LSD alters dynamic integration and segregation in the human brain

This fMRI study (2020) improves our understanding of how LSD changes brain function over time and how subjective effects (e.g. ego dissolution) map onto these changes.

Abstract

“Investigating changes in brain function induced by mind-altering substances such as LSD is a powerful method for interrogating and understanding how mind interfaces with brain, by connecting novel psychological phenomena with their neurobiological correlates. LSD is known to increase measures of brain complexity, potentially reflecting a neurobiological correlate of the especially rich phenomenological content of psychedelic-induced experiences. Yet although the subjective stream of consciousness is a constant ebb and flow, no studies to date have investigated how LSD influences the dynamics of functional connectivity in the human brain. Focusing on the two fundamental network properties of integration and segregation, here we combined graph theory and dynamic functional connectivity from resting-state functional MRI to examine time-resolved effects of LSD on brain networks properties and subjective experiences. Our main finding is that the effects of LSD on brain function and subjective experience are non-uniform in time: LSD makes globally segregated sub-states of dynamic functional connectivity more complex, and weakens the relationship between functional and anatomical connectivity. On a regional level, LSD reduces functional connectivity of the anterior medial prefrontal cortex, specifically during states of high segregation. Time-specific effects were correlated with different aspects of subjective experiences; in particular, ego dissolution was predicted by increased small-world organisation during a state of high global integration. These results reveal a more nuanced, temporally-specific picture of altered brain connectivity and complexity under psychedelics than has previously been reported.“

Authors: Andrea I. Luppi, Robin L. Carhart-Harris, Leor Roseman, Ioannis Pappas, David K. Menon & Emanuel A. Stamatakisa

Notes

This paper is a further analysis of data from the same participants (n=20, 75μg LSD) as were studied in Carhart-Harris and colleagues (2016).

Summary

1. Introduction

LSD, psilocybin and DMT are classic serotonergic psychedelics that induce a profoundly altered state of consciousness. Using functional MRI to study these drugs can provide insight into normal and abnormal brain function.

Evidence for the therapeutic potential of carefully administered psychedelics for improving mental health outcomes has been accumulating, and increased brain complexity may offer a proficuous way of indexing the psychological effects of mind-altering substances via their neurobiological effects.

Recent studies have shown that brain dynamics may represent the “common currency” between the mind and the underlying neurobiology, and that patterns of brain connectivity vary over time, switching between different dynamic sub-states with different relevance for cognition.

Brain integration and segregation are important properties of the mind and brain, relevant to our understanding of conscious experience. LSD and other psychedelics decrease the integrity of functional connectivity within resting-state networks, but increase FC between normally distinct RSNs.

These observations raise the intriguing possibility that LSD has a finer-grained temporal dimension to its effects on the human brain.

Graph theory provides a formal way to quantify integration, segregation and their dynamics in the brain. A crucial property of many natural and artificial networks is the so-called ‘smallworld’ organisation, which may represent a mark of optimal balance between global and local processing.

We combined graph theory with dynamic functional MRI connectivity to explore the time-resolved effects of LSD on brain function and the ensuing subjective alterations of consciousness in 15 healthy volunteers with previous psychedelic experience.

2.1.2. Recruitment

Twenty healthy volunteers with a previous experience using psychedelic drugs underwent two scans, 14 days apart. They were given a placebo or an active dose of LSD, and had a brief acclimation period in a mock MRI scanner before the resting-state scans were initiated.

The two BOLD scans included here were performed on a 3T GE HDx system. They were acquired with 3D fast spoiled gradient echo scans in an axial orientation.

BOLD-weighted fMRI data were acquired using a gradient echo planer imaging sequence, and 15 subjects were analyzed. One subject aborted the experiment due to anxiety, and four others were excluded for excessive head motion in the scanner.

