Divergent changes in perturbation-induced brain reconfiguration following depression treatment with psilocybin and escitalopram

The authors frame major depressive disorder as a condition marked by altered brain dynamics, with a tendency towards rigidity rather than flexible adaptation. They note that artificial perturbations of whole-brain models have been used to study brain sensitivity and transitions between states, but that previous work has focused more on immediate responses to perturbation than on how brain organisation itself reconfigures when perturbations are sustained. They also point to earlier studies in the same trial showing treatment-specific differences in susceptibility to perturbation after psilocybin and escitalopram, but say it remained unclear whether these differences extended to broader reorganisation of brain networks. Dagnino and colleagues therefore set out to test whether whole-brain reconfiguration under perturbation changes after treatment with psilocybin versus escitalopram in patients with major depressive disorder. To do this, they introduced the brain network reconfiguration index (NRI), intended to quantify how much a whole-brain model reorganises when perturbed, rather than simply how strongly it responds. The study used the same clinical cohort as earlier work and aimed to characterise treatment-related neural changes in a way that could complement conventional connectivity analyses.

The analysis used resting-state fMRI data from a double-blind randomised controlled trial of depression treatment, identified as NCT03429075. The extracted text indicates that the final sample for this analysis comprised 22 participants in the psilocybin arm and 20 in the escitalopram arm, with baseline and post-treatment eyes-closed resting-state fMRI scans. Mean age was 41.9 years in the psilocybin group and 38.7 years in the escitalopram group. The text does not clearly describe the full recruitment criteria or clinical inclusion and exclusion criteria in this section. Participants in the psilocybin arm received two doses of 25 mg psilocybin, spaced 3 weeks apart. Participants in the escitalopram arm received 1 mg psilocybin on the dosing days, apparently to standardise expectation, and then took daily capsules for 6 weeks and 1 day; these capsules contained inert placebo in the psilocybin arm and 10 mg escitalopram in the escitalopram arm. The authors state that all participants were told they would receive psilocybin, but were not told the dose. The treatment schedule included a second dosing day 3 weeks after the first, with no crossover in dosages between groups. Neuroimaging data were parcellated using the DK80 atlas, which comprises 62 cortical and 18 subcortical regions. The authors built individual whole-brain Hopf models, representing each brain area with a Stuart-Landau oscillator. They derived an initial generative effective connectivity (GEC) matrix for each participant by fitting the model to empirical fMRI data, using structural connectivity and a thermodynamic optimisation approach based on an “arrow of time” measure to capture directional asymmetry in information flow. To compute NRI, the researchers perturbed each brain area separately by increasing Gaussian noise while keeping the other areas fixed at 0.01, across stimulation intensities from 0.02 to 0.10 in steps of 0.01. They then reoptimised the perturbed model to produce a response GEC (GECr) and quantified NRI as the decorrelation, or inverse similarity, between GECr and the initial GEC. Higher NRI therefore indicates greater reorganisation under perturbation. The paper also reports complementary graph-theory analyses of network integrity after targeted node removal, using thresholded weighted and binarised GEC matrices across 5–100% top edges. Statistical testing used permutation-based Wilcoxon sign-rank tests with 1,000 permutations, a 0.05 significance threshold, and false discovery rate correction for multiple comparisons.

The main finding was that global NRI increased after psilocybin treatment and decreased after escitalopram treatment. This pattern was seen when NRI was averaged across all brain areas and stimulation intensities. The between-session differences were reported as significant across all stimulation intensities above 0.02 after correction for multiple comparisons, whereas 0.02 was described as very close to the unperturbed model and produced negligible perturbation. At the regional level, NRI across brain areas was higher after treatment in the psilocybin arm and lower after treatment in the escitalopram arm. The strongest absolute changes were concentrated mostly in subcortical regions and in areas linked to the ventral attention and limbic networks. In the psilocybin group, areas associated with the default mode network also showed prominent changes, whereas in the escitalopram group additional changes were observed in regions contributing mainly to the dorsal attention network. The authors also examined whether regional NRI was related to perturbability, a measure of how much a region orchestrates hierarchical brain dynamics. They report a significant positive Pearson correlation between nodal perturbability and NRI in both treatment groups and at both sessions, although the exact correlation coefficients are not given in the extracted text. In the complementary graph-theory analyses, the two interventions again showed opposite patterns. For the weighted network analysis, psilocybin was associated with a significant reduction in area under the curve of global efficiency after treatment across all thresholds, whereas escitalopram showed a significant increase across all thresholds. For the binarised analysis, both groups showed the same direction of change at thresholds of 5–15%, but the direction reversed at higher thresholds: psilocybin showed higher AUC at thresholds at or above 30%, while escitalopram showed lower AUC above 30%. In the node-removal analysis based on the distance to the largest drop in cluster size, psilocybin showed a higher distance after treatment for almost all thresholds, and escitalopram showed the opposite, with significant effects for most thresholds in the 35–50% range and for psilocybin also at 5%. The extracted text does not clearly provide exact p-values, effect sizes, or confidence intervals for the main NRI comparisons, beyond noting significance after multiple-comparison correction.

The authors interpret their findings as evidence that depression treatment with psilocybin and escitalopram leads to divergent changes in how the brain reorganises under perturbation. They argue that increased NRI after psilocybin reflects a shift away from rigid, maladaptive network configurations towards greater capacity for reorganisation, whereas decreased NRI after escitalopram suggests a more stabilised brain state closer to the unperturbed configuration. They present this as extending earlier findings from the same dataset showing increased susceptibility after psilocybin and reduced susceptibility after escitalopram. Dagnino and colleagues place the psilocybin findings in the context of theories of psychedelic therapy, including ideas about increased brain and mind plasticity, psychological flexibility, and the REBUS model, which proposes a temporary flattening of the brain’s energy landscape that enables escape from rigid attractors. They also note that the treatment-specific neural pattern resembles previously reported psychological differences: psilocybin has been linked to greater emotional intensity, responsiveness, and reductions in avoidance and anhedonia, whereas escitalopram has been associated with reduced emotional responsiveness and emotional blunting. The authors further interpret the positive association between regional NRI and perturbability as suggesting that brain reconfiguration is greatest when perturbation occurs in regions with stronger influence over hierarchical organisation. They relate this to earlier work indicating that core brain areas can shape how perturbations spread through the network. They also argue that their graph-theory analyses are consistent with the NRI results: after psilocybin, the network appears to contain weaker connections but a broader communicative structure, whereas escitalopram shows the opposite pattern. The discussion emphasises several strengths of the approach. The authors say whole-brain Hopf models capture fMRI dynamics with a useful balance between realism and complexity, and that their framework assesses reconfiguration under sustained perturbation rather than only immediate response. They also note that NRI is computed in a thermodynamic framework based on the “arrow of time”, and that unlike traditional node-removal approaches, their perturbation ordering is not predetermined. The main limitations they acknowledge are the relatively small sample, which limits statistical power and generalisability, the dependence of NRI estimates on the chosen brain parcellation, and the use of structural connectivity matrices derived from healthy participants rather than from the patients themselves. They suggest that future work should test the robustness of the findings in larger cohorts, across alternative atlases, and with more individualised structural connectivity data. The authors conclude that their work adds to knowledge about perturbation, depression, and treatment effects, and offers a new way to characterise whole-brain reconfiguration after intervention.