intake, where acute physiological psychometric changes only prevail at approximately 50%. Third, the effects of psilocybin contributing to cognitive and neural flexibility in patients suffering from depression revealed a surprising lack of correlation between improvement of depression and enhancement of cognitive capabilities. It is speculated that heightened flexibility, a major part of the therapeutic effects of psilocybin, creates a window of plasticity facilitating improvements, akin to escaping a dysfunctional rigid state (by flattening local minima as suggested in the REBUS model). This aspect is potentially not fully captured in a probabilistic learning task. Fig.The learning rate of psilocybin is noninferior to placebo. To introduce learning in uncertainty, the probabilistic component of the task returned a positive reward in 70% for the more lucrative symbol and a loss in 30% of the time, while for the less lucrative symbol, the percentage was reversed. Each subject solved eight tasks in total. Four on each measurement day (psilocybin or placebo). Each task consisted of sixty trials evenly distributed among two symbol pairs, leading to a total of 240 data points per trial and per condition. The conditions were a tupel of (subconscious, fearful), (subconscious, neutral), (conscious, fearful), and (conscious, neutral) (A). In the first trial, the more lucrative symbol was chosen with an accuracy of approximately 50%. Progressing through trials, the accuracy of choosing the "better" symbol increased logarithmically, reflecting a learning effect. Each learning curve was constructed from 30 subjects completing four tasks, each involving two symbol pairs, resulting in a total of n=240 per trial and substance and n=7200 per substance. (B) Overall learning curve with logarithmic smoothing of placebo versus psilocybin. Psilocybin is shown in blue and placebo in grey. The mean on each trial is depicted with standard errors. Only trial number 20 showed a significantly better result in favor of placebo. (C) Psilocybin did not significantly differ from placebo. Coefficient plot of placebo vs. psilocybin reflecting a lack of overall significant difference (0 ∈ 95% HPD). Asterisks denote statistical significance (*p < 0.05). Fig.: Learning curves of different conditions. Learning curves of subdivided conditions (duration of presentation, emotional cues), comprising 30 subjects solving two tasks with two symbol pairs, leading to n=120 per trial and condition to a total of n=3600 per condition. Psilocybin is shown in shades of blue and placebo in shades of grey. (A) Learning curve of subconscious compared to conscious cues with a (B) coefficient plot of pairwise marginal mean differences of direct comparisons at the top, showing significance between subconscious cues in favor of placebo, and psilocybin conscious and placebo conscious over psilocybin subconscious (0 ∉ 95% HPD respectively). (C) Learning curves of neutral and fearful cues with a (D) coefficient plot of pairwise marginal mean differences of direct comparisons at the top, showing significance between placebo neutral over psilocybin fearful (0 ∉ 95% HPD respectively).
Bar graphs of subgroups with standard errors (each with N=30, n=1800). Direct comparison led to subconscious neutral cues with psilocybin faring worse compared to placebo subconscious neutral (green, 0 ∉ 95% HPD), while group comparison with psilocybin led to conscious neutral cues performing better than both subconscious neutral (orange, 0 ∉ 95% HPD) and subconscious fearful (purple, 0 ∉ 99% HPD). (B) Coefficient plot of the pairwise marginal means of the performance comparison of direct subgroups, substance groups and mixed pairs, with significant differences highlighted in red. Legend: Pla = placebo, Psi = psilocybin, Con = conscious (47 ms), Sub = subconscious (33 ms), F = fearful, N = neutral. Asterisks denote statistical significance (*0 ∉ 95% HPD, **0 ∉ 99% HPD). Fig.: Placebo earned more money, and performed better when administered first, while higher dosage of psilocybin led to better results. Receiving placebo first led to higher scores. The group receiving psilocybin first scored significantly worse than those receiving placebo first and those psilocybin after placebo (0 ∉ 95% HPD) (A) and bar graph reflecting the difference in scores (B). Higher dosage of psilocybin led to higher scores than placebo. Coefficient plot of pairwise marginal means of different dosages (C). The 20 mg psilocybin group had significantly higher scores than the same participants receiving placebo (0 ∉ 99% HPD). The 15 mg placebo group achieved a significant score over the 15 mg psilocybin group (0 ∉ 99%). Although only limited analysis is possible due to the small sample size (N=2, n=60), the 10 mg-placebo group was superior to the 10 mg psilocybin group. Placebo led to higher earnings than with psilocybin. Boxplot of average earnings per task of psilocybin vs. placebo (D). With psilocybin, participants earned almost one-third less on average than placebo, although this difference did not reach statistical significance (p=0.09, Mann-Whitney-test, n=30, N=7200 per substance). Moreover, the total amount of monetary gain was fraught with probabilistic undulations, since participants were also probabilistically able to earn money by choosing the inferior symbol. The horizontal bar shows the median, and the black
Further information and requests for resources should be directed to and will be fulfilled by the Lead Contact, Andrea Casanova (andrea.casanova@bli.uzh.ch).
