Direct serotonin release in humans shapes aversive learning and inhibition

Understanding central serotonin (5-HT) has been a longstanding goal due to its role in many psychiatric drugs (SSRIs, MDMA, LSD) and its phylogenetically conserved influence on behaviours including feeding, sexual function, cognition, behavioural inhibition, memory, and aversive processing. Human studies manipulating synaptic 5-HT with SSRIs or tryptophan depletion have yielded inconsistent behavioural effects across reinforcement learning, behavioural inhibition, and memory. SSRIs have complex acute and subchronic effects, including autoreceptor-mediated feedback limiting 5-HT release and interactions with dopaminergic systems, complicating causal inference about serotonin’s specific behavioural functions. Selective serotonin releasing agents (SSRAs) provide an alternative probe that directly stimulate exocytic 5-HT release without broad monoaminergic efflux. Low-dose fenfluramine, recently licensed for epilepsy, appears to increase synaptic 5-HT rapidly and selectively without increasing extracellular dopamine, offering an opportunity to examine the direct effects of elevated synaptic 5-HT on aversive processing, behavioural inhibition, and memory. The study tests the hypothesis that fenfluramine would produce behavioural patterns opposite to tryptophan depletion, specifically reduced loss sensitivity, enhanced inhibition, and improved verbal memory.

Prior research indicates core functions of serotonin include behavioural inhibition, memory, and aversive processing. Human studies with SSRIs report mixed effects on reinforcement sensitivity (increases in reward, increases in loss, decreases in reward, or decreases in both), and variable impacts on inhibition and memory. Mechanistically, SSRIs’ effects are confounded by 5-HT autoreceptor feedback that reduces firing-dependent release and by dopamine transporter-mediated uptake and co-release, affecting striatal, prefrontal, and hippocampal circuits. Preclinical and human imaging work shows fenfluramine increases synaptic 5-HT robustly (up to ~200% vs baseline), reduces [18F]altanserin binding consistent with increased 5-HT, and is inactive at dopaminergic synapses at low doses. Tryptophan depletion tends to enhance punishment learning and aversive biases. Optogenetic and pharmacological studies suggest DRN serotonergic neurons may modulate aversive prediction errors more than reward, while non-serotonergic DRN stimulation can increase reward preference. The contradictory SSRI findings may reflect off-target effects beyond serotonin; SSRAs may isolate serotonin’s role, particularly in loss sensitivity and inhibition.

Design: Double-blind, randomized, placebo-controlled study with subchronic administration (8 ± 1 days). Participants: 56 randomized (28 fenfluramine, 28 placebo); final analyzed sample N = 53 young, non-clinical adults (mean age ~20.15; 32 female; fenfluramine n = 26; placebo n = 27). Inclusion: BMI 18–30, fluent English; Exclusion: recent recreational drugs (3-month washout; ≥1 year for MDMA), pregnancy/breastfeeding, medical contraindications. Screening: medical history, SCID-5, cardiovascular (BP, ECG), renal/liver panels, drug and pregnancy tests. Intervention: Fenfluramine hydrochloride (racemic) 15 mg b.i.d. or matched placebo, oral solution, for 8 ± 1 days. Randomization: stratified block (gender and task stimulus version); conducted by Clinical Pharmacy Support Unit (Oxford Health NHS FT). Ethics and preregistration: University of Oxford CUREC (MSD-IDREC R69642/RE004); ClinicalTrials.gov NCT05026398. Procedures: Baseline visit with cognitive/emotional task battery and questionnaires; supervised first dose with monitoring (BP, observations) and salivary cortisol at pre-dose, 1 h, 3 h; daily dosing and daily questionnaires; follow-up visit repeating tasks/questionnaires and allocation guess. Task battery: 1) Auditory Verbal Learning Task (AVLT) assessing declarative memory encoding and retrieval (accuracy). 2) Affective Interference Go/No-Go Task assessing behavioural inhibition under affective distractors (fearful/happy/neutral), with measures of response inhibition (no-go accuracy), go-trial accuracy and response times (impulsivity index), and set-shifting; drift diffusion modelling (DDM) estimated boundary separation (a), non-decision time (Ter), initial bias (za), drift rate (v), and drift criterion (dc). 3) Verbal n-back (0-,1-,2-,3-back) assessing complex verbal working memory (target accuracy, response time). 4) Probabilistic instrumental learning task (win vs loss trials; 70/30 probabilities) assessing reinforcement sensitivity; reinforcement learning (Q-learning) model estimated outcome sensitivity (rho) and learning rate (alpha, separately for win/loss). 5) Oxford Memory Task (visuospatial working memory: localisation speed, selection accuracy). 6) Contextual Cueing Task (implicit visual learning: accuracy, response times). Questionnaires: STAI-T, STAI-S, BDI-II, PANAS, VAS (daily), PDQ-D, Side Effects Profile. Data and Code: Raw/modelled datasets on Zenodo and GitHub; analysis code available. Statistical analysis: Baseline-adjusted mixed-effects ANCOVA on post-intervention outcomes (two-tailed), with baseline score as covariate and participant as random effect for multilevel structures; estimated marginal means (EMM) tests with Bonferroni-Holm correction; model diagnostics included residual normality, covariate independence and homogeneity of regression slopes; DDM fit with G^2 maximum likelihood using RT quantiles (10th, 30th, 50th, 70th, 90th); Pearson correlations for post-hoc analyses; chi-squared for allocation guess; linear mixed-effects models for longitudinal daily VAS and side effects, and for salivary cortisol across timepoints. Alpha = 0.05. A priori power: N = 52 (26/group) for 80% power.