2.2. Preprocessing

We followed the same preprocessing pipeline described in our previous publications ( Luppi et al., 2020 , 2019 ): removal of the first three volumes, functional realignment, slice-timing correction, identification of outlier scans for subsequent scrubbing, normalisation to Montreal Neurological Institute (MNI-152) standard space, spatial smoothing.

2.3. Denoising

We used the CONN toolbox to perform denoising and preprocessing, and followed the same denoising described in our previous publications. We avoided using global signal regression (GSR) in favour of the anatomical CompCor method, and used the cartographic profile method to identify integrated and segregated sub-states of dynamic functional connectivity.

The aCompCor method involves removing confounding effects such as white matter signal, cerebrospinal fluid signal, six subject-specific realignment parameters, and outlying scans using Ordinary Least Squares regression. Finally, the BOLD signal timeseries are band-pass filtered to eliminate both low-frequency drift effects and high-frequency noise.

2.4.1. Definition on regions of interest

To construct matrices of functional connectivity, 200 cortical regions were parcellated and 32 subcortical regions were augmented with the scale-200 version of the recent multi-scale local-global functional parcellation.

We replicated our analyses using two alternative ways of parcellating the brain, a finer-grained version of the multi-scale Schaefer functional cortical atlas and the Brainnetome atlas. The results pertain to the Schaefer-232 atlas.

2.4.2. Connectivity matrix construction

Following our previous work, functional connectivity was estimated as the Pearson correlation coefficient between the time-courses of each pair of ROIs, over the full scanning length.

2.4.3. Dynamic functional connectivity

Dynamic connectivity matrices were derived using an overlapping sliding-window approach, and consisted of one functional connectivity matrix for each timepoint. The tensors were obtained by convolving a rectangle of 22 TRs with a Gaussian kernel of 3 TRs, sliding with 1 TR step size.

2.5. Derivation of Integrated and Segregated sub-states

Shine et al. (2016) established a cartographic profile based on the graph-theoretical module assignments of each ROI, and subsequent work using the same methodology identified sub-states of higher integration or segregation over time.

Graph theory was applied to the dynamic FC matrices of each participant to identify modules, and a k-means clustering algorithm was used to assign each matrix to one of two clusters.

It is still possible to quantify the level of segregation in the primarily integrated sub-state, and the integration of the primarily segregated sub-state, even after dividing days in “hot” and “cold”.

For each subject, the proportion of time spent in each sub-state was quantified, and the entropy of the pattern of predominantly integrated and segregated sub-states was measured using a naive estimator.

2.6. Validation of dynamics against stationary null model

We used a Vector Autoregressive (VAR) model to generate surrogate timeseries for the resting brain, and applied sliding-windows dynamic functional connectivity followed by the cartographic profile to compare the proportion of time spent in the predominantly integrated sub-state between the empirical data and the stationary surrogates.

2.7. Characterisation of dynamic functional brain networks

Several properties of dynamic functional brain networks are sensitive to pharmacological perturbations of consciousness, but it is not known whether primarily integrated or segregated sub-states are especially susceptible to the effects of LSD.

2.8.1. Prevalence of Anticorrelations

Anaesthesia and disorders of consciousness suppress anticorrelations between brain regions, but this phenomenon is not uniform in time, and is primarily observed during predominantly integrated dynamic sub-states. The prevalence of anticorrelations in the psychedelic state induced by LSD was measured.

2.8.2. Structural-functional similarity

Previous results have shown that suppressed consciousness induced by anaesthetics corresponds to increased similarity between certain sub-states of dynamic functional connectivity and the underlying pattern of anatomical connections. In order to evaluate the similarity between functional and structural connectivity, we computed the Spearman correlation coefficient.

We computed the Hamming distance between the binarized connectivity patterns of each ROI in the functional and anatomical connectivity matrices as an alternative method to estimate the similarity between structural and functional patterns of connectivity.

A small-world network is a lattice network with short characteristic path lengths, and is theorised to support specialised processing. Its small-worldness is decreased during anaesthesia and in patients with disorders of consciousness.

