Spectral signatures of serotonergic psychedelics and glutamatergic dissociatives

Pallavicini and colleagues situate their work within a long-standing distinction between serotonergic psychedelics (SP) and glutamatergic dissociatives (GD), noting that both classes produce marked alterations of perception and self-experience yet have distinct primary molecular targets (5-HT2A receptor agonism for SP; NMDA receptor antagonism for GD). Earlier neuroimaging and pharmacological studies show overlapping system-level effects (for example, altered default mode network activity, reductions in lower-frequency oscillatory power and increased global connectivity/entropy), but there has been no quantitative, frequency-specific comparison of the acute neurophysiological signatures induced by representative compounds from the two classes. The study set out to compare MEG-derived, source-localised spectral power and long-range functional connectivity changes produced acutely by two serotonergic psychedelics (LSD and psilocybin) and one NMDA-antagonist dissociative (ketamine). The investigators aimed to: (1) identify band-limited power and connectivity changes for each drug versus placebo; (2) quantify the spatial similarity of those changes between drugs; and (3) test whether multivariate classifiers trained to distinguish a given drug from its placebo generalise to distinguish other drugs from their placebos. The approach emphasises frequency-specific patterns as potential convergent or divergent neurophysiological signatures tied to pharmacology and subjective effects.

The analyses used previously collected MEG datasets from three double-blind, placebo-controlled experiments. Recorded sample sizes were 15 participants for LSD, 19 for ketamine and 14 for psilocybin, with recording durations of roughly 5–7 min (LSD), 6–10 min (ketamine) and 2–5 min (psilocybin). All sessions included a corresponding saline placebo condition; participants had prior experience with at least one hallucinogen but not within 6 weeks. General exclusion criteria included age under 21, pregnancy, personal or immediate family history of psychiatric disorder, substance dependence, cardiovascular disease, claustrophobia or prior adverse response to a hallucinogen. Ketamine sessions additionally excluded smokers, women and participants with body mass index outside 18–30 kg/m2. LSD and psilocybin were administered intravenously at doses reported as 75 μg and 2 mg respectively, with MEG recorded 4 h after LSD infusion and immediately after psilocybin infusion. Ketamine was given as an initial bolus (0.25 mg/kg over 1 min) followed by a 40 min maintenance infusion at 0.375 mg/h. Ethical approval and informed consent were obtained for all studies. MEG data were acquired on a CTF 275-channel system sampled at 1200 Hz (0–300 Hz bandpass), with 29 reference channels for noise cancellation and analysis of synthetic third-order gradiometers. Data were band-pass filtered (1–150 Hz), downsampled to 600 Hz and segmented into 2 s epochs. Visual inspection and an automated muscle-artifact rejection algorithm removed contaminated epochs, and independent component analysis was used to remove residual ocular, cardiac and muscle components. Source modelling used individual MRIs and a linearly constrained minimum variance beamformer to generate broadband virtual sensor time-series at 90 cortical and subcortical regions defined by the automated anatomical labelling (AAL) atlas. Time–frequency analysis employed Hanning-windowed FFTs from 1–100 Hz at 0.5 Hz intervals, with results aggregated into six canonical bands: delta (1–4 Hz), theta (4–8 Hz), low alpha (8–10.5 Hz), high alpha (10.5–13 Hz), low beta (13–20 Hz) and high beta (20–30 Hz). Gamma was excluded because of susceptibility to muscle artefact. For each band, two‑tailed Student’s t-tests (p < 0.05, Benjamini–Hochberg false-discovery rate corrected) compared spectral power for each drug versus its placebo. Functional connectivity between each pair of AAL ROIs was computed as the linear correlation between envelopes of orthogonalised, bandpass-filtered source signals; orthogonalisation (via least-squares regression) removes spurious covariance from signal leakage while preserving interactions between independent sources. Mass univariate t-tests were run on both power and connectivity matrices; spatial similarity between drug-induced maps was assessed by computing Pearson correlations of the t-values across ROIs or ROI pairs. To probe multivariate, spatially distributed signatures, the team trained random forest classifiers (1000 trees, sqrt(number of features) features per split, Gini impurity, no maximum depth) to distinguish each drug from its placebo using either spectral power across the 90 ROIs or functional connectivity features. Classifier performance used five-fold cross-validation and area under the receiver operating characteristic curve (AUC). Statistical significance was estimated by 1000 label‑shuffling permutations to derive empirical p-values, with FDR correction. For connectivity they applied two approaches: global classifiers using all ROI–ROI values, and local classifiers trained on the connectivity profile of each single ROI versus all others (90 classifiers per drug and band, 1620 in total). Significant classifier performance was operationalised as AUC > 0.7 with p < 0.05 after FDR correction.

