Psilocybin for treatment-resistant depression: fMRI-measured brain mechanisms

Psilocybin, a serotonin 2A receptor agonist and the active prodrug psilocin, is showing clinical promise when administered with preparatory and integrative psychological support, but the neural mechanisms that underlie its antidepressant effects remain poorly understood. Earlier human imaging work has mainly characterised the acute psychedelic state, reporting a more entropic brain configuration with modular disintegration and enhanced global connectivity that correlates with aspects of the subjective experience such as ego-dissolution. Few studies, however, have examined brain function beyond the acute phase (>12 hours) in clinical populations, leaving a gap about post-acute or 'after-glow' changes that might relate to mood improvement and longer-term outcomes. Carhart-Harris and colleagues set out to characterise cerebral blood flow (CBF) and resting-state functional connectivity (RSFC) changes before versus one day after psilocybin treatment in patients with treatment-resistant depression (TRD). Patients received two oral doses (10 mg then 25 mg one week apart) within an open-label clinical protocol. The investigators tested whether post-treatment CBF and BOLD RSFC would be altered relative to baseline and whether such changes would correlate with immediate mood improvements and with clinical response at a 5-week endpoint. A priori regions of interest (ROIs) implicated in depression and its treatment were selected for focussed analyses, and broader network-level assessments of 12 canonical resting-state networks were also performed.

This was an open-label clinical study of psilocybin for treatment-resistant major depression, conducted under ethical and regulatory approvals in the UK. Nineteen patients underwent pre-treatment and one-day post-treatment fMRI scanning; after quality control, 16 subjects were retained for arterial spin labelling (ASL) CBF analyses and 15 for BOLD RSFC analyses. The primary clinical outcome measure was the 16-item Quick Inventory of Depressive Symptoms (QIDS-SR16); a revised 24-hour version of the QIDS-SR16 was used to index state mood at the post-treatment scan. Clinical relationships were assessed both contemporaneously (Pearson's r for change from pre-treatment to scan 2) and for longer-term outcome (patients were split into responders and non-responders at 5 weeks, defined as >50% reduction in QIDS-SR16, and one-tailed t-tests were used for imaging comparisons given directional hypotheses). Imaging acquisition used a 3T Siemens Tim Trio scanner with a 12-channel head coil. CBF was measured with ASL and functional connectivity with BOLD resting-state scans. Four analysis packages were used (FSL, AFNI, Freesurfer, ANTS) and preprocessing included motion correction, scrubbing using a frame-wise displacement (FD) threshold of 0.5 with exclusion if >20% volumes were scrubbed, spatial normalisation to 2 mm MNI space, band-pass filtering (0.01–0.08 Hz), and regression of nine nuisance regressors (six motion and three anatomical regressors: ventricles, draining veins, and local white matter). Seed-based RSFC analyses used four a priori seeds: bilateral parahippocampus (PH), ventromedial prefrontal cortex (vmPFC), subgenual anterior cingulate cortex (sgACC), and bilateral amygdala; seeds were derived from standard atlases or prior work. Subject-level GLMs were run and higher-level comparisons used mixed-effects (FLAME 1 + 2), cluster-corrected (z > 2.3, p < 0.05). Network-level analyses employed 12 canonical resting-state networks (RSNs) identified from independent component analysis on Human Connectome Project data. Within-network integrity was quantified by dual regression to derive mean parameter-estimate (PE) values per RSN and paired t-tests compared pre- and post-treatment conditions; Bonferroni correction was applied across 11 RSNs (DMN was treated as a priori and not corrected). Between-network connectivity was computed as symmetric 12 × 12 PE matrices using GLMs (rather than Pearson correlations) and paired permutation tests (5000 permutations) tested condition differences. An exploratory analysis also examined whether acute-session ratings of 'peak' or 'mystical-type' experience covaried with post-acute RSFC changes.