-Data reported in this paper may be shared by the lead contact upon request. -Original code was written for EmotLearn in Presentation® and may be shared by the lead contact upon request. -Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.
The experiment constituted a randomized, double-blind crossover placebo-controlled study using a low (10 mg; <50 kg body weight) mid (15 mg; <80 kg) or a high dose (20 mg; ≥80 kg) of psilocybin tested against a placebo (mannitol). A total of 30 white healthy right-handed (23 male and 7 female) volunteers with a mean age of 29 years were enrolled. Exclusion criteria were personal and family history of major psychiatric disease as defined in the DSM-V, any major medical conditions, family history of seizure disorder, current psychopharmacological treatment or use of medication, and pregnant or breastfeeding women. The
As part of the main study, participants were investigated double-blind and randomized twice at a 14-day interval with TMS-EEG after administration of psilocybin or placebo (mannitol) and approximately 240 min after substance intake solved a probabilistic learning task, EmotLearn, where the goal was to maximize virtual monetary reward while different emotional cues were presented. Participants were instructed on the specific process and goals of the learning task both on the screening day and on the test day before starting the task. Because of time restraints, no test session was held during the testing day. To counter order effects and minimize systematic bias, we applied complete and random counterbalancing by both randomizing the order of tests on a given day and randomly assigning a participant to one of the two substances.
EmotLearn is a probabilistic learning task programmed in Presentation® () to examine the computational processes behind the interaction between reward learning and subconscious/conscious emotional processing. The goal was to estimate how different emotional cues affect the learning rate under psilocybin or placebo. These emotional stimuli were female and male fearful and neutral faces taken from the Karolinska Directed Emotional Faces database (), presented in a random order. On any of the two measurement days, each participant solved four tasks consisting of sixty trials. On each task, a fixed condition pair was tested (subconscious-fearful, subconscious-neutral, consciousfearful, conscious-neutral). Overall, the thirty participants accumulated 7200 data points per substance (psilocybin, placebo). Each task consisted of an emotional cue followed by one out of two symbol pairs, presented thirty times each. The design was inspired by a study categorizing different presentation times into subconscious (33 ms) or conscious (47 ms) perception, reflecting two hypothesized pathways of emotional processing. A slower, more precise, cortical pathway (several visual stages such as the retina, lateral geniculate nucleus of the thalamus, primary visual area cortex, higher-order brain areas, and finally extending to the amygdala), and a faster, coarser, subcortical pathway (leading through the retina, superior colliculus, pulvinar nucleus of the thalamus, and extending to the amygdala). Each trial began with a fixation cross (1000 ms), followed by an emotional cue in the form of a face. Participants chose one of two symbols by pressing the left or right mouse button to progress to the next trial. Their choice was highlighted with a gray square around the symbol (1500 ms) followed by the outcome (1500 ms) in the form of +10 CHF reward or -10 CHF loss. Each symbol pair was endowed with a superior and an inferior symbol. The superior symbol had a higher probability (70%) of returning a positive reward (+10 CHF) and a lower probability (30%) of returning a loss (-10 CHF), while the opposite was true for the inferior symbol (70% for 10 CHF loss and 30% for 10 CHF reward). A task was concluded when a subject completed sixty choices (two symbol pairs presented thirty times each). Each symbol was randomly displayed an equal number of times on the left or right, with sixteen different symbol pairs randomly assigned to a task. Participants had to learn the stimulus-outcome associations and find the most lucrative strategy to maximize their virtual money reward, although no money was paid out and served as an indirect incentive. Participants usually determined their own learning J o u r n a l P r e -p r o o f heuristic as a win-stay, lose-switch strategy often applied in learning tasks. This probabilistic flair needed the participants to quantify uncertainty and flexibly update their beliefs in an environment of incomplete data.
Raw data from the logfiles consisting of ratios of correct choices between 0 and 1 reflecting the accuracy of choosing the more favorable symbol on each trial in the learning task were extracted with Microsoft Excel (Microsoft, version 16.44) and MATLAB (MathWorks, version 6.5). The Shapiro-Wilk test determined that the data from the learning task were not normally distributed (p < 0.05). Learning curves of accuracy of correct choices were smoothed with a Gaussian filter. All analyses were performed with R (R Core Team, version 4.3.1). For psychometric analysis of learning performance, we used the main dimension "Vigilance Reduction" of the 5-dimensional Altered States of Consciousness Questionnaire (5D-ASC)and the subdimension "Impaired Control and Cognition" of the 11 second-order scales of the 5D-ASC . The nonparametric scores were calculated with a Mann-Whitney test comparing scores after the psilocybin and placebo sessions. The significance level used α = 0.05 in the null hypothesis.
