LSD microdosing attenuates the impact of temporal priors in time perception

Over recent years, predictive processing accounts have been invoked to explain how psychedelics alter perception and cognition. One prominent formulation, the REBUS (Relaxed Beliefs Under Psychedelics) model, proposes that psychedelics relax the precision or weight of priors—expectations formed from past sensory experience—thereby increasing the relative influence of incoming sensory evidence and reducing perceptual biases. Time perception is amenable to Bayesian formulations in which global and local temporal priors bias interval estimates toward recent or average durations, producing systematic under- or over-reproduction of stimulus intervals. Sadibolova and colleagues set out to test whether microdoses of LSD reduce the temporal reproduction bias via a diminished influence of temporal priors, as predicted by REBUS. The paper reports pre-registered re-analyses of an earlier trial in which participants received LSD microdoses or placebo and performed a visual temporal reproduction task spanning subsecond to suprasecond intervals. The principal aim was to examine the roles of global and local priors, both unweighted and weighted by their precision (inverse variance), in accounting for the previously observed LSD-related reduction in under-reproduction of long intervals.

The analyses re-use data from a randomised, double-blind, placebo-controlled inpatient study. The participant sample comprised 48 English-speaking adults aged 55–75 (mean age = 62.92); 21 were female (44%) and 27 male (56%). Participants were randomly allocated to one of four groups (n=12 per group) receiving placebo (0 μg) or LSD microdoses (5, 10, or 20 μg). Two multivariate outliers (one from the 5 μg group and one from the 20 μg group) were removed, yielding a final sample of 46. Dosing and testing occurred as part of a larger safety/efficacy protocol. The LSD solution was prepared in distilled water under cGMP standards (manufacturer: Onyx Scientific Ltd). Participants completed the temporal reproduction task once post-dose on the fourth day of dosing. Each trial began with a 750 ms cue (“memorize”), followed by a jittered blank (425–650 ms) and then a target stimulus (a blue circle) of varying duration (the extracted text indicates stimulus intervals spanned subsecond to suprasecond durations, with earlier material specifying 0.8 to 4 s). After a 500 ms blank the response cue (“reproduce”) prompted participants to hold the space bar to reproduce the interval; a co-appearing blue circle remained until release. Participants completed one practice block and four experimental blocks of 27 trials each, for 108 experimental trials in total. Preprocessing removed bivariate outlier responses per participant using the median absolute deviation method (mean removed responses = 8.37%, SD = 2.95%, range 1.85–14.81%). Data were analysed with mixed-effects models (lme4 in R) reflecting the hierarchical design: participants (46 levels) nested within dose and cohorts (4 levels). The authors used a dichotomous drug variable (placebo versus LSD) in models alongside continuous predictors for stimulus interval and priors. Priors were operationalised as: a global prior (arithmetic mean of all preceding intervals from trial 3 onward) and three local priors (n-1, n-2, n-3 means). Continuous variables were mean-centred and categorical variables dummy-coded. Models were fitted with restricted maximum likelihood using the nlopt_ln_neldermead optimiser and tighter tolerances. Model selection used changes in log-likelihood, AIC and BIC; Kenward–Roger approximations provided p-values for fixed effects. Bootstrapped 95% confidence intervals were computed via bootMer with 100 iterations. The final random-effects structure included correlated by-subject within-dose within-cohort intercepts and slopes for stimulus intervals. Because the pre-registered analysis treated priors as stimulus history without precision, the authors additionally computed unregistered precision-weighted priors. Precision weights were defined as the inverse of variance normalised by the sum of inverse variances of prior and likelihood. Prior variance was estimated from variance of preceding trials (global) or preceding 2–3 trials (n-2, n-3); likelihood variance for each trial was estimated after regressing out prior influence (for global priors this involved regressing preceding responses on stimulus intervals and using residual variance adjusted by the squared slope). The n-1 prior was excluded from precision-weighted analyses because its trial-by-trial precision could not be reliably computed from the extracted text.

