Spectral signatures of serotonergic psychedelics and glutamatergic dissociatives

This study (n=48) found that there is significant overlap in the neural correlates of classic serotonergic psychedelics (psilocybin, LSD) and ketamine, even though the mechanism of action is not the same.

Abstract

“Classic serotonergic psychedelics are remarkable for their capacity to induce reversible alterations in consciousness of the self and the surroundings, mediated by agonism at serotonin 5-HT_2A receptors. The subjective effects elicited by dissociative drugs acting as N-methyl-D-aspartate (NMDA) antagonists (e.g. ketamine and phencyclidine) overlap in certain domains with those of serotonergic psychedelics, suggesting some potential similarities in the brain activity patterns induced by both classes of drugs, despite different pharmacological mechanisms of action. We investigated source-localized magnetoencephalography recordings to determine the frequency-specific changes in oscillatory activity and long-range functional coupling that are common to two serotonergic compounds (lysergic acid diethylamide [LSD] and psilocybin) and the NMDA-antagonist ketamine. Administration of the three drugs resulted in widespread and broadband spectral power reductions. We established their similarity by using different pairs of compounds to train and subsequently evaluate multivariate machine learning classifiers. After applying the same methodology to functional connectivity values, we observed a pattern of occipital, parietal and frontal decreases in the low alpha and theta bands that were specific to LSD and psilocybin, as well as decreases in the low beta band common to the three drugs. Our results represent a first effort in the direction of quantifying the similarity of large-scale brain activity patterns induced by drugs of different mechanism of action, confirming the link between changes in theta and alpha oscillations and 5-HT_2A agonism, while also revealing the decoupling of activity in the beta band as an effect shared between NMDA antagonists and 5-HT_2A agonists. We discuss how these frequency-specific convergences and divergences in the power and functional connectivity of brain oscillations might relate to the overlapping subjective effects of serotonergic psychedelics and glutamatergic dissociative compounds.”

Authors: Carla Pallavicini, Martina G. Vilas, Mirta Villarreal, Federico Zamberlan, Suresh Muthukumaraswamy, David Nutt, Robin L. Carhart-Harris & Enzo Tagliazucchi

Summary

Classic serotonergic psychedelics induce reversible alterations in consciousness of the self and the surroundings, mediated by agonism at serotonin 5-HT2A receptors. N-methyl-D-aspartate (NMDA) antagonists induce subjective effects that overlap in certain domains with those of serotonergic psychedelics, suggesting similarities in the brain activity patterns induced by both classes of drugs.

1. Introduction

Several categories of compounds have been used to elicit altered states of consciousness, including serotonergic psychedelics (SP) and N-methyl-D-aspartate (NMDA) receptor antagonist glutamatergic dissociatives (GD2). These compounds have been investigated in humans at the molecular and systems level, as well as in terms of behavioral changes and subjective effects.

Both SP and GD elicit their pharmacological action by non-selectively binding to receptors associated with different endogenous neurotransmitters. SP and GD act as partial agonists of certain serotonin (5-HT]) receptor subtypes, while GD acts as non-competitive antagonists at NMDA receptors. Ketamine and PCP are likely agonists at 5-HT2A and dopamine D2 receptors, and they enhance 5-HT2A receptor-mediated vasoconstriction.

SP and GD have different primary molecular sites of action, but their effects are similar in that they decrease the functional integrity of the default mode network and increase its functional connectivity with other brain systems. LSD and psilocybin reduce broadband oscillatory power, while ketamine increases gamma and theta power in anterior regions, and decreases theta power in posterior areas. Both drugs increase global connectivity and entropy.

We used magnet-encephalography to measure changes in brain oscillations induced by the administration of different psychoactive drugs. We then trained multivariate machine learning models to distinguish each drug from the corresponding placebo using frequency-specific features, and evaluated the accuracy of generalizing each classifier to distinguish other drugs from the placebo conditions.

2. Materials and methods

2.1. Participants and experimental design

MEG data were collected from 15 participants for LSD, 6 participants for ketamine and 14 participants for psilocybin. Participants were excluded if they were younger than 21 years old, pregnant, had a cardiovascular disease, suffered from claustrophobia, blood or needle phobia, or had any medical condition rendering them unsuitable for the study.

MEG recordings were performed 4 h after LSD and 2 min after psilocybin and ketamine infusions, respectively. All experiments were approved by a UK National Health Service research ethics committee.

2.2. Data acquisition and pre-processing

For psilocybin and ketamine, participants lay in a supine position, and pulse rates and blood oxygenation levels were monitored throughout all acquisitions. Whole-head MEG recordings were made using a CTF 275- channel radial gradiometer system sampled at 1200 Hz (0 – 300 Hz band-pass).