Reinforcement learning: Fenfluramine reduced optimal choices in loss trials and decreased modelled outcome sensitivity for losses, with slower choices during loss conditions. - Optimal choice: group × condition interaction F[1,50] = 5.14, p = 0.03, ηp² = 0.07; loss EMM = −8.62 ± 3.18, p < 0.01, d = −0.75; reward EMM = 0.68 ± 3.18, p = 0.83. - Outcome sensitivity (rho): group × condition F[1,50] = 5.73, p = 0.02, ηp² = 0.10; loss EMM = −0.90 ± 0.43, p = 0.04, d = −0.57; reward EMM = 0.10 ± 0.43, p = 0.82. - Learning rate (alpha): no significant group effects (F[1,50] = 0.92, p = 0.34; interaction F[1,50] = 1.22, p = 0.27). - Time to choice: group × condition F[1,50] = 5.52, p = 0.02, ηp² = 0.11; loss EMM = 246.0 ± 95.6 ms, p = 0.01, d = 0.71; reward EMM = 13.9 ± 95.6 ms, p = 0.89. Behavioural inhibition and impulsivity (Affective Go/No-Go): Fenfluramine improved response inhibition, increased cautious decision criterion, and reduced choice impulsivity (longer go-trial RT), especially under aversive interference. - Response inhibition (no-go accuracy): main effect F[1,47] = 11.26, p < 0.01, ηp² = 0.15; overall EMM = 9.69 ± 2.63, p < 0.001, d = 0.60; positive interference EMM = 11.25 ± 4.56, p = 0.015, d = 0.69; aversive interference EMM = 9.25 ± 4.56, p = 0.044, d = 0.58; control EMM = 8.58 ± 4.56, p = 0.062. - Go-trial accuracy: no significant group effect (F[1,47] = 0.83, p = 0.37). - Signal detection criterion (c): main effect F[1,47] = 13.54, p < 0.001, ηp² = 0.19; overall EMM = 0.08 ± 0.02, p < 0.001, d = 0.39; discriminability not significantly different. - Choice impulsivity (go-trial RT): main effect F[1,47] = 22.00, p < 0.001, ηp² = 0.27; overall EMM = 17.2 ± 2.72 ms, p = 3.73e−09, d = 1.03; interaction F[2,95] = 3.22, p < 0.05, ηp² = 0.08; largest effect under aversive distractors EMM = 21.3 ± 4.71 ms, p < 0.0001, d = 1.28 (vs control EMM = 14.6 ± 4.71 ms, p < 0.01, d = 0.88; positive EMM = 15.4 ± 4.71 ms, p < 0.01, d = 0.93). - Correlations: go RT with response inhibition r = 0.61, p < 0.001; with criterion c r = 0.77, p < 0.001. Drift diffusion modelling: Under aversive interference, fenfluramine shifted initial bias toward the no-go boundary, consistent with reduced impulsivity. - Initial choice bias (za): group × condition F[2,95] = 3.45, p = 0.03, ηp² = 0.06; aversive EMM = −0.33 ± 0.15, p = 0.03, d = −0.60; control EMM = −0.01 ± 0.15, p = 0.97; positive EMM = −0.16 ± 0.15, p = 0.31; other parameters (a, v) not significantly different. Memory: Fenfluramine enhanced verbal processing speed at highest working memory load and improved delayed recall accuracy; no effects on visuospatial working memory or implicit visual learning. - Verbal n-back target RT: group × condition F[3,143] = 3.66, p = 0.01, ηp² = 0.05; 3-back EMM = −118.2 ± 50.2 ms, p = 0.02, d = −0.67; 0-back EMM = 35.9 ± 50.2, p = 0.48; 1-back EMM = −16.7 ± 50.2, p = 0.74; 2-back EMM = −51.8 ± 50.2, p = 0.30; accuracy not different (F[1,47] = 0.70, p = 0.41). - AVLT delayed recall: group × condition F[2,1474] = 6.23, p = 0.01, ηp² < 0.01; delayed recall EMM = 0.84 ± 0.35, p = 0.02, d = 0.34; learning EMM = −0.07 ± 0.14, p = 0.63; distraction recall EMM = −0.90 ± 0.70, p = 0.20. - No significant group effects on Oxford Memory Task and Contextual Cueing Task. Self-report and physiological measures: No significant group effects on most subjective measures; PANAS negative items baseline imbalance noted but no post-intervention differences; allocation guess not different; no differences in salivary cortisol across initial dosing period.