Replication of the original effect: Mixed-effects analyses replicated the earlier finding that LSD reduced the under-reproduction bias for supra-second intervals. A model including stimulus interval as a fixed effect showed a positive relationship between stimulus duration and reproduced duration (β = .17, 95% CI [.07, .27], SE = .06, t = 3.04, p = .004), corresponding to an approximate .64 s increase in reproduced intervals per 1 s increase in stimulus duration across conditions. Relative to placebo, reproduced intervals under LSD were, on average, longer by .24 s and increased by an additional .17 s per 1 s increment, indicating a shift toward more veridical timing under LSD. Global priors (unweighted): Including the unweighted global prior in models produced mixed indicators of model fit: AIC and Kenward–Roger tests suggested improvement while BIC worsened by >10 points, making evidence inconclusive. A three-way interaction testing whether the global prior differentially affected long-interval reproduction across drug groups was non-significant (β = .01, 95% CI [-.46, .39], SE = .21, t = .03, p = .98). The extracted text reports that, paradoxically, reproduced intervals under LSD showed a steeper alignment with the global prior than placebo (i.e. apparently greater reliance on the unweighted global prior), although this pattern did not provide clear statistical support for LSD moderating unweighted global priors. Similar patterns were reported for unweighted n-2 and n-3 priors. Precision-weighted priors: When priors were weighted by their precision, a different picture emerged. Higher precision of local priors (n-2, n-3) was associated with stronger local-history biases—reproduced intervals decreased with increasing local prior precision weights. Crucially, once precision-weighted local priors (n-2) were accounted for, the previously observed LSD-mediated reduction in under-reproduction was no longer statistically significant: the drug effect estimate after controlling for the precision-weighted n-2 prior was β = -0.01 (95% CI [-.05, .03]) and across-condition differences for individual long intervals were not statistically significant after Holm–Bonferroni correction (ps > .05). The authors report that the reduction in reproduced intervals driven by increasing precision weights occurred more slowly in the LSD group than in placebo, and that controlling for precision-weighted local priors eliminated the between-group differences in reproduction of long intervals. From these analyses the investigators infer that reduced precision-weighting of local temporal priors underlies the altered time perception observed with LSD microdoses.

Sadibolova and colleagues interpret their re-analysis as broadly consistent with predictive-processing formulations such as REBUS: LSD microdosing appears to lessen the influence of local temporal priors on reproduced intervals, thereby reducing the under-reproduction bias for longer stimulus durations. The authors note that their finding is specific to precision-weighted local priors—unweighted global priors showed an inconclusive or even counterintuitive pattern—so incorporating prior precision is critical for assessing how psychedelic drugs modulate the balance between top-down expectations and bottom-up sensory evidence. The paper situates these behavioural findings within neurocognitive and neurochemical hypotheses. REBUS posits that psychedelics increase excitability of deep-layer pyramidal neurons expressing 5-HT receptors, disrupting synchrony and lowering prior precision so that ascending prediction errors have greater impact. The authors acknowledge that LSD also engages dopaminergic and glutamatergic systems; changes in striatal dopamine might alternatively explain reduced precision-weighting of temporal priors, and complex interactions between serotonin and dopamine could underlie timing effects. An additional account—the cortico‑striatal‑thalamocortical (CSTC) model—was discussed: LSD may reduce thalamic gating and increase bottom-up information flow, functionally equivalent to raising likelihood precision relative to priors. The claustrum was highlighted as another candidate region for future investigation given its dense 5-HT receptor expression and putative role in temporal integration. Concerning microdosing, the authors argue that small LSD doses typically produce minimal phenomenology, making expectancy effects an unlikely explanation for the observed reduction in under-reproduction. They recommend future studies that explicitly manipulate temporal priors across a range of doses and probe neurochemical and neural mechanisms (including thalamic and claustral contributions) to discriminate between serotonergic, dopaminergic, and network-level interpretations. The investigators also acknowledge that the neurocognitive and neurochemical instantiation of these timing effects remains underspecified and call for targeted mechanistic work. No separate conclusion section is present in the extracted text.