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

Psilocybin with psychological support is showing promise as a treatment model in psychiatry but its therapeutic mechanisms are poorly understood. Here, cerebral blood flow (CBF) and blood oxygen-level dependent (BOLD) resting-state functional connectivity (RSFC) were measured with functional magnetic resonance imaging (fMRI) before and after treatment with psilocybin (serotonin agonist) for treatment-resistant depression (TRD). Quality pre and post treatment fMRI data were collected from 16 of 19 patients. Decreased depressive symptoms were observed in all 19 patients at 1-week post-treatment and 47% met criteria for response at 5 weeks. Whole-brain analyses revealed post-treatment decreases in CBF in the temporal cortex, including the amygdala. Decreased amygdala CBF correlated with reduced depressive symptoms. Focusing on a priori selected circuitry for RSFC analyses, increased RSFC was observed within the default-mode network (DMN) post-treatment. Increased ventromedial prefrontal cortex-bilateral inferior lateral parietal cortex RSFC was predictive of treatment response at 5-weeks, as was decreased parahippocampal-prefrontal cortex RSFC. These data fill an important knowledge gap regarding the post-treatment brain effects of psilocybin, and are the first in depressed patients. The post-treatment brain changes are different to previously observed acute effects of psilocybin and other ‘psychedelics’ yet were related to clinical outcomes. A ‘reset’ therapeutic mechanism is proposed.

'desegregation' 14 . These effects have been found to correlate with important aspects of the 'psychedelic experience' , including 'ego-dissolution' 14,17,19 , and were predictive of post-acute changes in the personality domain 'openness' 20 . To our knowledge, only one other very recent study has investigated >12 hour post-acute effects of psychedelics on human brain function (although see 12 and now 21 ), and few have looked at anatomical changes possibly related to psychedelic use 22,23 .
The present study focused on changes in brain function before versus after psilocybin in patients with treatment-resistant depression who received two doses of the drug (10 mg followed by 25 mg, one-week apart) as part of an open-label clinical trial. Arterial spin labelling (ASL) and blood oxygen level dependent (BOLD) resting state functional connectivity (RSFC), were used to measure changes in cerebral blood flow (CBF) and functional connectivity before (baseline) and one-day after treatment with psilocybin (i.e. one day after the 25 mg dose). It has been suggested that the days subsequent to a psychedelic experience constitute a distinct phase, referred to as the 'after-glow' , that is characterised by mood improvements and stress relief 24 . The rationale for scanning one-day post-treatment was to capture brain changes related to this so-called after-glow that might correlate with current mood improvements and/or longer-term prognoses. We predicted that resting-state CBF and FC would be altered post treatment and correlate with immediate and longer-term clinical improvements.
With regards to 'longer-term' clinical outcomes, we chose to focus on a 5-week post-treatment endpoint due to a virtual 50:50 split between responders and non-responders at this time-point (QIDS-SR16) and that none of the patients went on to additional (and thus, confounding) treatments within this time frame. A select number of regions of interest were chosen a priori for CBF and RSFC analyses due to previous work implicating their involvement in depression and its treatment, e.g [25][26][27] .

Results
Nineteen patients with diagnoses of treatment resistant major depression completed pre-treatment and one-day post-treatment fMRI scanning. Excessive movement or other artefact meant that three patients were removed from the ASL analyses and four from the RSFC (SI Appendix), leaving sample sizes of 16 (mean age = 42.8 ± 10.1 y, 4 females) and 15 (mean age = 42.8 ± 10.5 y, 4 females) for the ASL and BOLD analyses, respectively.
Treatment with psilocybin produced rapid and sustained antidepressant effects. For the patients included in the ASL analysis (minus one patient whose scan 1 rating was not collected), the mean depression score (QIDS-SR16) for the week prior to the pre-treatment scan was 16.9 ± 5.1, and for the day of the post-treatment scan, it was 8.8 ± 6.2 (change = −8.1 ± 6, t = −5.2, p < 0.001). The mean QIDS-SR16 score at baseline (screening) was 18.9 ± 3, and for 5-weeks post-treatment, it was 10.9 ± 4.8 (change = −8 ± 5.1, t = −6.3, p < 0.001).