We employed a Bayesian mixed-effects regression model with both fixed and random effects considering an interaction term and applied a prior distribution. Since our main accuracy measure was the proportion of correct choices bounded between 0 and 1, we used a beta distribution for the dependent variable. For each group hierarchy (placebo vs. psilocybin, cue category (fearful vs. neutral, subconscious vs. conscious, dosages) and subgroup (e.g. placebo conscious neutral vs. psilocybin subconscious fearful)), we fitted multiple models in R using a hierarchical Bayesian method with Hamiltonian Markov chain Monte Carlo sampling using the No-U-Turn Sampler and implemented in Stan (Version 2.17.3). The model was run with four chains, each with 2000 iterations, and the first 1000 iterations of each chain were discarded as warm-up, leaving a total of 4000 post-warmup draws for inference. Models were compared with different priors, and the mean of the prior was calculated with the logit function. Posterior distributions were interpreted using the 95% Bayesian credible interval called the highest posterior density interval
Create a free account to open full-text PDFs.
Mason, N. L., Kuypers, K. P. C., Reckweg, J. et al. · Neuropsychopharmacology (2020)
Preller, K. H., Vollenweider, F. X. · Behavioral Neurobiology of Psychedelic Drugs (2016)
Smigielski, L., Kometer, M., Scheidegger, M. et al. · Human Brain Mapping (2020)
Vollenweider, F. X., Preller, K. H. · Nature Reviews Neuroscience (2020)
Carhart-Harris, R. L., Giribaldi, B., Watts, R. et al. · New England Journal of Medicine (2021)
Davis, A. K., Barrett, F. S., May, D. G. et al. · JAMA Psychiatry (2021)
Schindowski, E. M., Jungwirth, J., Schuldt, A. et al. · EClinicalMedicine (2023)
Raison, C. L., Sanacora, G., Woolley, J. D. et al. · JAMA (2023)
Holze, F., Gasser, P., Müller, F. et al. · Biological Psychiatry (2023)
Mertens, L. J., Preller, K. H. · Pharmacopsychiatry (2021)
Ly, C., Greb, A. C., Cameron, L. P. et al. · Cell Reports (2018)
Calder, A. E., Hasler, G. · Neuropsychopharmacology (2022)
Grimm, O., Kraehenmann, R., Preller, K. H. et al. · European Neuropsychopharmacology (2018)
Spriggs, M. J., Murphy-Beiner, A., Murphy, R. et al. · Psychological Medicine (2022)
Umbricht, A., Vollenweider, F. X., Schmid, L. et al. · Neuropsychopharmacology (2003)
Carter, O., Burr, D. C., Pettigrew, J. D. et al. · Journal of Cognitive Neuroscience (2006)
Kometer, M., Schmidt, A., Bachmann, R. et al. · Biological Psychiatry (2012)
Rucker, J. J., Marwood, L., Ajantaival, R. L. J. et al. · Journal of Psychopharmacology (2022)
Vollenweider, F. X., Vollenweider-Scherpenhuyzen, M. F. I., Bäbler, A. et al. · NeuroReport (1998)
Vollenweider, F. X., Smallridge, J. W. · Pharmacopsychiatry (2022)
Pokorny, T., Preller, K. H., Kraehenmann, R. et al. · European Neuropsychopharmacology (2016)
Wall, M. B., Demetriou, L., Giribaldi, B. et al. · American Journal of Psychiatry (2025)
Carhart-Harris, R. L., Friston, K. J. · Pharmacological Reviews (2019)
Kraehenmann, R., Preller, K. H., Scheidegger, M. et al. · Biological Psychiatry (2015)
Carhart-Harris, R. L., Nutt, D. J. · Journal of Psychopharmacology (2017)
Barrett, F. S., Doss, M. K., Sepeda, N. D. et al. · Scientific Reports (2020)
Quednow, B. B., Kometer, M., Geyer, M. A. et al. · Neuropsychopharmacology (2011)
Carhart-Harris, R. L., Erritzoe, D., Williams, T. et al. · PNAS (2012)
Bravermanová, A., Viktorinová, M., Tylš, F. et al. · Psychopharmacology (2018)
Duerler, P., Brem, S., Fraga-González, G. et al. · Cerebral Cortex (2021)
Hasler, F., Grimberg, U., Benz, M. A. et al. · Psychopharmacology (2004)
Holze, F., Becker, A. M., Kolaczynska, K. E. et al. · Clinical Pharmacology and Therapeutics (2022)
Doss, M. K., Považan, M., Rosenberg, M. D. et al. · Translational Psychiatry (2021)
Studerus, E., Gamma, A., Vollenweider, F. X. · PLOS ONE (2010)