The mixed serotonin receptor agonist psilocybin reduces threat-induced modulation of amygdala connectivity

Serotonin (5-HT) signalling is central to neural circuits involved in emotion processing, and Kraehenmann and colleagues have previously reported that psilocybin (a 5-HT2A receptor agonist) attenuates amygdala blood oxygen level-dependent (BOLD) responses to threat-related visual stimuli and is linked to acute mood enhancement. The authors motivate the present work by noting uncertainty about whether psilocybin alters bottom-up (visual→amygdala→prefrontal) or top-down (prefrontal/amygdala→visual) effective connectivity within the visual–limbic–prefrontal network that underlies threat processing. To address this gap the study re-analysed earlier fMRI data with dynamic causal modelling (DCM) and Bayesian model selection (BMS). The investigators set out to test three alternative mechanisms for threat-induced modulation of connectivity (bottom-up only, top-down only, or both), and to determine which connectivity changes account for the previously observed psilocybin-related reductions in amygdala and primary visual cortex (V1) activation during threat processing.

Twenty-five healthy, right-handed volunteers (16 males; mean age 24.2 ± 3.42 years) were recruited and screened to exclude psychiatric, neurological, cardiovascular, and substance-use disorders. The study used a randomised, double-blind, placebo-controlled, cross-over design; participants received either oral placebo or 0.16 mg/kg psilocybin in two imaging sessions at least 14 days apart. Mood was assessed with PANAS and the state portion of the STAI before and about 210 minutes after drug administration. Scanning took place 70–90 minutes post-dose to coincide with the plateau of psilocybin effects. During fMRI, subjects performed an amygdala reactivity task comprising alternating 24-s blocks of threat-picture discrimination, neutral-picture discrimination and shape discrimination (baseline). Stimuli were drawn from the IAPS set; there were 24 threat and 24 neutral pictures arranged in triplets, and each block presented six images for 4 s each. Behavioural responses (choice via MR-compatible button device) did not alter trial duration and no performance feedback was given. Functional images were acquired on a 3 T scanner (EPI, TR = 2.5 s, TE = 30 ms, 3 × 3 × 3 mm voxels) and preprocessed in SPM12b (realignment, coregistration, unified segmentation normalisation, 8 mm smoothing). First-level GLMs modelled threat, neutral and shape blocks and produced threat-minus-shapes contrast images for second-level analyses. DCM for fMRI (SPM12b, DCM12) was applied to infer directed (effective) connectivity between three right-hemispheric regions of interest chosen a priori and from the GLM: right primary visual cortex (rV1, x=12,y=-82,z=-7), right amygdala (rAMG, x=24,y=-1,z=-13) and right lateral prefrontal cortex (rLPFC, x=54,y=32,z=20). Regional time series were extracted from 10 mm spherical volumes centred on subject-specific suprathreshold voxels within a 10 mm radius of the group maxima, summarised by the first eigenvariate. The authors report they could not extract an rLPFC time series in two subjects and that they restricted DCM analyses to subjects showing significant responses; the extracted text does not clearly report the final sample size included in the DCM analyses. A three-node base model with bidirectional endogenous connections V1↔AMG and AMG↔LPFC was defined, with visual stimuli entering at V1. The modulatory effect of threat was modelled in three variants corresponding to the hypotheses: modulation of bottom-up connections, modulation of top-down connections, or modulation of both. Group-level random-effects BMS was used to adjudicate the most plausible model per drug condition, and random-effects Bayesian model averaging (BMA) produced subject-specific connectivity estimates weighted by posterior model probability. Connectivity estimates (in Hz) were entered into repeated-measures ANOVAs (factors: drug [psilocybin, placebo] and connection type) to test drug effects on endogenous and modulatory parameters; significant ANOVA effects were followed up with Duncan's multiple range tests. Psilocybin–placebo differences in direct V1 input were tested with paired t-tests. Finally, Pearson correlations tested relationships between psilocybin-induced connectivity changes and changes in behavioural measures (reaction time, accuracy) and mood scores (PANAS, STAI). A two-tailed p < 0.05 threshold was used for statistical significance.

