Distinct brain responses to psilocybin and escitalopram in depression captured by the Fluctuation-Dissipation Theorem

Major depressive disorder is described as a highly burdensome and heterogeneous condition, and the authors note that existing antidepressants such as SSRIs, including escitalopram, often have only modest efficacy, limited adherence, relapse, and side effects. Against this backdrop, psilocybin has emerged as a candidate treatment, but the brain mechanisms underlying its effects, and how these compare with conventional antidepressants, remain incompletely understood. Previous fMRI research has mainly examined the acute psychedelic phase, often in healthy participants, leaving uncertainty about post-acute effects in depression and about how changes in brain organisation relate to different treatments. Dagnino and colleagues set out to compare psilocybin and escitalopram in patients with major depressive disorder by examining whole-brain hierarchy and non-equilibrium dynamics through a thermodynamic framework based on violations of the Fluctuation-Dissipation Theorem (FDT). They aimed to determine whether the two interventions produce distinct brain reconfigurations after treatment, whether these effects can be measured at whole-brain and regional levels, and whether baseline brain measures relate to clinical response. The paper is presented as an analysis of data from a double-blind randomised controlled trial.

The study analysed data from a double-blind randomised controlled trial, NCT03429075, in which participants with unipolar major depressive disorder were allocated to psilocybin or escitalopram. Eligibility required a clinician-confirmed diagnosis of unipolar MDD and a score of at least 16 on the 21-item Hamilton Depression Rating Scale. Exclusion criteria included personal or immediate family history of psychosis, serious suicide attempts, risky physical health conditions, pregnancy, MRI contraindications, SSRI contraindications, or prior escitalopram use. Treatment resistance was not used as an inclusion or exclusion criterion. Participants underwent screening by telephone and further medical assessment. A total of 59 patients were recruited, with 30 allocated to psilocybin and 29 to escitalopram. The final imaging sample comprised 22 participants in the psilocybin arm and 20 in the escitalopram arm. The main reasons for exclusion were missing post-treatment scans, excessive head motion, adverse reactions leading to discontinuation in the escitalopram arm, cannabis use, and COVID-19 lockdown-related loss to follow-up. Mean age was 41.9 years in the psilocybin group and 38.7 years in the escitalopram group; 8 participants were female in the psilocybin arm and 6 in the escitalopram arm. The treatment schedule included baseline resting-state fMRI with eyes closed, a first dosing day with either 25 mg psilocybin or 1 mg psilocybin in the escitalopram arm as a blinding procedure, a second dosing day three weeks later with the same respective dosing pattern, and six weeks plus one day of daily capsules after the first dosing day. The psilocybin arm received inert placebo capsules, whereas the escitalopram arm received 10 mg escitalopram capsules, starting with one capsule daily for three weeks and then two capsules daily thereafter. Participants were told they were receiving psilocybin but were not informed about dosage. Post-treatment resting-state fMRI was collected three weeks after the second dosing day. Clinical outcome assessment used the Beck Depression Inventory (BDI-1A) at baseline and at 2, 4 and 6 weeks after the first dosing day, although the paper states that the primary clinical outcome for the parent trial was the QIDS-SR-16. The imaging analysis focused on pre- and post-treatment resting-state fMRI. Data were parcellated into 80 regions using the DK80 atlas, and pre-processing used standard pipelines in FSL, AFNI, Freesurfer and ANTs, including motion correction, slice timing correction, registration, scrubbing, band-pass filtering and nuisance regression. For the main analysis, the authors built individual whole-brain models based on a supercritical Hopf bifurcation and derived a generative effective connectivity (GEC) matrix for each participant. They then quantified deviation from the FDT as a measure of non-equilibrium dynamics by perturbing each node and estimating how spontaneous fluctuations related to responses under small perturbations. They also computed a perturbability map, network-level summaries across the Yeo resting-state networks plus subcortical areas, segregation using Louvain modularity on both functional connectivity and GEC matrices, and asymmetry as the proportion of asymmetric interactions after thresholding the absolute difference between each matrix and its transpose. To explore treatment-response prediction, they applied support vector machine classifiers in the escitalopram group using several feature sets: global brain connectivity, GEC in-weight, GEC out-weight, GEC total weight, and the FDT-derived perturbability map. They used leave-one-out cross-validation shuffled 1,000 times, with feature ranking based on training-set separation. Finally, they tested associations between baseline brain measures and BDI change using Pearson correlation, and overall statistical testing relied on permutation-based t-tests with 1,000 permutations and false discovery rate correction for network analyses.