Whole-brain CBF was calculated pre and post treatment and contrasted (Fig. 1). Only decreases in CBF were observed post treatment (vs pre), and these reached statistical significance in the left Heschl's gyrus, left precentral gyrus, left planum temporale, left superior temporal gyrus, left amygdala, right supramarginal gyrus and right parietal operculum (Table S1). Based on previous findings of increased amygdala blood flow and metabolism in depression 25 , reductions in amygdala CBF were compared with the reductions in depressive symptoms between scan 1 and 2 (i.e. decreased depressed mood at the time of scanning), and a significant relationship was found (r = 0.59; p = 0.01). After splitting the sample into responders and non-responders at 5-weeks post-treatment, and then comparing CBF changes in a t-test, no significant difference was found (t = 0.11; p = 0.46).
Increased vmPFC RSFC was observed with the bilateral inferior-lateral parietal cortex (ilPC) post-treatment. This effect did not correlate with reductions in depressive symptoms between scan 1 and 2 (r = −0.26; p = 0.17) but did predict treatment response at 5 weeks, with responders showing significantly greater vmPFC-ilPC RSFC increases than non-responders (t = 2.1; p = 0.03).
Decreased PH RSFC was observed with a PFC cluster incorporating the lateral and medial prefrontal cortex. This effect did not correlate with reductions in depressive symptoms between scan 1 and 2 (r = 0.08; p = 0.38) but did relate to treatment response at 5 weeks, with responders showing significantly greater PH-PFC RSFC decreases than non-responders (t = −1.9, p = 0.04). Amygdala RSFC was not significantly altered post treatment.
Lastly, based on indications from previous work 4,5,10 we explored the possibility that the quality of the acute 'psychedelic' experience may have mediated the post-acute brain changes. We focused on a rating scale factor related to 'peak' or 'mystical' experience and used scores for the high-dose psilocybin session as a covariate in a Figure 1. Whole-brain cerebral blood flow maps for baseline versus one-day post-treatment, plus the difference map (cluster-corrected, p < 0.05, n = 16). Correlation chart shows post-Treatment changes in bilateral amygdala CBF versus changes in depressive symptoms (r = 0.59, p = 0.01). One patient failed to completed the scan 2 QIDS-SR16 rating, reducing the sample size to n = 15 for the correlation analysis. In all of the images, the left of the brain is shown on the left. PH RSFC analysis. The PH was specifically chosen due to previous work implicating its involvement in related states 14 . Results revealed that patients scoring highest on 'peak' or 'mystical' experience had the greatest decreases in PH RSFC in limbic (e.g. bilateral amygdala) and DMN-related cortical regions (e.g. the PCC). See the supplementary file for the relevant maps and discussion.

Discussion
The present study goes some way to addressing an important knowledge gap concerning the post-acute brain effects of serotonergic psychedelics. Its findings suggest that changes in brain activity observed just one-day after a high dose psychedelic experience are very different to those found during the acute psychedelic state. Specifically, whereas the acute psychedelic state in healthy volunteers is characterised by modular disintegration 14,15,28 and global integration 14,19,29 , there are trends towards modular (re)integration and minimal effects on global integration/segregation post psilocybin for depression. Relating the blood flow findings to what has been seen previously in the acute psychedelic state is somewhat more complicated due to inconsistencies in this literature -likely due to analysis approaches and interpretation 14,15,30 : Here we saw decreased CBF bilaterally in the temporal lobes, including the left amygdala one-day post treatment. Decreased absolute CBF in subcortical and high-level association cortices have been previously reported with intravenous (I.V.) 15 and now oral psilocybin 30 but increased CBF and metabolism have also been reported with I.V. LSD 14 , oral psilocybin 31 , and oral ayahuasca 32 .
Much recent research has focused on the involvement of the default-mode network in psychiatric disorders 33 , and particularly depression 34,35 . We previously observed decreased DMN functional integrity under psilocybin 15 and LSD 14 , and others have with ayahuasca 28 . Here however, increased DMN integrity was observed one-day post treatment with psilocybin, both via seed (i.e. vmPFC and sgACC) and network-based approaches. Previous work has suggested that increased DMN integrity may be a marker of depressed mood and specifically, depressive rumination 34,36 . On this basis, increased DMN integrity post psilocybin may be surprising. The post-treatment increases in within-DMN RSFC and sgACC-PCC RSFC did not relate to symptom improvements but vmPFC-ilPC RSFC did (see Fig. 3). This apparent divergence from previous findings 36,37 is intriguing, and deserves further discussion (below).