Bayesian model selection indicated that the full model—where threat modulates both forward and backward connections in the V1–AMG–LPFC network—best explained the data under both treatments. Exceedance probabilities for the full model were 97% under placebo and 62% under psilocybin. Parameter estimates were obtained via Bayesian model averaging for each treatment. Coupling in the dynamic models is reported in units of Hz, with typical baseline connections on the order of 0.1–0.5 Hz. A 2-way repeated-measures ANOVA on modulatory parameters found no main effect of drug (F1,22 = 3.10, p = 0.09, ηp2 = 0.12), a significant main effect of connection type (F3,66 = 3.94, p = 0.01, ηp2 = 0.15), and a significant drug × connection-type interaction (F3,66 = 2.84, p = 0.04, ηp2 = 0.11). Post-hoc tests showed that the threat-induced modulation of the AMG → V1 connection was significantly reduced after psilocybin compared with placebo (p = 0.01, Duncan corrected). There were no significant psilocybin effects on endogenous (baseline) connections or on direct input parameters (all p > 0.05). Correlation analyses between psilocybin-minus-placebo changes in the AMG → V1 modulatory coupling and corresponding changes in behavioural measures (reaction time, accuracy) or mood scores (PANAS positive/negative, STAI state anxiety) revealed no significant associations (all p > 0.05).

Kraehenmann and colleagues interpret their findings as showing that psilocybin attenuates amygdala-dependent top-down tuning of visual cortex during threat processing. The reduction in the threat-induced modulatory effect of the AMG → V1 connection under psilocybin offers a mechanistic account for previously observed decreases in amygdala and visual-cortex responses to threat and for acute shifts in emotional bias away from negative valence. The authors situate the winning full model—featuring reciprocal V1↔AMG and AMG↔LPFC connections—within prior DCM literature that similarly favours bidirectional modulation during negative emotion processing. They note that feedback from the amygdala can tune visual processing to prioritise salient threat signals and that excessive amygdala-driven tuning may bias perception towards negative stimuli, a process implicated in anxiety and depression. Reducing that amygdala-to-visual cortex signalling may therefore decrease threat sensitivity and free capacity in visual processing to represent non-threat or positive information. Contrary to their hypothesis, psilocybin did not increase inhibitory top-down connectivity from LPFC to AMG. The authors propose two non-mutually exclusive explanations: first, the psilocybin-induced reduction in amygdala activation might arise from a direct effect within the amygdala, where 5-HT2A receptors are abundant and agonists have direct action; second, a reduced amygdala response would lessen the need for LPFC regulation, yielding no observable increase in LPFC→AMG coupling. They further acknowledge that other prefrontal regions (medial PFC, anterior cingulate, orbitofrontal cortex) could mediate prefrontal regulation of the amygdala; task design and the lack of GLM activation in those regions in the present dataset make such effects less likely to have been detected here. The authors acknowledge several limitations: the DCM network was deliberately parsimonious and excluded other regions known to participate in threat processing (ACC, OFC, fusiform gyrus); analyses were restricted to a right-hemispheric subnetwork for statistical efficiency, so left-hemisphere prefrontal influences could have been missed; and the extracted text does not clearly state the final sample size included in the DCM analyses after excluding subjects without sufficient regional activation. For future work they recommend broader and bilateral effective connectivity models and tasks that differentially recruit left and right prefrontal regions. Finally, the authors contrast psilocybin's network-level effects with those reported for conventional serotonergic antidepressants (for example, (S)-citalopram), suggesting that while both modalities may normalise amygdala hyper-reactivity to threat, they appear to act on different network interactions. They highlight potential therapeutic implications, proposing that attenuating amygdala-driven top-down visual tuning could help reduce negative cognitive biases in mood and anxiety disorders and may facilitate exposure-based psychotherapy in conditions such as post-traumatic stress disorder.

The authors conclude that acute psilocybin administration decreases threat-related top-down connectivity from the amygdala to the visual cortex, which may reduce threat sensitivity in visual processing. This mechanistic effect could underlie psilocybin's capacity to shift emotional bias away from negative valence and has potential therapeutic relevance for mood and anxiety disorders in which persistent negative cognitive biases and heightened sensitivity to aversive stimuli are central features.