Source modelling was performed using the Fieldtrip toolbox on MRI data for each participant. Broadband virtual sensor time-series were constructed using a linearly constrained minimum variance beamformer.

2.3. Time-frequency analysis

Time-frequency analysis was performed using Hanning windowed Fast Fourier transforms and six canonical spectral bands were split into six groups: delta, theta, low alpha, high alpha, low beta and high beta. Gamma activity was excluded from the analysis.

2.4. Functional connectivity analysis in source space

The functional connectivity between regions was determined by computing the linear correlation between the envelopes of the orthogonalized time series, which were obtained by subtracting the optimal linear prediction from the original time series.

2.5. Linear correlation between statistical parametric maps of drugs vs. placebo

Mass univariate tests were conducted to compare the spectral power and functional connectivity of MEG signals in different frequency bands, and similarities in the spatial patterns of changes were assessed by computing the linear correlation coefficient.

2.6. Generalization of multivariate machine learning models between drugs

We trained machine learning models to distinguish LSD, ketamine and psilocybin from the corresponding placebos, and evaluated the accuracy of the models using a five-fold cross-validation procedure. We then applied the same model to distinguish another drug from the placebo.

Multivariate classifiers were based on random forests, which were implemented in scikit-learn. The random forest algorithm creates an ensemble of decision trees based on a randomly chosen subset of the features, and the probability of a new sample belonging to each class is determined by the aggregated “vote” of all decision trees. We trained a random forest with 1000 decision trees and expanded the trees until all leaves were pure (i.e. no maximum depth was enforced). The hyperparameters are detailed in the documentation.

We trained a random forest classifier to recognize drugs from placebos, and then evaluated the performance using a five fold cross-validation procedure. We then constructed an empirical p-value by counting the amount of times the accuracy of the scrambled classifier was greater than the original classifier.

We built 1620 classifiers based on the functional connectivity values of individual regions of interest (ROIs) and repeated the procedure for each drug and frequency band. These classifiers produced significant accuracy values (p 0.05, FDR corrected).

3. Results

We observed significant differences in spectral power between drugs and placebo in all frequency bands, except high beta. LSD was the only drug to produce significant decreases in all examined frequency bands and also presented the largest effect sizes.

We tested the similarity of the spatial distributions of spectral power decreases between drugs using a linear model. The correlation values were generally low, implying that the changes in MEG spectral power induced by a compound vs. the placebo were not spatially similar.

We trained multivariate machine learning classifiers to distinguish LSD, psilocybin and ketamine from the placebo based on the spectral power decreases at all source space ROIs. The classifiers could generalize to distinguish other drugs from the placebo with significant accuracy.

We computed functional connectivity between 90 source space ROIs for all frequency bands and conditions, and found that LSD significantly reduced connectivity values for all frequency bands, with the most widespread changes in low beta, followed by theta and high beta.

We repeated the multivariate analysis using data from different compounds and observed less training/evaluation pairs with significant AUC values. LSD presented significant connectivity differences for all spectral bands and theta, low and high alpha, and low and high beta bands.

We used functional connectivity values between individual ROIs and all remaining ROIs to train and evaluate random forest classifiers that could distinguish between LSD and psilocybin as well as between LSD and the placebo condition.

4. Discussion

Consistent with previous findings, both SP and GD decreased the power of broadband oscillations as measured with MEG. These decreases might reflect more general processes that can be caused by the action of NMDA antagonists as well.

Power spectrum changes were similar after the administration of ketamine and LSD, except for high beta, which diverged between the drugs. This is consistent with previous reports of different spatial patterns of glucose consumption during the acute effects of SP and GD.

Three drugs here investigated resulted in decreased functional connectivity in the default mode network (DMN), consistent with previous reports of diminished DMN connectivity elicited by psychedelics associated with the experience of ego dissolution.

Low to moderate doses of ketamine and SP elicit different subjective effects, mainly in the visual domain. The functional connectivity profile of individual occipital ROIs allowed to distinguish LSD from placebo, as well as the generalization towards the classification of psilocybin vs. placebo, and vice versa.

LSD produced significant decreases in functional connectivity in all frequency bands, as revealed by mass univariate statistical tests. This could be related to the association between the BOLD signal and gamma band activity, which was excluded from the analyses due to muscular artifacts.

We excluded activity in the gamma band from our analyses due to the possibility of muscular artifacts, but future studies should address the challenge of recording sufficiently clean gamma activity under the effects of SP.

We introduced a framework to assess whether the brain activity elicited by psychoactive drugs can be generalized to other drugs. This method could be valuable for the study of new substances.