It should be noted that findings of elevated within-DMN RSFC in depression are not entirely consistent in the literature [38][39][40][41] . For example, using a DMN-focused analysis, precuneus-DMN RSFC 39 was found to be lower in patients than in healthy controls, and normalised after treatment with electroconvulsive therapy (ECT) -and only in responders 39 -consistent with the present findings. Lower precuneus-DMN RSFC in depression was also seen in a separate study and the degree of this abnormality correlated with autobiographical memory deficits 40 . In another study, lower PCC-dmPFC and PCC-ilPC RSFC were seen in first-episode depressed patients relative to healthy controls 41 . In the present study, we saw increased within-DMN RSFC post treatment with psilocybin, and increased vmPFC-bilateral ilPC RSFC was predictive of treatment response at 5 weeks (Fig. 3). These findings suggest a commonality in the antidepressant action of ECT and psilocybin 39 in which DMN integrity is decreased acutely (at least by the latter 14,15,28 ) and increased (or normalised) post-acutely, accompanied by improvements in mood. This process might be likened to a 'reset' mechanism in which acute modular disintegration (e.g. in the DMN) enables a subsequent re-integration and resumption of normal functioning.
Recent meta-analyses of studies of resting-state CBF in depression have yielded relatively mixed results 34,42 , although findings of increased thalamic 34,42 and sgACC metabolism are relatively consistent 34 . Here, we did not find any post-treatment changes in thalamic or sgACC CBF with psilocybin, either in whole-brain or ROI-based analyses. We did observe decreased CBF bilaterally in the temporal cortex however, including the left medial temporal lobe and specifically, the left amygdala. Given previous findings of elevated resting-state amygdala CBF and metabolism in mood disorders 25,43,44 , the reduction in amygdala CBF observed here, and its relation to symptom severity, could be viewed as a possible remediation effect. Moreover, generalised decreases in CBF are (again) consistent with what has been previously reported with ECT 45 , i.e. most studies have documented an increase in CBF in the acute 'ictal' state, including in the amygdala 45 ; however, the post-ictal period is characterised by decreased CBF, and often in those regions that were most perfused during seizure 45 . Acutely increased CBF has previously been reported with ayahuasca 32 and LSD 15 and increased glucose metabolism has been observed in the acute state with oral 31 but not I.V. psilocybin 15 . Thus, a post-acute reversal of acute increases in CBF could be seen as consistent with the post-treatment 'reset' mechanism proposed above -although recent work has laid into question whether oral psilocybin does indeed cause increases in brain absolute CBF 30 . It would be challenging (but not impossible) to carry out acute and post-acute imaging in future trials of psilocybin for depression, and this may be necessary if the 'reset' model is to be properly tested. In such a study, we would advise focusing on BOLD RSFC (and perhaps simultaneous EEG-related measures) rather than CBF, due to RSFC and EEG offering more direct and reliable indices of brain activity and function than more difficult to interpret measures such as CBF. The inclusion of a healthy control group, exposed to a consistent treatment procedure, would further strengthen the design of such a study, as would the inclusion of a placebo and/or active comparator arm.
The present study's other major positive finding was a decrease in RSFC between the bilateral parahippocampus and the PFC, an effect that (like increased vmPFC-ilPC RSFC) was predictive of treatment response at 5 weeks. Curiously, a post-hoc exploratory analysis suggested that acute 'peak' or 'mystical-type' experiences under psilocybin may mediate the post-acute changes in parahippocampal RSFC (including decreased PH-PCC RSFC). Focusing on parahippocampal-PFC RSFC, this has generally been found to be elevated in depression 46 , and consistently so across the duration of a resting-state scan 47 . Prefrontal-limbic circuitry has been linked with top-down suppression of affective responsiveness 48 and lower resting-state amygdala-vmPFC RSFC in combination with amygdala hyperfusion was found to relate to state-anxiety in healthy individuals 43 , corroborating separate findings 49 . Seven days of citalopram has been found to reduce amygdala-vmPFC 50 and dorso-medial PFC-left hippocampal RSFC 51 in healthy volunteers, somewhat consistent with the present findings.