At the whole-brain level, psilocybin and escitalopram produced opposite changes in FDT deviation. After psilocybin, global deviation from the FDT increased significantly, whereas after escitalopram it decreased significantly. The paper reports p<0.05 for psilocybin and p<0.01 for escitalopram in the global analysis, and Figure 2 is described as showing within-group before-versus-after differences of p=0.044 for psilocybin and p=0.004 for escitalopram. An ANOVA supported treatment as the main factor driving these differences, while responder status and BDI change did not explain them. Correlations between change in FDT deviation and change in BDI were not significant when examined within each treatment arm or across both arms together. Regional analyses showed that, in both pre- and post-treatment scans, the highest FDT deviations were mainly in somatomotor and ventral attention areas, together with some subcortical regions, whereas the lowest values were mainly in limbic areas and some default mode and subcortical regions. Changes after treatment generally followed the same global pattern: most areas increased in the psilocybin arm and decreased in the escitalopram arm. For psilocybin, the largest increases were mainly in somatomotor regions; for escitalopram, the largest decreases were in somatomotor, dorsal attention and ventral attention regions. At the resting-state network level, psilocybin showed a trend towards increased FDT deviation across all networks, with significant changes in somatomotor, dorsal attention, ventral attention and default mode networks, although these did not survive multiple-comparisons correction. Escitalopram produced significant decreases across all networks, with several effects surviving correction, especially in ventral attention and subcortical networks. Segregation analysis yielded no significant findings when applied to functional connectivity matrices alone, but the GEC matrices showed significant opposite changes: modularity decreased after psilocybin and increased after escitalopram, both with p<0.001. The authors describe this as more overlapping communities after psilocybin and greater segregation after escitalopram. Correlations between change in GEC modularity and change in BDI were not significant within either treatment arm, although a positive correlation appeared when both groups were pooled. The authors interpreted this pooled association as masking the opposite treatment-specific effects. Asymmetry analyses showed that global asymmetric interactions increased after psilocybin and decreased after escitalopram across thresholds, matching the pattern observed for FDT deviation. Intra-module and inter-module asymmetry followed the same direction in each arm. When examining responders versus non-responders, psilocybin responders generally showed increases in perturbability across most cortical regions, whereas non-responders had more mixed and heterogeneous patterns. In the escitalopram arm, both responders and non-responders generally showed decreases in perturbability, but the changes were larger in responders. Subcortically, some regions did not follow the overall treatment trend in either arm, including the globus pallidus internal segment and caudate in the psilocybin group, and the globus pallidus internal segment, hippocampus, subthalamic nucleus and nucleus accumbens in the escitalopram group. For treatment-response prediction in escitalopram, the support vector machine performed best when using the FDT-derived perturbability map, reaching a maximum accuracy of 85.50% with one feature. This outperformed the GBC and the different GEC-based classifiers, which achieved maximum accuracies of 35.55%, 52.95%, 58.25% and 56.85% depending on the feature set. The most informative regions across classifiers were mainly in limbic, default mode and subcortical areas, with additional contributions from visual, somatomotor and ventral attention regions. Finally, baseline asymmetry within the somatomotor network was significantly positively correlated with change in BDI in the psilocybin arm, meaning that lower baseline asymmetry was associated with greater clinical improvement. No significant relationship was found for escitalopram or when both treatment groups were combined.

The authors argue that their thermodynamic framework captured distinct post-treatment reconfigurations of hierarchical brain dynamics in depression. They interpret psilocybin as increasing non-equilibrium dynamics, asymmetry and reduced segregation, whereas escitalopram produced the opposite pattern across the same measures. They present these opposing effects as evidence that the two interventions may reduce depressive symptoms through different brain-level mechanisms, extending previous analyses from the same trial cohort. They place their findings in the context of earlier psychedelic research, noting that acute psilocybin effects in healthy participants have often been associated with reduced temporal asymmetry, increased complexity and a loosening of hierarchical constraints, whereas their study examined post-acute effects in depressed patients. The authors suggest, cautiously, that the post-acute increase in non-equilibrium after psilocybin may reflect recalibration of hierarchical organisation relative to the depressive baseline state, rather than simply mirroring the acute psychedelic state. They also note that future work should examine healthy versus depressed states and compare onset, peak and post-acute phases more directly. At the regional level, the authors emphasise the prominence of somatomotor and ventral attention regions, with additional involvement of default mode and dorsal attention networks, and they link these systems to depression-related processes such as embodied symptoms, cognitive vulnerability, self-referential thought and rumination. They interpret the responder analyses as showing that psilocybin responders tended to move towards greater perturbability across most areas, whereas escitalopram responders and non-responders generally showed reductions. For modularity, they interpret the decrease after psilocybin and increase after escitalopram as consistent with increased integration versus greater segregation, respectively, and they note that the GEC-based approach differentiated the groups more clearly than classical functional connectivity. The authors also highlight the baseline predictor findings, particularly the relationship between lower baseline somatomotor asymmetry and greater clinical improvement under psilocybin, and the superior classification performance of the FDT-based measure for escitalopram response. They frame these results as relevant to precision psychiatry and biomarker development. They acknowledge several limitations. The analysis used an 80-region DK80 parcellation, which they describe as a compromise between computational cost and spatial resolution, and there is no single agreed standard for fMRI parcellation. They also relied on a structural connectivity template from a separate healthy cohort rather than individual connectomes. In addition, the sample size was relatively small, limiting generalisability and statistical power. The authors conclude that larger replication studies would improve robustness and help clarify the value of the approach for understanding and optimising depression treatment.