Dissociation of local versus global cortical responsiveness across wakefulness, NREM sleep, and anesthesia.

Panels illustrate that although local neuronal and high-frequency responses remain relatively preserved between wake and sleep or under anesthesia, large-scale cortical propagation is selectively disrupted during both NREM sleep and anesthetic-induced unconsciousness. (A) Time-frequency spectrograms of human auditory cortex (80-200 Hz gamma band) in the same stimulus window during wakefulness (left) and NREM sleep (right), showing robust local gamma responses in both states. Modified from Hayat et al., 2022. (B) Trial-averaged gamma-band amplitude (dB) over time (solid lines) and baseline confidence intervals (dashed lines; wake: magenta, sleep: green), demonstrating preserved local gain but attenuated feedback modulation during sleep. Modified from Hayat et al., 2022. (C) Intracellular recordings of a cat neocortical neuron during slow-wave sleep (SWS; left) and quiet wakefulness (WAKE; right). Top traces, surface EEG; bottom traces, membrane potential showing stimulus-triggered down-state (arrow) followed by recovery and firing. Reproduced from Steriade et al., 2001. (D) Summary of local excitability measures in SWS versus wake: number of spikes evoked by a standardized current pulse (left), somatic input resistance (middle), and hyperpolarization area (right). Reproduced from Steriade et al., 2001. (E) Transcranial magnetic stimulation (TMS)-evoked potentials and source-reconstructed cortical maps in four human subjects during wakefulness (left) and NREM sleep (right), showing widespread propagation when awake and collapse to a local response during sleep. Modified from Massimini et al., 2005. (F) TMS-EEG responses in human subjects in awake (blue) vs midazolam sedation (red): top, individual single-trial waveforms superimposed (blue/red) with average (black); bottom, cortical source maps at representative latencies, revealing preserved local activation but loss of global spread. Modified from Ferrarelli et al., 2010.

© 2001, American Physiological Society. Figure 1C and D are reproduced from Steriade et al., 2001 with permission from the American Physiological Society. It is not covered by the CC-BY 4.0 license and further reproduction of this panel would need permission from the copyright holder.

Responsiveness in local circuits across different conditions: in vitro experiments and models.

(A) Neocortical slices activity recorded with 16-channel multielectrode array (MEA). Single pulses of electrical stimulation were applied to the infragranular layers. Raw local field potential (LFP) traces during the regime of slow oscillations (SO) at the top and multiunit activity (MUA) at the bottom. The same is displayed below (blue) for desynchronized activity, awake-like, in the presence of neurotransmitters. On the left, recordings of 5s of spontaneous SO (top), and NE+CCh (bottom). On the right, single traces and averaged (red) LFP (top) and MUA (bottom) responses to electrical stimulation representing spatiotemporal responses across the slice. Modified from Barbero-Castillo et al., 2021. (B) Responsiveness of cortical slices under three different conditions: SO, NE+CCh, and Ka. The responses to electrical stimulation recorded by a single channel are shown both in the LFP and MUA in the first two rows, respectively. The black lines indicate the average response for each condition. Below, the corresponding binary matrix of significant responses to the stimulation [SS(x,t)]. (C) Histogram showing the population values of sPCI averaged across slices (n=14). Modified from D’Andola et al., 2018. (D) Averaged LFP (top) and MUA (bottom) responses to electrical stimulation during spontaneous SO (left), desynchronized activity (middle), and with bath application of200nM of GBZ (right). Binary matrices of significant sources of activity [SS(x,t)] following electrical stimulation delivered to neocortical slices (bottom). (E) Population values of sPCI shown as boxplot for increasing GBZ concentrations (50, 100, 150, and 200nM GBZ *p=<0.05). (F) Model of a cortical slice consisting of pyramidal (blue) and inhibitory (red) neurons arranged in a 50 × 50 square lattice. (G) Responsiveness of the in silico model reproduces what is observed experimentally during (from left to right) SO, SO + blocking GABAA, SO + blocking GABAB, and desynchronized activity. Single spontaneous sLFP (top), averaged sLFP (middle top), MUA (middle bottom), and binary matrices of significant sources of activity [SS(x,t)] (bottom) are shown for each condition. (H) Population sPCI computed on the response generated by the model for inhibited and disinhibited cortical networks. The gray area represents the model predictions. Modified from Barbero-Castillo et al., 2021.

Responsiveness across brain states at the circuit level.