In conclusion, here we document for the first time, changes in resting-state brain blood flow and functional connectivity post-treatment with psilocybin for treatment-resistant depression. Decreased blood flow was found to correlate (in the amygdala) with reductions in depressive mood. Increased within-DMN RSFC was observed post-treatment, using both seed and network-based analyses, and specific increases in RSFC between the vmPFC and bilateral ilPC nodes of the DMN were greatest in individuals who maintained treatment-response at 5 weeks. Finally, decreased PH-PFC RSFC was observed post-treatment and this was also predictive of treatment-response at 5 weeks. An exploratory post-hoc analysis revealed that acute 'peak' or 'mystical' experience during the high-dose psilocybin session was predictive of these changes in PH RSFC.
This study is limited by its small sample size and absence of a control condition. Moreover, correction for multiple testing was applied to the full RSN but not the specific (hypothesis-based) ROI analyses. Future research with more rigorous controls should serve to challenge and develop the present study's findings and inferences. Assessing the relative contributions of, and potential interactions between, the different treatment factors (e.g. the drug and the accompanying psychological support) may be a particularly informative next step.  Imaging vs clinical outcomes. To explore relationships between significant imaging outcomes and the main clinical outcomes, we chose to focus on changes in depressive symptoms from: 1) pre-Treatment to scan 2 (i.e. one-day post-treatment), and 2) pre-Treatment to 5 weeks post-Treatment. The primary clinical outcome measure, the 16-item Quick Inventory of Depressive Symptoms (QIDS-SR16) was chosen for this purpose. Relationships between imaging outcomes and contemporaneous decreases in depressive symptoms were calculated using a standard Pearson's r, and relationships with the longer-term (i.e. at 5 weeks post-treatment) changes in depressive symptoms were calculated by splitting the sample into responders (>50% reduction in QIDS-SR16 scores) and non-responders at this time-point, and then performing a one-tailed t-test on the relevant imaging outcomes (one-tailed as directionality was unequivocally implied by the direction of the significant imaging outcome). We used a revised version of the QIDS-SR16 for 24-hour measurement for the post-treatment scan in order to get a contemporaneous, state-related index of depressive symptoms at this time-point.

Method
Anatomical Scans. Imaging was performed on a 3 T Siemens Tim Trio using a 12-channel head coil at Imanova, London, UK. Anatomical images were acquired using the ADNI-GO (Alzheimer's Disease Neuroimaging Initiative, Grand Opportunity 52 ) recommended MPRAGE parameters (1 mm isotropic voxels, TR = 2300 ms, TE = 2.98 ms, 160 sagittal slices, 256 × 256 in-plane FOV, flip angle = 9 degrees, bandwidth = 240 Hz/pixel, GRAPPA acceleration = 2). BOLD Pre-processing. Four different but complementary imaging software packages were used to analyse the fMRI data. Specifically, FMRIB Software Library (FSL) 53 , AFNI 54 , Freesurfer 55 and Advanced Normalization Tools (ANTS) 56 were used. Fifteen subjects were used for this analysis: one subject was discarded from the analysis due to an injury in parietal cortex and three subjects were discarded due to high levels of head movement. Principally, motion was measured using frame-wise displacement (FD) 57 . The criterion for exclusion was subjects with >20% scrubbed volumes with a scrubbing threshold of FD = 0.5. For the 15 subjects that were used in the analysis, there was no significant difference in the mean FD (meanFD before = 0.179 ± 0.088, mean-FD after = 0.158 ± 0.084, p = 0.23). The mean percentage of scrubbed volumes for before and after treatment was 4.6 ± 5% and 3.5 ± 5.2%, respectively (p = 0.56). The maximum of scrubbed volumes for before and after treatment was 17.3% and 17.7%, respectively. The following pre-processing stages were performed: 1) removal of the first three volumes; 2) de-spiking (3dDespike, AFNI); 3) slice time correction (3dTshift, AFNI); 4) motion correction (3dvolreg, AFNI) by registering each volume to the volume most similar, in the least squares sense, to all others (in-house code); 5) brain extraction (BET, FSL); 6) rigid body registration to anatomical scans (BBR, FSL); 7) non-linear registration to 2 mm MNI brain (Symmetric Normalization (SyN), ANTS); 8) scrubbing 58 -using an FD threshold of 0.5, scrubbed volumes were replaced with the mean of the surrounding volumes. 9) spatial smoothing (FWHM) of 6 mm (3dBlurInMask, AFNI); 10) band-pass filtering between 0.01 to 0.08 Hz (3dFourier, AFNI); 11) linear and quadratic de-trending (3dDetrend, AFNI); 12) regressing out 9 nuisance regressors (all nuisance regressors were band-pass filtered with the same band-pass filter as above): out of these, 6 were motion-related (3 translations, 3 rotations) and 3 were anatomically-related (not smoothed). Specifically, the anatomical nuisance regressors were: 1) ventricles (Freesurfer, eroded in 2 mm space), 2) draining veins (DV) (FSL's CSF minus Freesurfer's Ventricles, eroded in 1 mm space) and 3) local white matter (WM) (FSL's WM minus Freesurfer's subcortical grey matter (GM) structures, eroded in 2 mm space). Regarding local WM regression, AFNI's 3dLocalstat was used to calculate the mean local WM time-series for each voxel, using a 25 mm radius sphere centred on each voxel 59 .

