Effects of psilocybin on probabilistic choice behavior.

A. After a variable amount of time (3-30 sec), a trial began with illumination of the nosepoke lights. Animals then made a left or right action (choice), which could immediately result in reward or no reward. Animals were required to make a feeder entry to initiate the next inter-trial interval. B. An example session showing the underlying probabilities of reinforcement for left or right choices. Reinforcement probabilities for each action could be one of three combinations. Two blocks were designated as low uncertainty as they contained an optimal choice (e.g. left or right optimal for block 1 and 3 above, respectively). In high uncertainty blocks neither choice was optimal (block 2). Blocks pseudorandomly transitioned to a different block. C. Representative choice data (moving 5 trial average) from a subject. Shading represents the underlying block with R=reward, r= choose right. Small hashes at the top and bottom indicate a trial was unrewarded while large hashes indicate a trial was rewarded. D. Win-stay/lose-shift for low uncertainty blocks (mean ± SEM, lines indicate individual rats). Animals used recent outcomes to drive choices as evident by outcome sensitive stay probabilities (*p<.05 effect of outcome, RM ANOVA). E. Win-stay/lose-shift for high uncertainty blocks (mean ± SEM, lines indicate individual rats). Win-stay/lose shift behaviors were outcome sensitive and unchanged by psilocybin (*p<.05 effect of outcome, RM ANOVA). F. Choice variability for low uncertainty blocks (mean ± SEM, lines indicate individual rats). There was no difference between baseline and psilocybin sessions. Note: higher entropy values signify higher variability. G. Choice variability for high uncertainty blocks (mean ± SEM, lines indicate individual rats). Psilocybin increased the entropy metric for choice variability (*p<.05, paired t-test). H. Transition analysis of choice behavior for low uncertainty blocks (mean ± SEM). Psilocybin enhanced asymptotic (i.e.pre-transition) optimal choice. When contingencies transitioned to lower reinforcement probability for the optimal side, animals adapted by decreasing preference for said side (*significant intercept effect of psilocybin, #p<.05 significant effect of reversal phase, generalized mixed-effects model). I. Mean ± SEM and individual subject averages for the pre-transition window (*p=.022, paired t-test one-tailed). J. Mean ± SEM and individual subject averages for the 1-5 trial post transition window (p=.036, paired t-test). K. Mean ± SEM and individual subject averages for the 6-10 trial post transition window (p=.70, paired t-test). Lines in bar plots indicate individual rats. N=7-8 rats.

Computational modeling of choice behavior indicates psilocybin shifts learning mechanisms.

A. Schematic of the reinforcement learning process with outcome dependent learning rates (η+/-), a value independent bias parameter (α), and an inverse temperature (β) to influence choice policy via Q learning. Choices (ct) produce an outcome (Rt) which is compared to the prior Q value for a given chosen side (Qc(t-1)) to form a prediction error (PE). Learning rates then control how much the PE is used to update Qc. The Q value for the chosen and unchosen (Qun) options are transformed into choice probabilities via a logistic function. B. Representative observed choice behavior (5 trial average, light grey) and simulated model predictions for that subject (dark grey). Shading indicated probability of reward given a right choice. C. Simulated choice averages (mean and 95% CI) around transitions in reinforcement probability from low uncertainty for baseline and psilocybin sessions. D. Boxplots of model simulations (median and IQR, whiskers indicate 5-95 percentile) for asymptotic optimal choice in low uncertainty blocks (trials −4-0). The model recovered increases in optimal choice seen after psilocybin treatment (*p=.025, one-tailed t-test). E. Boxplots of model simulations (median and IQR, whiskers indicate 5-95 percentile) for choice entropy in high uncertainty blocks. The model recovered increases in choice entropy seen with psilocybin treatment (* p<.01, one-tailed t-test). F-I. Posterior parameter distributions for the group level effect of psilocybin (n=4000 estimates from 4 chains). Below the distribution: red circles indicate the mean, thick black bars indicate 80% highest density interval (HDI) and thinner black bars indicate 95% HDI. F. Temperature was not affected by psilocybin (0∊75%HDI). G. Bias was attenuated by psilocybin (0∉80%HDI). H-I. Psilocybin shifted learning rates by increasing rewarded learning rate (0∉80%HDI) and decreasing unrewarded learning rate (0∉90%HDI).

Psilocybin bidirectionally shifts population and subcluster related activity to reward and no reward outcomes.

