Network control energy reductions under DMT relate to serotonin receptors, signal diversity, and subjective experience

Abstract

Psychedelics offer a profound window into the human brain through their robust effects on perception, subjective experience, and brain activity patterns. The serotonergic psychedelic N,N-dimethyltryptamine (DMT) induces a profoundly immersive altered state of consciousness lasting under 20 min, allowing the entire experience to be captured during a single functional magnetic resonance imaging (fMRI) scan. Using network control theory, we map energy trajectories of 14 individuals undergoing fMRI during DMT and placebo. We find that global control energy is reduced after DMT injection compared to placebo. Longitudinal trajectories of global control energy correlate with longitudinal trajectories of electroencephalography (EEG) signal diversity (a measure of entropy) and subjective drug intensity ratings. At the regional level, spatial patterns of DMT's effects on these metrics correlate with serotonin 2a receptor density from positron emission tomography (PET) data. Using receptor distribution and pharmacokinetic information, we recapitulate DMT's effects on global control energy trajectories, demonstrating control models can predict pharmacological effects on brain dynamics.

Fig. 1. Time-resolved network control analysis of the human brain during a pharmacologically induced alteration of consciousness.

a Fourteen individuals were scanned twice per day on two different days (two weeks apart), receiving either DMT or saline placebo at each of these separate days in a single-blind, counterbalanced design (see “Participants and study design” section for details). On each day, a 28-min-long eyes-closed resting-state EEG-fMRI scan was performed with DMT/placebo intravenously administered at the end of the 8th minute. On the same day, identical scanning sessions were performed where participants were asked to rate the subjective intensity of drug effects at the end of every minute. b Here, we deploy a time-resolved network control analysis of the brain’s trajectory through its activational landscape. The position in the landscape is illustrated here as a 3D vector containing regional BOLD signal amplitude at a given time t. We compute a control energy time-series from the regional activity vector time-series by modeling transitions between adjacent regional activity vectors (x₀ and x_f, respectively) using a linear time-invariant model within a network control theory framework. In this framework, the state of the network x(t), here a vector of regional BOLD activations at time t, evolves over time via diffusion through the brain’s weighted structural connectome A, the adjacency matrix. In order to complete the desired transition from the initial (x₀) to the target state (x_f), input (u) is injected into each region in the network. Varying control strategies (reflected in the matrix B) may be deployed, wherein different regions are assigned varied amounts of control within the system. Integrating input u(t) at each node over the length of the trajectory from x₀ to x_f yields region-wise control energy, and summing over all regions yields a global value of control energy required to complete the transition.

Fig. 2. Global control energy is reduced after DMT injection compared to after placebo injection, and negatively correlates with signal diversity and subjective drug intensity ratings.

a Group-average global control energy time-series for the DMT and placebo (PCB) conditions. Nearly two-thirds of post-injection control energies (61.7%) were found to be significantly lower under DMT compared to placebo (n = 14 subjects; gray boxes reflect cluster-corrected significant time-points; see “Global control energy analyses” section for details). b Differences in global control energy and EEG signal diversity between the DMT and PCB conditions are negatively correlated over the 28 min scans (n = 838 time-points; Spearman’s R = −0.35, p_perm < 0.0001), indicating that lower demand for fMRI-based global control energy was associated with increased EEG-based signal diversity of brain activity. c Differences in global control energy between the DMT and PCB conditions were averaged over one-minute intervals in order to compare with subjective drug intensity ratings (0–10) collected at the end of each minute (the latter of which were obtained during a separate fMRI from the one used to calculate global control energy). We found a negative correlation over time between intensity ratings and differences between the DMT and placebo conditions’ global control energies (n = 28 time-points; Spearman’s R = −0.42, p_perm 0.0166). Solid lines are group means and corresponding shaded boundaries reflect the standard error of the mean (SEM).

Fig. 3. Regional control energy and its temporal correlation with signal diversity and drug intensity are associated with serotonin 2a receptor maps.

a Regional control energy metrics. (left) The change in regional control energy in the 8 min after DMT injection, relative to the 8 min prior to the injection. (middle) Each region’s control energy time-series over the course of the full 28 min DMT scans correlated with global signal diversity from EEG during the same scans. (right) Regional control energy during the DMT scans was averaged over one-minute windows corresponding to the timing of subjective drug intensity ratings from separate scans. The windowed control energy time-series for each region was then correlated with the subjective drug intensity ratings. b Each of the regional metrics in (a) were then correlated with the cortical spatial map of the serotonin 2a receptor derived from PET. The strength of these correlations were compared against null correlations with 10,000 cortical spin permutations of the 2a receptor map. c Scatter plots of the three cortical regions’ metrics (n = 100 regions) from (a) and serotonin 2a receptor density from (b).

Fig. 4. Dominance analysis reveals the highest relative importance of the serotonin 2a receptor in DMT-related changes in cortical activity metrics.

Three separate dominance analyses were performed using cortical values from five PET-derived serotonin receptor and transporter spatial densities as input variables and each cortical metric from Fig. 3a as the output. Dominance analysis assesses the relative importance of each input in explaining the output variable’s variance while controlling for the contributions of other predictors in multiple regressions. Displayed is the percent relative importance given to each receptor/transporter map for explaining the variance in each cortical metric, as determined by dominance analysis. n = 116 regions; 5-HT = serotonin (5-hydroxytryptamine).

Fig. 5. Global control energy time-series for the DMT condition is simulated using only placebo fMRI data, coupled with simulated DMT plasma concentrations and 2a receptor density information.

Pharmacokinetic modeling yields an estimate of DMT concentration over the length of the 28 min scans. Here, we specifically used predicted ‘brain-effect compartment’ concentrations from a previously validated model using plasma concentration sampling and EEG. Multiplying DMT concentration over time by regional PET-derived serotonin 2a densities yields an estimate for DMT’s impact on each brain region over time, which can be used as a time-varying control strategy. In order to simulate the impact of DMT on the global control energy time-series, we use each participant’s placebo fMRI data and apply the time-varying control strategy via inclusion in the diagonal of matrix B. Prior to DMT injection, the control strategy (diagonal in B) is uniform, as is the case for all previously calculated energy metrics. Solid lines are group means, and corresponding shaded boundaries reflect the standard error of the mean (SEM). CE = control energy; a.u. = arbitrary units.