BOLD fMRI Resting State
Seed-based RSFC. Based on prior hypotheses, 4 seeds were chosen for these analyses: 1) the bilateral PH, vmPFC, sgACC and bilateral amygdala. The PH seed was constructed by combining the anterior and posterior parahippocampal gyrus from the Harvard-Oxford probabilistic atlas, which was then thresholded at 50%. The vmPFC seed was the same as one previously used by our team in analyses of the acute effects of LSD 60 , psilocybin 61 and MDMA 62 . The sgACC seed was a 5 mm sphere centred at ±2 28 -5 (MNI_152 coordinates) based on 63 . Bilateral amygdala seed was based on Harvard-Oxford probabilistic atlas, threshold at 50%. Mean time-series were derived for these seeds for each RS scan. RSFC analyses were performed using FSL's FEAT for each subject. Pre-whitening (FILM) was applied. A higher level analysis was performed to compare pre-treatment and post-treatment conditions using a mixed-effects GLM (FLAME 1 + 2), cluster corrected (z > 2.3, p < 0.05). MRIcron was used to display the results.

Integrity (within-RSN RSFC).
Network integrity was calculated for each RSN for both pre-treatment and post-treatment. All 20 HCP ICA components were entered into FSL's dual regression analysis 65 . The first step of the dual regression used the components as regressors applied to the 4D BOLD datasets for each subject, resulting in a matrix of time-series for each ICA. The second step involved regressing these time-series into the same 4D scan data to get a subject-specific set of spatial maps (parameter estimate (PE) images). For each subject and for each condition, within each of the 12 RSNs of interest (threshold = 3), the mean PE across voxels was calculated. This mean PE represents the integrity value. Subsequently, paired t-tests were used to calculate the difference in integrity between conditions for each RSN (Bonferroni corrected for 11 RSNs, with no correction for DMN as we had a prior hypothesis).

Segregation (between-RSN RSFC).
Between-RSN RSFC was calculated in a similar manner to previous analyses involving acute LSD 60 and psilocybin 66 . Specifically, a 12 × 12 matrix was constructed representing RSFC between different RSN pairs. For each subject and for each condition, the time-series for the relevant pair of RSNs, was entered into a GLM, resulting in a PE value representing the strength of functional connectivity between them. GLM was used rather than correlation coefficients because differences between Pearson's correlations could be a result of either signal or noise differences; therefore, it is preferable to perform regression and look for pre-treatment and post-treatment differences on the PE 67 . The GLM was estimated twice: 1) each RSN as a dependant variable in one model, and 2) each RSN as an independent variable in the second model. These two PE values were then averaged together, to generate a symmetric 12 × 12 matrix (Fig. 4b). Three 12 × 12 matrices were created as follows: 1) the group mean PE values for pre-Treatment treatment, 2) the group mean PE values for post-Treatment treatment, and 3) paired t-test to compare the PE values for the two conditions, pre-Treatment and post-Treatment treatment (two-tailed, 5000 permutations).