Elevating synaptic 5-HT via a selective releasing agent produced a clear reduction in sensitivity to aversive outcomes during instrumental learning, opposite to effects observed with tryptophan depletion, supporting a causal role for serotonin in modulating loss sensitivity. Reward processing remained unaffected, aligning with preclinical evidence that serotonergic DRN stimulation does not necessarily enhance reward learning, and contrasting with some SSRI findings potentially driven by dopaminergic modulation and emotional blunting. Behavioural inhibition improved through increased caution, with reduced choice impulsivity and a DDM-identified bias toward the no-go boundary under aversive interference. This pattern fits theoretical accounts of serotonin as an inhibitory neuromodulator engaged particularly in aversive contexts, yet the benefits here extended to neutral contexts as well, suggesting a broader role in limiting impulsive actions. Memory effects were selective to verbal domains, with faster processing at high working memory load and improved delayed recall, consistent with literature on 5-HT’s contributions to verbal learning. The use of fenfluramine, relatively selective for serotonergic release at low doses, helps disentangle serotonin-specific effects from SSRI-related off-target systems. Findings have implications for understanding central serotonin’s role in decision-making under aversive conditions and suggest potential avenues for targeting loss sensitivity and impulsivity in psychiatric disorders without exacerbating anhedonia.

Directly increasing synaptic serotonin with fenfluramine in healthy adults reduced sensitivity to loss outcomes, enhanced behavioural inhibition by promoting cautious decisions, shifted bias toward impulse control during aversive interference, and improved verbal memory recall and processing speed. These results clarify serotonin’s role in guiding decision-making across aversive and neutral contexts and demonstrate the utility of selective serotonin releasing agents as experimental probes. Future work should test SSRAs in clinical populations characterized by negative biases and cognitive impairments (e.g., depression, bipolar disorder, schizophrenia), investigate neurochemical specificity with PET/SPECT, examine longer treatment durations and dose–response effects, and explore therapeutic potential for disorders of impulsivity (e.g., ADHD).

Interpretation is bounded by the healthy, young, non-clinical sample and short subchronic dosing period. While fenfluramine is relatively selective at low doses, ancillary pharmacology (5-HT2A/B/C, sigma-1 receptor modulation) and metabolite norfenfluramine could contribute off-target effects. Differences across reinforcement learning paradigms (model-free vs model-based, reversal) limit direct comparison with prior SSRI/TRP studies. The DDM parameters showed bias effects primarily under aversive interference; accuracy was similar across groups for go trials, potentially constraining parameter differences. Ecological implications of reduced loss sensitivity depend on context; despite reduced optimal choices in loss trials, total monetary outcomes did not differ. Further neuroimaging is needed to confirm regional serotonergic specificity.