We adopted the measure of small-world propensity recently developed by Muldoon et al. (2016), which quantifies the extent that different networks exhibit small-world structure, while accounting for network density.

We bound both measures of fractional deviation between 0 and 1, and ensured that the resulting values of small-world propensity were also bounded between 0 and 1.

Small-world propensity is a measure of how closely a network resembles a small-world network. A high small-world propensity indicates better adherence to the requirements of a small-world network.

Unlike the small-world index, small-world propensity is not intended to determine in absolute terms whether a network exhibits small-world structure.

2.8.4. Functional network construction

We constructed brain networks by thresholding functional connectivity matrices, setting all above-threshold edge weights to unity, and assigning unit weight to each edge. This ensured that any observed differences would be due solely to the networks’ topology.

To ensure that the results would not be dependent on the threshold level used, we thresholded each FC matrix at density levels ranging between 10% and 25%, sampled in steps of 5%. We also repeated our analysis including all non-negative edges to account for biological meaning.

2.8.5. Network functional complexity

We also applied the recently introduced measure of “Functional Complexity” to quantify the functional diversity of functional brain networks. This measure is the difference between the observed distribution p(r ij ) and the uniform distribution, and is calculated as the sum of the differences over the bins.

2.9. Statistical analysis

Since head motion was significantly higher in the LSD condition than in the placebo condition, we included mean framewise displacement as a covariate of no interest in our statistical analyses.

2.9.1. Network based statistic

The network-based statistic (NBS) approach was used to investigate the statistical significance of LSD-induced alterations on the functional brain networks, for time-averaged functional connectivity and for the predominantly integrated and segregated substates. This nonparametric statistical method achieves superior power compared to mass-univariate approaches.

2.9.2. Correlations with subjective experiences

Correlations between brain measures and subjective experiences were obtained using Spearman’s non-parametric rank-based correlation coefficient , implemented in the R package ggstatsplot . We also investigated the possible effects of motion on our results.

3.1. Preserved temporal balance of integration and segregation under LSD

LSD alters two fundamental properties of the brain: integration and segregation. We employed an a priori clustering of dynamic functional connectivity into two sub-states, for both LSD and placebo, and found that the proportion of time spent in the globally integrated sub-state during the placebo condition was consistent with the data.

The proportion of time spent in each brain sub-state was preserved under the effects of LSD.

The proportion of time spent in the predominantly integrated sub-state was significantly greater for empirical data than for stationary data generated by a VAR model for each participant in the placebo and LSD conditions, indicating the presence of dynamics in the functional connectivity.

LSD increases the entropy of certain dimensions of brain function, but the temporal alternation of the two sub-states did not exhibit significant changes in its temporal entropy under the effects of LSD compared with placebo.

We investigated whether the effects of LSD on brain function are manifested during dynamic sub-states of high or low integration or segregation, and whether these effects are different from the effects of LSD on time-averaged functional connectivity.

3.2. Time-specific effects of LSD on brain connectivity

LSD-induced network alterations were evident in both time-averaged functional connectivity and the predominantly segregated substate of dynamic functional connectivity. The predominantly segregated sub-state displayed both increases and decreases in functional connectivity, including most of the disconnections from prefrontal cortex.

During the predominantly segregated sub-state, functional connectivity of somatomotor and auditory cortices increased, while connectivity of key nodes of the default mode network decreased. The posterior cingulate displayed increased connectivity with the executive control network, and the thalamus displayed increased connectivity with both cortical and subcortical structures.

3.3. Structural-functional untethering and reduced anticorrelations in the segregated sub-state

Changes in the strength of specific network edges can be used to understand how alterations in one’s state of consciousness arise from perturbations of the brain’s network organisation.

Our results indicated that there was no significant change in the similarity between time-averaged functional connectivity and the underlying structural connectome, but that there were moments in time (characterised by especially high network segregation) during which LSD induces the opposite effect to anaesthesia, allowing FC to diverge more freely from anatomical constraints.