A-B. Population averaged firing rate during the peri-action epoch for trials which resulted in reward or no reward (mean ± SEM). Psilocybin altered response trajectories by elevating the length of reward response and inhibiting no reward related response relative to baseline (n=207-222 units from 7-8 rats; mixed effects model interactions p < .02). C. Schematic demonstrating reward and no reward responses of two example units. These signals were correlated between each unit pair and subjected to clustering analysis based on reward and no reward response profiles (see methods). Right - Unit by unit correlation matrix sorted by the results of agglomerative cluster analysis. D. Two subtypes of reward-preferring neuronal clusters were found based on their temporal pattern of activation (depicted as Type I and Type II). Psilocybin selectively increased the relative proportion of Type I units (* p=.01, Fisher exact test, Benjamin-Hochberg corrected). E. Mean and 95% CI for peri-action activity of each subtype (n=71-105 units). F. Mean and 95% CI relative change between reward and no reward response shows preference for reward which was not influenced by psilocybin. G. Two subtypes of no reward preferring neuronal clusters were found, and psilocybin selectively decreased the relative proportion of Type I units (* p=.03 Fisher exact test, Benjamin-Hochberg corrected). H. Mean and 95% CI for peri-action activity of each subtype (n=90-94 units). I. Mean and 95% CI relative change between reward and no reward response shows preference for no reward, the magnitude of which was not influenced by psilocybin. J. A final cluster of neurons showed rapid decay in activity immediately at action/outcome, and this cluster was less prevalent during acute psilocybin (* p=.04, Fisher exact test, Benjamin-Hochberger corrected). K. Mean and 95% CI for peri-action activity of the final cluster (n=69 units). L. While decreases were observed in activity for rewarded and unrewarded trials, preference scores indicate that magnitude of this decrease was lower for unrewarded trials (mean and 95% CI).

Psilocybin attenuates choice related encoding at multiple timescales.

A. Mean spike rate and raster plot for a representative single unit which encodes choice direction (left/right). B. Fraction of task encoders showing significant choice encoding at each time bin over the peri-action period. Psilocybin attenuated encoding prior to the action being executed. C. Mean spike rate and raster plot for a representative single unit which encodes outcome. D. Fraction of task encoders showing significant outcome encoding at each time bin over the peri-action period. E. Mean spike rate and raster plot for a representative single unit which encodes outcome differentially based on choice. F. Fraction of task encoders showing significant choice x outcome encoding at each time bin over the peri-action period. Psilocybin attenuated encoding following outcome delivery. G. Mean spike rate and raster plot for a representative single unit which encodes the outcome from the prior trial. H. Fraction of task encoders showing significant prior outcome encoding at each time bin over the peri-action period. Colored circles above plots indicate significant binomial test (p<.01 Benjamin-Hochberg corrected) for baseline or psilocybin. Black ‘x’ above plots indicates significant cluster corrected chi2 test between baseline and psilocybin. Solid black bars on x-axis indicate significant session difference for that 0.5 sec time chunk (GLM with Benjamin-Hochberg corrected post-hoc contrast).

Encoding of total and relative value signals during acute psilocybin.

A. Mean and SEM of relative value estimates in the first 10 trials of a block and the last 5 trials of the block. Data are color coded by block type and symbols reflect psilocybin (circles) or baseline (diamond) session. B. Mean spike rate and raster plot for representative single unit with relative value encoding. For visualization, data are colored by terciles of value estimates. C. Fraction of choice-value encoders showing significant relative value encoding at each time bin over the peri-action period. D. Mean and SEM of total value estimates in the first 10 trials of a block and the last 5 trials of the block. Data are color coded by block and symbols reflect psilocybin (circles) or baseline (diamond) session. E. Mean spike rate and raster plot for representative single unit with total value encoding. For visualization, data are colored by terciles of value estimates. F. Fraction of choice-value encoders showing significant total value encoding at each time bin over the peri-action period. Colored circles above plots depict significant binomial test (p<.01 Benjamin-Hochberg corrected) for baseline or psilocybin.

Psilocybin attenuates decoding of choice in the peri-action period.

A. Decoding choice (chosen side) before action execution was attenuated by psilocybin across a range of pseudo-population sizes (mean ± 95%CI, *p<.05 main effect of psilocybin). Squares near the 50% chance line are the results of shuffle analysis (#p<.05 versus shuffle). B. Full population decoding (mean ± 95% CI, n = 20 runs with 207-222 units) demonstrated that decoding performance for choice was highest before and at action execution in baseline sessions. Decoding performance was highest at action execution in psilocybin sessions. Lighter shading indicates mean ± 95%CI for shuffle runs. C. Box plots (median ± whiskers for the 5-95 percentile) showing effects of removing a specific cluster (see Figure 3) from choice decoding analysis during the pre-action period (−1-0 s). The largest losses in accuracy were observed when decay neurons were ablated (*p<.05 action decay vs each other cluster, Benjamin-Hochberg corrected).