(A) Top, firing rate v(t) from a network of excitatory spiking neurons. Arrows and dashed lines, time of perturbation induced by a sudden change in the synaptic current. Middle, Average adaptation level a(t). Bottom, post-perturbation trajectories in the (a, v) plane showing the state-dependent nature of the responses. (B) Bifurcation diagram showing the different activity regimes displayed by spiking neuron network simulations as a function of the firing rate adaptation and the number of incoming excitatory Poisson processes from other (external) areas. Solid lines show the transitions predicted by the theory while the colors indicate the different dynamical regimes the “finite size” spiking neural network exhibits in simulations. (C) Left, the firing rate of excitatory spiking neurons is shown for three simulations associated with different positions on the phase diagram of panel A. The black arrow indicates the moments in which the network receives a precise excitatory stimulus. Right, the same trajectories are shown on the ‘Firing Rate’ vs. ‘Adaptation Current’ plane. (D) Left and right columns show two different asynchronous irregular (AI) states in a network of 10,000 adaptive exponential (AdEx) integrate-and-fire neurons (8000 RS cells and 2000 FS cells) in two different network states. The response to a Gaussian-distributed excitatory input (top) is indicated in the raster of units (middle). The instantaneous excitatory firing rate is shown in the bottom (noisy curves), together with the AdEx mean-field model (continuous curves). (E) Layered spiking network model covering 4 × 4 mm2 at biologically realistic neuron density. Activity is evoked at t = 700 ms by a spatially and temporally confined thalamic pulse. Evoked activity propagates through the cortical network as displayed in the spike raster plot and sequence of snapshots of spatiotemporally binned firing rates. (A: modified from Linaro et al., 2011; B,C: modified from Cattani et al., 2023; D: modified from di Volo et al., 2019; E: modified from Senk et al., 2024.

Cortical responses occurring as traveling waves.

(A) Spontaneous traveling waves occurring in the awake monkey primary visual cortex, imaged by VSD. Each panel shows a phase latency map for three examples of spontaneous traveling waves. (B) Traveling waves evoked by visual inputs. Three examples of the same visual input and the traveling wave evoked. (C) Model of traveling waves using a 2D array of AdEx mean-field units. The top row shows snapshots of the simulated VSD signal for a single visual input evoking a traveling wave. The bottom row shows the response evoked by more complex inputs in the same conditions. (D) Cortical area modeled as a lattice of cortical assemblies. Changing both adaptation and background excitation, different slow-wave activities are induced which are reminiscent of in vivo observations under deep and light anesthesia (Pazienti et al., 2022). A focal stimulation randomly involving one cortical assembly (red dots) elicits slow activation waves with a probability depending on the past activity of the cortical field. The stimulation likely elicits a spiral wave (top right snapshots) under deep-like anesthesia state, whilst a planar wave (bottom right snapshots) is evoked under the light-like anesthesia state. For both C and D, the population models were composed of excitatory and inhibitory IF neurons incorporating spike-frequency adaptation and being excited by an external source of excitatory neurons. (A,B: modified from Muller et al., 2014; C: modified from Zerlaut et al., 2018; D: modified from Pazienti et al., 2022 and Galluzzi et al., 2025).

Cortical mesoscale responsiveness at different brain state levels in animal and human data and simulations.

(A) State characteristics of mesoscale unit firing patterns in monkey PFC recorded using Utah-arrays. (B) Similar recordings in monkey PPC. In both areas, multi-unit spiking activity was monitored across the three levels of consciousness (quiet wakefulness—red, light anesthesia—green, deep anesthesia—blue). Black dots represent neuronal spikes. Sparsely-firing units during wakefulness were discarded, and this population was fixed for analysis during the anesthetic states. The red, green and blue rectangles denote the path of the Up state (P(UP) vs P(DOWN)), as estimated by a two-state Gaussian hidden Markov model. Progressively, the spiking activity of the network reorganizes into distinct periods of high and low activity, with UP and DOWN states (C) Field activity propagation within the PFC after electrical stimulation. (D) Field propagation in PPC. The red, green and blue curves represent the modulation in signal energy of the broadband (0.1-200Hz) LFPs over distance. The inset bar graphs show the total rate of change in signal energy modulation for proximal (first three spatial bins) and distal (last three spatial bins) populations. In the PFC, signal energy modulation is maintained across proximal and distal populations during wakefulness but decreases as a function of the depth of anesthesia. However, in the PPC, the signal energy modulation decreases with distance in all three states, with similar modulation strengths during wakefulness and deep anesthesia (modified from Dwarakanath et al., 2025). (E,F) Responsiveness to electrical stimulation increases during pre-ictal state and is inhibited during ictal state. (E) Experimental data. The top time trace (David, 2007) shows SEEG responses in the anterior hippocampus during 1Hz stimulation of the amygdala. The amplitude of the response grows until the occurrence of the seizure (around 24 s). The middle panel (Russo et al., 2023) shows response to stimulation in the medial temporal lobe. Stimulation time points are marked with vertical magenta dashed lines. The two bottom plot rows show this response averaged over all trials and overlaid for all recording contacts. (F) Theoretical framework. The bifurcation diagram of the Jansen & Rit model with three insets showing outputs of the model operating in three different dynamical regimes (marked with roman numbers) differing only in the value of the model’s baseline input (abscissa of the main plot). In each case a stimulation pulse with amplitude 10 [s−1] and duration 200 ms was delivered at time t = 0. The black arrows link the value of the baseline input (60 [s− 1], 110 [s−1] and 120 [s−1]) with the corresponding generated time traces. Modified from Jedynak et al., 2017.