Spectral power: All three drugs produced widespread reductions in spectral power relative to placebo, most consistently in the low and high alpha bands and in the low beta band. Psilocybin additionally reduced high beta power, while ketamine uniquely reduced delta power. LSD produced the most extensive and largest effect-size decreases across nearly all examined bands; bilateral thalamic ROIs showed reductions in all bands except high beta. No significant increases in spectral power were reported. Spatial similarity of power changes: Simple linear comparisons of the drug-versus-placebo t-value maps yielded generally low Pearson correlations across drugs, indicating limited spatial similarity of power changes. Notable exceptions were moderate positive correlations: ketamine versus LSD in the low alpha band (R = 0.55), psilocybin versus LSD in the theta band (R = 0.44), LSD versus psilocybin in low alpha (R = 0.32) and delta (R = 0.37), and LSD versus ketamine in delta (R = 0.38). All reported correlations reached statistical significance after FDR correction. Multivariate power classifiers: Random-forest models trained on spectral power features distinguished drugs from placebo with significant accuracy, and several classifiers generalized across compounds. Classifiers trained on psilocybin data could discriminate LSD from placebo across all frequency bands. Classifiers trained on LSD generalized to ketamine for all bands except high beta. Cross-generalisation was particularly consistent in the low and high alpha bands and the low beta band, suggesting these bands carry information shared between drugs. Functional connectivity: LSD significantly reduced source-space functional connectivity across all examined frequency bands, with the greatest extent in low beta, followed by theta and high beta. Psilocybin and ketamine showed sparser connectivity reductions, predominantly in low and high beta. Correlations of connectivity-change maps were more positive than for power: for example, psilocybin versus LSD in the low alpha band showed R = 0.67. Multivariate connectivity classifiers: Classifiers using connectivity features showed less cross-generalisation than power-based models. LSD-trained connectivity classifiers achieved significant AUCs in bands where LSD produced the largest connectivity differences (theta, low/high alpha, low/high beta). Psilocybin-trained classifiers generalized to LSD in theta, low alpha and low beta; the low beta band exhibited the most frequent generalisation across drug pairs. At the ROI level, functional connectivity profiles of occipital regions in the low alpha band distinguished LSD from placebo and generalised between LSD and psilocybin; parietal and frontal nodes associated with the default mode network also contributed to successful generalisation between these two SPs.

Pallavicini and colleagues interpret their findings as indicating both convergent and divergent spectral signatures across serotonergic psychedelics and an NMDA-antagonist dissociative. A common landmark across LSD, psilocybin and ketamine was a decrease in broadband spectral power and reductions in source functional connectivity. However, frequency-specific patterns revealed that classifiers generalised more readily between LSD and psilocybin than between those SPs and ketamine, especially in alpha-band connectivity and occipital circuitry. The authors relate decreases in low-frequency power (particularly alpha and beta) to processes such as cortical disinhibition and increased signal entropy, phenomena previously linked to 5-HT2A receptor agonism but shown here to be also elicitable by NMDA antagonism. They note theta-band changes did not survive multiple-comparison correction, although uncorrected testing suggested posterior theta decreases consistent with previous reports. High beta activity was an exception, diverging between SPs and ketamine and aligning with prior observations that high-beta or gamma-related measures may map differently onto metabolic and hemodynamic signals. Connectivity analyses clarified convergences and divergences: low alpha connectivity decreases in occipital regions were specific to LSD and psilocybin and did not generalise well to ketamine, which accords with the stronger visual phenomenology reported under SPs and the established role of alpha rhythms in visual gating and cortical inhibition. Conversely, low-beta connectivity reductions generalised across drug classes and are discussed as potentially indexing shared effects on self-related processing, given associations between beta-band synchrony, default mode network dynamics and ego‑related experience. The authors acknowledge several limitations that temper interpretation. Gamma-band activity was excluded because of muscle artefact, which limits comparison with fMRI BOLD findings and prior ketamine gamma reports; the extracted text also implies dose- and pharmacology-related differences (for example, LSD produced the largest effects), so some divergences may be quantitative rather than categorical. Methodologically, univariate linear models captured only part of the similarity structure, and multivariate classifiers provided complementary sensitivity to distributed patterns. Finally, the authors propose that their machine‑learning generalisation framework could be applied to characterise new psychoactive compounds by mapping their induced brain states onto those of better-known substances, while emphasising the need for further work to address gamma-band recording challenges and dose/administration differences.