Anaesthesia and disorders of consciousness induce reduced anticorrelations between brain regions. This phenomenon is mainly observed during the predominantly integrated dynamic sub-state, and differs across dynamic sub-states of functional connectivity.

LSD reduces the proportion of negative edges across the brain, but this effect is sub-state specific: no effect was observed during the primarily integrated substate, whereas a significant reduction was observed during the primarily segregated substate.

3.4. Increased functional network small-worldness under LSD

LSD has the opposite effect on human brain networks than expected, which suggests that small-world organisation is compromised in a dynamic fashion during loss of consciousness.

LSD induces an increase in the small-world propensity of time-averaged functional brain networks, but only when considering binary networks comprising only the strongest connections. However, when taking into account the weight of connections between brain regions, an increase in small-world propensity could be detected during both the predominantly segregated and predominantly integrated substates.

3.5. Increased functional complexity in the segregated sub-state

Recent research indicates that aspects of spatial and temporal complexity of the brain are enhanced during the psychedelic state induced by LSD, and decreased when consciousness is lost due to anaesthesia or severe brain injury. Therefore, brain complexity may provide a useful avenue for grounding the mind-altering effects of LSD.

Our results indicate increased functional complexity during the psychedelic state induced by LSD, but only during the primarily segregated sub-state.

3.6. Robustness to parcellation type and size

To ensure robustness, we replicated our results using two alternative ways of defining brain network nodes: the Schaefer functional cortical atlas (400 cortical ROIs instead of 200) and the Brainnetome atlas (210 cortical and 36 subcortical ROIs), obtained from anatomical and functional connectivity.

3.7. Correlations with subjective experiences

We found several significant correlations between LSD-induced changes in brain measures and subjective ratings, but none were reproducible across all three parcellations. However, dynamic analysis revealed consistent significant correlations between subjective ratings and our brain measures of interest.

Correlations were observed between the change in weighted small-world propensity, ASC subjective ratings of blissful state, ASC complex imagery and the VAS score quantifying the feeling of ego dissolution in the predominantly integrated sub-state.

We note that none of our brain measures significantly correlated with mean framewise displacement, when using the Schaefer parcellations.

4. Discussion

We combined fMRI dynamic functional connectivity and graph theory to investigate how the classic serotonergic psychedelic LSD modifies brain function and its dynamics under task-free, eyes-closed resting-state conditions.

LSD’s consciousness-altering effects show prominent temporal aspects, and do not influence the overall allocation of time between predominantly integrated and segregated sub-states.

LSD suppresses anticorrelations between brain regions during the predominantly segregated sub-state, whereas reduced anticorrelations are also reliably observed during loss of consciousness induced by anaesthesia or brain injury. This reconciles apparently opposite observations.

The state of altered consciousness induced by LSD corresponds to an additional, abnormal increase in functional complexity of the brain. This increase is only observed during moments when the brain is characterised by a predominantly segregated pattern of functional connectivity.

LSD induces additional decoupling of functional connectivity from the underlying structural connectome during the predominantly segregated sub-state, which is already more decoupled from structural connectivity than the integrated sub-state. This evidence is also consistent with previous results obtained using an alternative method to study brain dynamics, known as connectome harmonic decompositions. These results reveal a psychedelic-induced decoupling of function from structure.

The reduced similarity between structural and functional connectivity observed during the psychedelic state induced by LSD indicates that the organism’s expectation about which brain regions should be exchanging information with each other is less constrained than usual by the presence or absence of an underlying anatomical connection.

The effects of LSD allow the brain to explore functional connectivity patterns that go beyond those dictated by anatomy, resulting in unusual beliefs and experiences.