Macroscale responsiveness at different brain state levels in experiments and models.

(A-E) Brainstate dependent probability of late stimulation response. (A) Stimulation (60 ms air puff, left); recording field of view (2 mm scale bar, middle); Allen Mouse Brain Atlas applied to parcellate among cortical areas (right). (B) Group-averaged response to sensory stimulation under deep (upper) and medium (lower) isoflurane anesthesia. Under medium anesthesia, a stronger secondary response is typically observed. (C) Exemplary sensory-evoked calcium activity under medium anesthesia showing an early and a late response. (D) PCI of sensory-evoked response, showing a decrease in complexity with increasing anesthesia levels. E. Late response probability in deep (lighter colors) and medium (darker colors) anesthesia (12 cortical regions on both hemispheres. F-I. Whole-brain model of mouse in TVB, using AdEx mean-field models. Slow oscillations induced by increasing the Spike Frequency Adaptation value (parameter b); at low SFA the network expresses an asynchronous irregular regime. (F) Calcium signals in the mouse TVB model. (G) Calcium signal simulation for different anesthesia depths (top graph: asynchronous wake-like activity). (H) Power spectra of TVB signals. (I) response to stimulation (left) and PCI Estimation with focal stimulation (right). J. Interactive simulation of state-dependent spontaneous and evoked waves. (J) Propagation patterns evoked by stimulation in a mean-field mouse model inferred from calcium signals. The same network generates rich, state-dependent repertoires of spontaneous and evoked wave propagation patterns. Dorsal view of the cortical hemisphere model (pixel size 100μm, 25 mm2 field of view). EBRAINS LINK: https://wiki.ebrains.eu/bin/view/Collabs/interactive-exploration-of-brain-states.

Brain responsiveness across scales and species.

(A1-A4) Live, data from (Arena et al., 2021; Casarotto et al., 2016); Comolatti et al., 2025). (A1) from left to right, the scalp-EEG response to TMS stimulation in human subjects in two conditions (e.g. wake vs. sleep or healthy vs. UWS), their topographical representation and the associated PCI values (triangles) with respect to a distribution obtained from a benchmark population (modified from Casarotto et al., 2016). (A2,3) the scalp-EEG and the intracranial EEG (respectively) response to SPES in wakefulness and NREM sleep from the same human subject, their topographical representation and the associated PCI values (triangles) with respect to a distribution obtained from a benchmark population (modified from Comolatti et al., 2025). (A4) PCI measurements in rodents. Intracranial EEG responses to perturbations by brief electrical stimulation from an intracortical electrode in area M2, during wakefulness and general anesthesia (propofol) in the same rat, their topographical representation, and the associated PCI-ST values (triangles) (modified from Arena et al., 2021). (B1) Spontaneous local field potential activity from mice was recorded during SO with a superficial 32- channel multielectrode array (MEA) placed on the cortical surface (scale 500 μm). The pink circle indicates the location of the stimulation electrode. (B2) shows three representative recordings carried out at three different levels of anesthesia (deep, in dark blue, medium in blue, cyan in light blue respectively, color coding consistent in this panel). (B3) shows the average frequency of the SO. (B4-6) show the spatiotemporal MUA responses of all 32 channels (left) and the binary matrix (right) during the first 2s after stimulus onset during three different anesthesia levels (respectively, deep, medium and light). The spatial profile is shown at the bottom of each matrix on a visual representation of the recording MEA at three different time points (t1 to t3). Overall magnitude of perturbational complexity values under evoked (B7); Friedman p=0.0046; Wilcoxon Deep-Midp=0.19, Deep-Light p=0.0078, MidDeep p=0.15 and spontaneous (B8); Friedman p=0.19 conditions in our population of mice. B1-B8 modified from Dasilva et al., 2021.