After data quality exclusions, 16 patients contributed to ASL CBF analyses (mean age 42.8 ± 10.1 years, 4 females) and 15 to BOLD RSFC analyses (mean age 42.8 ± 10.5 years, 4 females). Treatment with psilocybin was associated with rapid and statistically significant reductions in depressive symptoms. For the ASL sample, mean QIDS-SR16 dropped from 16.9 ± 5.1 (week prior to pre-treatment scan) to 8.8 ± 6.2 at the day of the post-treatment scan (change = -8.1 ± 6, t = -5.2, p < 0.001). From screening baseline (18.9 ± 3) to 5 weeks post-treatment the mean score fell to 10.9 ± 4.8 (change = -8 ± 5.1, t = -6.3, p < 0.001). In the BOLD sample, mean change values were -7.3 ± 5.3 (scan 1 to scan 2) and -8.2 ± 5.2 (baseline to 5 weeks), both highly significant (p < 0.001). Across the full cohort, all 19 patients showed some reduction in depressive symptoms at 1 week, with 12 meeting the response criterion, and at 5 weeks 47% met criteria for response. Whole-brain ASL contrasts revealed only decreases in absolute CBF post-treatment; significant clusters included left Heschl's gyrus, left precentral gyrus, left planum temporale, left superior temporal gyrus, left amygdala, right supramarginal gyrus and right parietal operculum. A decrease in amygdala CBF correlated with contemporaneous reductions in depressive symptoms between pre-treatment and one-day post (r = 0.59, p = 0.01). When responders and non-responders at 5 weeks were compared for CBF change, no significant group difference was detected (t = 0.11; p = 0.46). Seed-based RSFC analyses produced several regionally specific effects. Increased sgACC–posterior cingulate cortex/precuneus (PCC) RSFC was seen post-treatment, but this change did not correlate with symptom reductions at scan 2 nor predict 5-week response. Increased vmPFC–bilateral inferior-lateral parietal cortex (ilPC) RSFC was observed post-treatment; this increase did not correlate with immediate symptom change but was greater in responders at 5 weeks (t = 2.1; p = 0.03). Decreased parahippocampus (PH)–prefrontal cortex (PFC) RSFC was also observed post-treatment and was associated with treatment response at 5 weeks (t = -1.9; p = 0.04), though it did not correlate with immediate symptom change. Amygdala RSFC showed no significant post-treatment change. Network-level results indicated increases in within-network RSFC for the default-mode network (DMN) (t = 2.7, p = 0.018), dorsal attention network (DAN) (t = 2.2, p = 0.042) and posterior opercular network (POP) (t = 2.7, p = 0.016) after treatment; however, these effects did not survive Bonferroni correction (revised α = 0.0042) and did not correlate with clinical outcomes. Between-network analyses found decreased DMN–right frontoparietal (rFP) RSFC (t = -3.6, p = 0.0031) and increased sensorimotor–rFP RSFC (t = 2.2, p = 0.045), but these did not survive false-discovery-rate correction and were not related to symptom change or response. In an exploratory analysis, higher scores on an acute 'peak' or 'mystical' experience factor from the high-dose session were associated with larger decreases in PH RSFC with limbic regions (including bilateral amygdala) and DMN-related cortical regions (including the PCC).

The investigators interpret their findings as evidence that post-acute brain changes one day after a high-dose psilocybin session differ markedly from the acute psychedelic state. Whereas the acute state is commonly associated with modular disintegration and increased global integration, the post-treatment pattern showed trends towards modular reintegration and minimal effects on global integration or segregation. In particular, decreases in CBF were found bilaterally in temporal cortex including the left amygdala, and the magnitude of amygdala perfusion reduction related to immediate mood improvement. Given prior reports of elevated amygdala blood flow and metabolism in depression, the authors view this as potentially remedial. An increase in DMN integrity was observed post-treatment across seed and network analyses, which contrasts with previous demonstrations of decreased DMN integrity during the acute psychedelic state. The authors note that reports of DMN abnormalities in depression are mixed; some studies have shown reduced DMN connectivity in patients that normalises following successful treatment such as electroconvulsive therapy. On this basis, Carhart-Harris and colleagues propose a 'reset' model whereby an acute period of modular disintegration permits subsequent re-integration or normalisation of networks like the DMN, accompanied by clinical improvement. Supporting this view, greater post-treatment increases in vmPFC–ilPC RSFC (a DMN node pairing) and greater decreases in PH–PFC RSFC were both predictive of sustained response at 5 weeks. The discussion acknowledges inconsistencies in the literature regarding acute and post-acute CBF changes across different psychedelics and administration routes, and suggests that future studies aiming to test the reset model should include both acute and post-acute imaging, ideally emphasising BOLD RSFC and EEG as more direct indices of brain function than CBF. Key limitations reported by the authors are the small sample size and the absence of a control condition; they also note that correction for multiple comparisons was applied to the full RSN analysis but not to the specific hypothesis-driven ROI tests. Finally, the investigators recommend that future work use more rigorous controls and disentangle the relative contributions of drug effects and the accompanying psychological support.