Negative affective bias in depression following treatment with psilocybin or escitalopram - a secondary analysis from a randomized trial

The authors situate the study against persistent shortcomings of current antidepressant strategies, notably selective serotonin reuptake inhibitors (SSRIs): partial efficacy, adverse effects and relatively slow onset of action. There is increasing interest in psychedelic compounds such as psilocybin — a 5-HT2A receptor agonist — which have shown promise in recent trials for producing relatively rapid antidepressant effects. However, the neuropsychological mechanisms by which psilocybin might reduce depressive symptoms are unclear, and it is not established whether these mechanisms overlap with those of conventional antidepressants such as escitalopram. G. and colleagues set out to compare psilocybin and escitalopram with respect to effects on emotional information processing, using an experimental medicine framework that views early shifts in emotional bias (the relative processing of positive versus negative emotional information) as a potential common mechanism of antidepressant action. Specifically, the study tested whether two doses of psilocybin plus placebo or a low psilocybin dose plus 6 weeks of escitalopram would produce a more positive behavioural bias on the Facial Emotion Recognition Task (FERT), and whether changes in that bias related to clinical change in depressive symptoms over the trial and at one-month follow-up.

This secondary analysis used data from a Phase II, double-blind, randomised controlled trial in patients with long-standing, moderate-to-severe major depressive disorder. Participants were randomised 1:1 to either: (a) two doses of 25 mg psilocybin given three weeks apart plus six weeks of daily placebo (psilocybin group, N=30), or (b) two doses of 1 mg psilocybin given three weeks apart plus six weeks of daily oral escitalopram (escitalopram group, N=29; 10 mg for the first 3 weeks, 20 mg for the following 3 weeks). All participants received the same preparatory and integration psychological support. Randomisation was performed by individuals not part of the research team and participants were informed they would receive psilocybin without dose disclosure to limit expectation effects. Behavioural assessment used the Facial Emotion Recognition Task (FERT), administered at baseline (visit 1) and at the primary post-treatment endpoint (visit 6; six weeks after first dosing and three weeks after the second dose). The FERT presents faces displaying six basic emotions (anger, disgust, fear, happiness, sadness, surprise) morphed across intensities; primary outcomes were accuracy, misclassifications (how often a face was classified as a particular emotion), and reaction time for correct responses. Signal detection measures (target sensitivity and response bias) were also computed. Emotions were grouped into negative (anger, disgust, fear, sadness) and positive (happy, surprise); an improvement in negative bias was defined as a statistically significant reduction over time in metrics reflecting preferential processing of negative versus positive emotions. Clinical symptomatology was measured with the Quick Inventory of Depressive Symptomatology-Self Report (QIDS-SR-16) at baseline, week 6 (primary clinical endpoint) and monthly thereafter, including a remote assessment at week 10. Behavioural data were analysed in IBM SPSS using a three-way mixed ANOVA with group (psilocybin vs escitalopram) as the between-subjects factor and visit (baseline, post-treatment) and emotion/valence as within-subjects factors. Analyses were conducted on an intention-to-treat basis, with additional sensitivity analyses reported for per-protocol participants. The extracted text does not clearly report detailed handling of missing data or the exact multiple-comparison correction strategy used for post-hoc tests.