The segregated sub-state of LSD involves a large-scale reorganisation of functional connectivity between posterior and frontal regions, especially the left anterior medial prefrontal cortex, which is implicated in the cognitive process known as reality monitoring. In patients with schizophrenia, the anterior mPFC is both hypoactive and disconnected from posterior regions during reality monitoring tasks. This may be a result of attenuated engagement of top-down reality monitoring processes during the predominantly segregated sub-state, as postulated by the REBUS/Anarchic Brain hypothesis.

The entropic brain hypothesis, predictive processing framework, psychoanalytic and filtration accounts all suggest that psychedelics perturb adaptive mechanisms that normally constrain perception, emotion, cognition, and self-reference.

LSD induces increased small-world propensity in both integrated and segregated dynamic sub-states, highlighting the important role of weak edges in the small-world topology of brain networks.

The brain appears to shift the balance of information processing towards a more localised pattern under the effects of LSD, by combining high local clustering (facilitating sharing of information at a local level) and a short characteristic path length.

The authors found that the predominantly integrated state exhibited increased clustering coefficient but decreased characteristic path length after ingestion of ayahuasca, and that small-world propensity also exhibited time-specific effects. This finding merits special attention because ego dissolution experienced during the psychedelic experience predicts positive clinical outcomes.

Our results suggest that the predominantly integrated sub-state may be especially relevant for supporting the feeling of integrity of one’s sense of self, when considering the intrinsically integrative nature of the self.

4.1. Limitations

The present study has a number of limitations, and it will be important to replicate these results using other datasets and larger samples to ensure that our understanding of the effects of LSD on brain function and dynamics is generalisable.

Although the study design was placebo-controlled, it is plausible that volunteers became aware of whether they took LSD or placebo, but we consider it unlikely that this may represent a confound for any of our findings.

Although the data was collected using an inert placebo, future studies may benefit from the use of an ‘active placebo’ condition. This would allow for a more stringent test of the entropic/anarchic brain hypothesis and also provide a way to control for the effects of arousal on dynamic functional connectivity.

Although head motion may represent a potential confound, global signal regression (GSR) has been shown to obscure important neurobiological information, and reduce the test-retest reliability of graph-theoretical metrics and intersession reliability of dynamic brain state properties. We used the aCompCor denoising method, excluded participants with unacceptably high levels of motion, and included mean framewise displacement as a covariate of no interest in our statistical analyses to mitigate the potential confound of residual differences in head motion between conditions.

The relation between functional and structural connectivity was estimated from population-average diffusion imaging templates of the Human Connectome Project, and the effects of LSD on this relation are therefore to be interpreted as deviations from the population average, rather than from the specific individual’s anatomical connectome.

The proportion of time spent in the primarily integrated sub-state is preserved under LSD, despite recent evidence that a dynamic sub-state of high global synchrony is visited more frequently under the effects of psilocybin.

The authors of this paper used sliding-windows FC and LEiDA to study the dynamics of brain connectivity, and found that there are several possible brain sub-states. These sub-states are not obvious to map onto the canonical resting-state networks identified by Lord and colleagues.

Future research should use EEG and MEG to study the dynamics of brain activity, and subjective experience should be sampled more regularly. This will allow a more direct relation between brain states of predominant integration and segregation, and subjective experiences induced by LSD and other psychedelics.

4.2. Conclusion

LSD makes globally segregated brain sub-states more complex and decoupled from structural constraints, and reduces functional connectivity of the anterior medial PFC, which is thought to subserve processes of reality monitoring.

LSD has time-dependent effects on the dynamics of brain function, and may exert its psychedelic effects differently at different points in time.

Data and code availability statement

Functional MRI data from the original study are available from OpenNeuro, and code for the cartographic profile, small-world propensity, and CONN toolbox are freely available online.

Acknowledgements

This work was supported by the Gates Cambridge Trust Scholarship, the Canadian Institute for Advanced Research (CIFAR), the Cambridge Biomedical Research Centre and NIHR Senior Investigator Awards, the Stephen Erskine Fellowship at Queens’ College, Cambridge, the Imperial College President’s Scholarship, and the Alex Mosley Charitable Trust.