Modeling responsiveness at the whole-brain level.

(A) The brain network model is constructed using a connectome derived from empirical DTI data, equipping the nodes with three different neural mass models to ask mechanistic questions: the phenomenological Hopf model (B,C) using response to stimulation to find support for either fluctuating (subcritical) or oscillatory (supercritical) regimes, the Montbrió Pazo Roxin Model (G,H,I) relating the dynamical features of spontaneous activity to the PCI, and the biologically most interpretable AdEx mean field (D,E,F) model investigating the role of neuronal adaptation in the complexity of the whole-brain response to stimulus. (B) The Hopf model was tuned to the sub- and super-critical regime and systematically perturbed with continuous stimulus applied with varying strength to the homotopic nodes. (C) The relative PCI after the stimulus varied across the nodes and stimulus strength across the nodes in the sub-critical regime, but there was almost no structured response in the super-critical regime. (D) Spatio-temporal propagation of the stimulus in the wake-like (low adaptation) and sleep-like (high adaptation) states, color codes time to significant deviation from baseline after the stimulus. (E) Excitatory firing rate of the stimulated brain region before and after the stimulus in the wake- and sleep-like dynamics. (F) Perturbation complexity index (PCI) of the whole-brain AdEx model in simulated anesthesia, either by NMDA block or by GABA-A potentiation, as well as in NREM sleep (high adaptation) condition. (G) For the MPR model, the working points with respect to the fluidity of spontaneous activity and Lempel-Ziv complexity of the stimulus response overlap for the parameter G that scales the network coupling, and are shifted for the noise a. (H) Four measures of the spontaneous EEG track the PCI in the preliminary dataset of 18 subjects (left to right, top to down): fluidity computed as circular correlation in sliding 3 s window, Lempel-Ziv complexity of the z-scored and binarized signal, bursting potential, and number of unique activation patterns. C modified from Sanz Perl et al., 2021, F modified from Sacha et al., 2025; G-H modified from Breyton et al., 2024.

State-dependent modulation of synaptic responses and model simulation.

(A) Auditory-evoked local field potential (LFP, top) and intracellular postsynaptic potential (PSP, bottom) recorded in rat primary auditory cortex. (B) LFP (top) and intracellular recording (bottom from auditory cortex during slow oscillations. (C) PSPs evoked by a 72 dB auditory stimulus during Down (red) and Up (black) states, illustrating the amplification of weaker inputs in the active network state. Averaged traces below. (D) Normalized PSP amplitude (relative to the Down-state response) plotted as a function of stimulus intensity for intracortical (IC) electrical stimulation. Circles, Up state; squares, Down state. Note potentiation at low intensities and attenuation at high intensities during Up states. (E) In a thalamocortical model, modulation factor (ratio of Up- to Down-state PSP amplitude) plotted against the Down-state PSP amplitude for IC (intracortical, dashed line) and thalamocortical (TC; solid line) stimulation. The TC pathway shows a larger overall gain modulation, reflecting the combined effects of thalamic and cortical network excitability. Modified from Reig et al., 2015.

(A) Butterfly plot of 60 channels (black traces) showing averaged TMS-evoked potentials from 150 trials during wakefulness. (B) Voltage maps at selected latencies, ranging from maximum (+100%) to minimum (−100%) values. (C) A weighted minimum norm inverse solution using a three-sphere BEM model estimates cortical currents. (D) Nonparametric bootstrap statistics identify significant TMS-evoked sources. (E) A binary spatiotemporal map [SS(x,t)] is constructed, with 1 indicating significance. Sources are sorted by total activity post-stimulus. PCI is computed as the Lempel-Ziv complexity of SS, normalized by its source entropy. The TMS stimulation site is marked with a green star. Modified from Casali et al., 2013.

Microstructure of brain responses.

Fine structure of spontaneous and evoked patterns of neural firing, in awake and anesthetized mice. Top: scheme of two-photon imaging of mice together with sound stimuli (50 different sounds were presented). The same population of neurons was imaged in awake and anesthetized mice. Left panels: patterns of response in awake mice. The bottom graph shows a compact representation of the correlated patterns of neural firing ("neuronal assemblies") that appear spontaneously or appear in response to the 50 different sounds. The color codes for the correlation between patterns. Right panels: same representation for anesthetized mice. Adapted from (Filipchuk et al., 2022).