The final analysed sample comprised 59 participants (n=30 psilocybin, n=29 escitalopram). Protocol non-completion occurred in both groups, largely due to adverse events, medication non-adherence, and COVID-19–related disruptions; all randomised participants were included in the intention-to-treat analyses. The authors report no statistically significant group × time or group × time × emotion interactions on any FERT outcome measures, indicating no detectable behavioural differences between psilocybin and escitalopram on the task (reported F and p ranges: accuracy F's < 0.52, p's > 0.475; misclassifications F's < 0.11, p's > 0.746; reaction times F's < 2.23, p's > 0.141; response bias F's < 0.89, p's > 0.349; target sensitivity F's < 0.58, p's > 0.448). Across both treatment groups combined, there was evidence of a reduced negative bias from baseline to week 6. For accuracy, a significant time × emotion interaction was reported (F(1,55)=30.44, p<0.001, η2=0.36) with post-hoc tests indicating improved recognition for negative emotions (reported t(56)=6.39, p<0.001, d=0.85; the extracted text does not clearly report the full confidence interval for this effect). Misclassifications also showed a significant time × emotion interaction (F(1,55)=5.17, p=0.027, η2=0.086), with a reduction in misclassifications of negative emotions post-treatment (reported t(56)=2.81, p<0.001, d=0.37, 95% CI: 0.15–0.89). Directional analyses indicated both a decrease in positive→negative substitutions over time (F(1,55)=37.61, p<0.001, η2=0.41) and an increase in negative→positive substitutions (F(1,55)=18.67, p<0.001, η2=0.25). Reaction times for accurate trials showed a time × emotion interaction (F(1,55)=5.11, p=0.028, η2=0.085); participants were faster post-treatment for negative emotions (t(56)=3.28, p=0.002, d=0.43) and even faster for positive emotions (t(56)=5.57, p<0.001, d=0.74). Response bias and target sensitivity also demonstrated significant time × emotion interactions (response bias: F(1,55)=7.50, p=0.008, η2=0.12; target sensitivity: F(1,55)=29.45, p<0.001, η2=0.35), with post-hoc tests indicating reduced response bias and lower target sensitivity for negative emotions post-treatment. Regarding associations with clinical change, there were no significant concurrent correlations between change in negative affective bias at week 6 and change in depressive symptoms at week 6 (QIDS-SR-16). Longitudinally, in the escitalopram group only, a decrease in misclassifications of positive faces as negative at week 6 correlated with subsequent depression improvement at week 10 (r(18)=0.498, p=0.025), whereas this association was not observed in the psilocybin group (r(17)=-0.195, p=0.423); the difference between these correlation coefficients was reported as statistically significant (Z=2.20, p=0.028). No other significant associations were reported in the extracted text. Sensitivity analyses restricted to per-protocol participants were conducted, but the extracted text does not report detailed outcomes for those analyses.

G. and colleagues interpret the findings as showing that both psilocybin and escitalopram are associated with a shift toward more positive (or less negative) behavioural biases in emotional processing over six weeks, as measured by the FERT, and that there were no detectable differences between the two treatments on these behavioural measures. This provides experimental medicine evidence for overlap in at least some cognitive effects of a psychedelic treatment and a conventional SSRI, despite different pharmacologies and dosing regimens. The authors place these results in the context of models proposing that early shifts in emotional information processing precede and contribute to later clinical improvement. They note that, consistent with these models, change in negative bias at week 6 predicted later clinical benefit (week 10) in the escitalopram arm, supporting an environment × bias mechanism for SSRI action. By contrast, no such longitudinal association was found for psilocybin; the authors suggest this may indicate that change in emotional bias is less central (or only one among several mechanisms) to psilocybin's therapeutic effects. They discuss possible mechanistic differences, including direct 5-HT2A agonism, effects on neural or network flexibility, and more rapid effects on neural plasticity with psilocybin, as potential explanations for differences in timing and pathways to clinical change. The authors also reference neuroimaging findings from the larger trial and other studies showing divergent neural responses: escitalopram was associated with reduced BOLD responses to faces (including amygdala), whereas psilocybin showed different or more heterogeneous amygdala and connectivity effects in acute and post-acute windows. The authors acknowledge several limitations. The trial lacked an inert placebo arm, limiting causal inference about treatment effects versus non-specific factors. Sample bias may have favoured participants more open to receiving psilocybin, potentially affecting generalisability. Practice effects on the FERT and the possibility of being underpowered to detect between-group differences are also noted. There were protocol deviations and missing data related to adverse events and COVID-19 disruptions. Functional unblinding was not assessed, and although expectancy did not predict response in the psilocybin arm, unmeasured expectancy or unblinding effects could still influence outcomes. Strengths cited include use of a well-validated, pharmacologically sensitive assay and deployment of an experimental medicine approach that can be less susceptible to placebo effects. The authors conclude that further studies examining early temporal effects and neuropsychological pathways for both interventions are required to clarify shared and distinct mechanisms and their relation to clinical benefit.