Psychedelics and schizophrenia

This review article (2009) offers hypothesis (currently being tested) on how psychedelics work and how research in this field and research on schizophrenia overlaps.

Abstract

“Research on psychedelics such as lysergic acid diethylamide (LSD) and dissociative drugs such as phencyclidine (PCP) and the symptoms, neurochemical abnormalities and treatment of schizophrenia have converged. The effects of hallucinogenic drugs resemble some of the core symptoms of schizophrenia. Some atypical antipsychotic drugs were identified by their high affinity for serotonin 5-HT_2A receptors, which is also the target of LSD-like drugs. Several effects of PCP-like drugs are strongly affected by both 5-HT_2A and metabotropic glutamate 2/3 receptor modulation. A serotonin–glutamate receptor complex in cortical pyramidal neurons has been identified that might be the target both of psychedelics and the atypical and glutamate classes of antipsychotic drugs. Recent results on the receptor, signalling and circuit mechanisms underlying the response to psychedelic and antipsychotic drugs might lead to unification of the serotonin and glutamate neurochemical hypotheses of schizophrenia.”

Authors: Javier González-Maeso & Stuart C. Sealfon

Summary

Abstract

Schizophrenia and states induced by certain psychotomimetic drugs share some physiological and phenomenological properties, but differ in fundamental ways. A computational model based on the predictive processing framework can explain the differences between the two states.

1. Introduction

This study aims to compare neuroimaging data from patients suffering from schizophrenia and healthy subjects under the effects of two psychoactive substances: the classical psychedelic lysergic acid diethylamide (LSD) and the dissociative drug ketamine (KET).

KET is thought to reproduce positive and negative symptoms of schizophrenia in humans, and its mechanism of action is thought to reproduce a key element of the molecular pathophysiology of schizophrenia.

Both psychotomimetic drug states and schizophrenia are associated with marked changes in large-scale neural dynamics. However, the effects of these changes are different, and a parsimonious account of the differences is still lacking.

A promising approach to gain insights into the mechanisms driving the core similarities and differences between psychotomimetic drug states and schizophrenia is to leverage principles from the predictive processing (PP) framework of brain function.

PP has been used to explain perceptual alterations observed in both psychotomimetic drug states and psychiatric illnesses, with a focus on schizophrenia. It has also been used to explain the action of psychedelics, most notably through the REBUS model.

This paper replicates and extends findings on neural diversity and information transfer under two psychotomimetic drugs and in schizophrenia using EEG and MEG recordings, and proposes a model-based perspective on how these conditions alter conscious experience.

100 2.1. Data acquisition and preprocessing

Data from 29 patients with schizophrenia and 38 age-matched healthy control subjects were obtained from the Bipolar-Schizophrenia Network on Intermediate Phenotypes (BSNIP) database . The subjects underwent two scanning sessions: one after drug administration and another after a placebo (PLA).

Preprocessing steps were performed using the Fieldtrip and EEGLAB libraries, and the data was segmented into 125 epochs of 2 seconds. The data was then bandpass-filtered between 1 – 100 Hz, and downsampled with phase correction to 250 Hz (EEG) and 300 Hz (MEG).

We use two complementary metrics to analyze neural activity: Lempel-Ziv complexity and transfer entropy. These metrics capture different aspects of neural dynamics and have been used across a wide range of states of consciousness.

Lempel-Ziv complexity (LZ) is a measure of pattern diversity that has been associated with unconscious states, psychedelics and musical improvisation.

To calculate LZ, first transform a given 160 signal into a binary sequence, then scan the resulting binary sequence using the 165 LZ76 algorithm, and finally divide the number of distinct patterns by log2 (T )/T to yield an estimate of the signal’s entropy rate.

We consider transfer entropy (TE) to assess the dynamical interdependencies between ROIs, and use this to analyse the structure of large-scale brain activity. TE is computed in terms of Shannon’s mutual information, and is equal to the product of the activity of two given ROIs.

State-space models with Gaussian innovations can be used to estimate the TE between pairs of ROIs, and the resulting network can be tested for statistical differences across groups.

2.3. Statistical analysis

For the schizophrenia dataset, group-level differences were estimated via linear models. A quadratic model was selected if it was preferred over a linear model by a log-likelihood ratio test.

Multiple comparisons were addressed by using the Network-Based Statistic (NBS) method, which identifies clusters of differences where a particular null hypothesis is consistently rejected.

220 2.4. Computational modelling

A Bayesian state-space model was developed to interpret the LZ and TE findings observed on the neuroimaging data. It is based on the postulate that neuronal populations across the brain can interpret their afferent signals.

This model represents neural activity within a larger hierarchical processing structure, and minimises prediction error signals throughout the hierarchy by updating top-down predictions.

Within this model, schizophrenia and psychedelic conditions are modeled as different types of disruption to Bayesian inference. This has important consequences for the behaviour of the model.

The dynamics of the system can be described as a recurrent update between predictions and prediction errors. The model is calibrated using data from the primary visual cortex, and then the schizophrenia condition is modelled by setting the model parameters.

LSD, ketamine and schizophrenia all show increased signal diversity, as measured by LZ, across the three datasets. This is due to increased bottom-up prediction errors and increased top-down transfer entropy.

