Conceptual framework of the study.

(a) Consciousness is a continuum and can be explored with drug-induced coma of various depths (Xenon, Propofol > Ketamine > Wakefulness). We hypothesize a correspondence between the variations in complexity found with PCI and the dynamics of spontaneous activity across the spectrum of consciousness. (b) We sketch various patterns of spatio-temporal activity reflecting changes in perturbational complexity from left to right. In (c), we show the conceptual shapes of corresponding manifolds of brain activity responsible for different sizes of the functional repertoire (number of wells) and associated with consciousness. (d) The brain is modeled as a network of neural masses coupled by an empirical connectome. This whole-brain model serves as a platform to simulate resting state activity (bottom left) and cortical stimulation (top left, example of firing rate time series with applied stimulus). Dynamical properties of the simulations are studied and compared with data features of human empirical recordings of spontaneous activity (bottom right, EEG during wakefulness and under three anesthetics) and stimulation (top right, TMS-EEG protocol performed in the same conditions).

Dynamical regimes and cascades.

Examples of firing rate time series of the model depending on the strength of interaction between nodes (global coupling, G) and noise intensity (σ). On the left column (a), the interactions are weak (G = 0.27, σ = 0.022 (bottom), σ = 0.056 (top)) and the activity is sparse. On the right column (c), connections are tighter (G=0.65, sigma-bottom=0.022, sigma-top=0.056) and a stable coactivation pattern appears in an ordered fashion. In the middle (b) (bottom), global coupling and noise are at optimal values (G=0.56, sigma=0.036) and allow the emergence of structured patterns (coactivation cascades) of different sizes and durations. Below each time series, the blue line plot shows a quantification of cascades. It corresponds to the absolute value of the mean signal after z-scoring each node’s activity. The AUC is the area under the blue line, it is a quantification of the presence of co-activation cascades.

Exploration of stimulation and spontaneous activity in the parameter space.

Metrics of spontaneous activity and following a perturbation in the parameter space of the model (G, sigma). Change of complexity after a stimulation measured by the sPCI (top left). Fluidity of the dynamics (bottom left) of spontaneous firing rate activity measured by V ariance(dF C) with a sliding window of 3s and 1s step-size. The optimal point or working point where fluidity is maximal (bottom left, red arrow). The number of cascades (top middle) measured by the AUC of the z-scored absolute mean signal. Global Activation Potential (GAP) (top-right) is assessed by the fastest change in the residual sum of squares across sources’ membrane potential. Complexity of spontaneous activity (bottom middle) measured by the Lempel-Ziv complexity of binarized firing rate activity (with a threshold at r = 0.7). And the size of the functional repertoire (bottom right) defined by the number of unique patterns of binarized firing rate activity (with a threshold at r = 0.7).

Resting-state metrics on EEG during anesthesia and wakefulness.

Subject-specific distributions of (a) fluidity, (b) the size of the functional repertoire, (c) Lempel-Ziv complexity, and (d) GAP, between wakefulness (blue) and anesthesia (red). From left to right in each panel: Ketamine group, Propofol group, and Xenon group. Values and distributions (kernel density estimations) were obtained by randomly sampling with replacement a minute of signal within each subject’s recording (50 samples drawn per subject per condition, 1 point per sample). Fluidity was calculated on the full recordings for each subject.

Predictive power of resting-state metrics and PCI.

(a) Cross-plots between the PCI obtained experimentally during a TMS-EEG protocol and each metric on spontaneous recordings (functional repertoire, complexity, fluidity, and GAP). Complexity and the size of the functional repertoire were normalized by the length of the recording in minutes. (b) Classification accuracy of a Support Vector Machine classifier with a linear kernel to distinguish either between anesthesia and wakefulness (downward orange triangles) or between conscious report and no report (upward blue triangles). Dashed lines represent the benchmark performances achieved by PCI classification (100% for consciousness and 87% for anesthesia).

Dynamical regimes and stimulation.

(a) Time series of firing rate activity showing the effect of stimulation in the regime corresponding to maximal fluidity (G = 0.56). On the left, a coactivation cascade is disrupted by the stimulus (sPCI<1) and on the right a cascade is created (sPCI>1). (b) In the regime of weak coupling (G = 0.27) the stimulus has little to no effect (sPCI=1). (c) In the regime of strong coupling (G = 0.65), the effect of stimulus is also almost inexistant (sPCI=1). (d) The distribution of the minimum sPCI value (across trials) in parameter space reveals a peak around G = 0.55. In the low noise regimes, the low values of sPCI are due to a stimulation artifact inducing strong change in the normalizing factor of the sPCI (see Methods).