Our results show that 300 LZ increases in all three datasets, and that the effects are widespread throughout the brain, with the effects in schizophrenia patients being more pronounced in frontal and parietal regions. However, controlling for the medication status of each schizophrenia patient was crucial to obtain results that match prior work.

We measured the transfer entropy (TE) between pairs of ROIs in the brains of LSD, KET, and schizophrenia patients, and found that KET and SCZ patients showed a ubiquitous decrease in TE, whereas LSD and KET patients showed a marked increase in TE.

Controlling for antipsychotic use was key to revealing differences between healthy controls and schizophrenia patients. However, unlike LZ, the correlation between antipsychotic use and TE between certain ROI pairs did not survive correction for multiple comparisons.

3.3. Computational model reproduces experimental results

We have seen that subjects under the effects of two different psychotomimetic drugs display increased signal diversity and reduced information flow in their neural 365 dynamics, whereas schizophrenia patients display increased complexity but also increased information flow. We have now shown that complementary perturbations to the precision terms of the predictive processing model reproduce these findings.

We repeated the analysis on the model but explored different variations of the precision terms, but neither change reproduced the experimental findings.

4. Discussion

In this paper we analysed MEG and EEG data from healthy subjects under the effects of LSD and ketamine, as well as from 400 schizophrenia patients and healthy control subjects. We found that both drugs increase signal diversity but decrease information transfer, with information transfer being higher in schizophrenia patients.

The symptoms of schizophrenia can be understood as alterations to processes of Bayesian inference. This is supported by a growing number of experimental findings, including an enhanced confirmation bias, impaired reversal learning, and a greater resistance to visual illusions.

Most studies on schizophrenia are task-based and focus on differentiating perceptual learning behaviours between healthy controls and schizophrenia patients. However, resting-state studies are lacking and provide insights into resting-state neural activity under schizophrenia.

4.2. Beyond unidimensional accounts of top-down vs bottom-up processing

We found that spontaneous brain activity increases in the psychotomimetic drug condition, while it decreases in schizophrenia, and that this increase is explained by a bias favouring bottom-up over top-down processing.

The increased transfer entropy from frontal to posterior brain areas observed under schizophrenia may be explained by aberrant Bayesian inference in which bottom-up influences become stronger, depending on which precision terms are involved.

The bottom-up vs top-down dichotomy between schizophrenia and psychosis is too simplistic, and multi-dimensional approaches may shed more light on this issue.

4.3. Limitations and future work

Our results agree with the canonical PP account of psychosis, but some reports have suggested a stronger influence of priors over sensory signals. The simple computational modelling developed here cannot account for these results.

Regarding the empirical analyses, it is important to note that they are subject to a few limitations due to the nature of the data used. Additionally, future work should examine how power spectra across the different conditions relate to the findings presented.

While the measures discussed here capture significant differences between schizophrenia patients and healthy controls, more work needs to be done. The number of antipsychotics taken by the patients was used as a proxy to the missing dosage data, but it is difficult to disentangle this effect from potential confounds. The neural underpinnings of positive and negative symptoms may be different, and future studies could explore these differences.

We used the number of antipsychotic medications being used by each patient as a proxy measure for their medication load. Our tentative results suggest that antipsychotic use may bring the patients’ neural dynamics closer to the range of healthy controls.

4.4. Final remarks

In this paper we have compared brain activ-580 ity in individuals with schizophrenia, a classic 5-HT2A receptor agonist psychedelic, LSD, and an NMDA antagonist dissociative, ketamine, with placebo. We have proposed a computational model that recapitulates the empirical findings through distinct alterations to optimal Bayesian inference.

This study combines information-theoretic analyses of experimental data and computational modelling to study the relationship between neural dynamics and high-level brain functions.

This work was supported by the Imperial College President’s PhD Scholarship, the Wellcome Trust, the Ad Astra Chandaria foundation, the Psychedelic Research Group, Imperial College London, the Ralph Metzner Chair of the Psychedelic Division, Neuroscape at University of California San Francisco, and the European Research Council.

The LSD study was approved by the National Research Ethics Service committee London-West London, and the KET study was approved by a UK National Health Service research ethics commit-640 tee.

Data and code availability statement

Raw MEG data from schizophrenia patients and healthy controls is available in the Har-645 vard Dataverse repository, and open-source implementations are available online.

650 Appendix A. Further details on the predictive processing model

This appendix outlines how to interpret a Bayesian inference process in terms of the joint dynamics of pre-655 diction and prediction error.

A brain region can use all its previous signals to generate an optimal prior estimation of the hidden cause of a new signal.

When all the distributions in the right-hand side of the equation above are Gaussian, the posterior is easily calculable using Bayes’ rule for Gaussian variables.

The Kalman gain parameter, which depends only on its previous value h t1 and the prediction error t, is used to update the prediction in Eq. (A.1) to the next step.

From the definition of t and Ft, it can be seen that increasing s leads to a higher t , thus increasing the bottom-up influence of prediction errors.

We explored additional variations of the computational model reported above, and found that increasing state precision or decreasing sensory precision (both cases of increased top-680 down influence) lead to decreased LZ in the prediction error signals.

The proxy measure for strength of antipsychotic medication used by the patients works well to differentiate the neural activity between brain regions.

We present a preliminary analysis of the relationship between the number of antipsychotics and the positive PANSS scores. The results suggest that the medications affect the self reported symptoms.