Dynamic reconfiguration of frequency-specific cortical coactivation patterns during psychedelic and anesthetized states induced by ketamine

The introduction frames ketamine as pharmacologically distinct from canonical anesthetics and classic psychedelics: it antagonises NMDA receptors and HCN1 channels and produces psychedelic or anesthetic effects depending on dose. Prior work has shown ketamine produces frequency-specific electrophysiological changes and network-level haemodynamic effects, but most studies used time-averaged measures that discard fast temporal structure. Measures of local signal complexity (e.g. Lempel-Ziv complexity) indicate dose-dependent effects of ketamine on signal diversity, but these reduce rich spatiotemporal patterns to a single scalar and typically probe only short time windows. The researchers set out to characterise how ketamine modulates the repertoire of spontaneous, frequency-specific cortical coactivation patterns over longer timescales using high-density EEG source-localised to cortical parcels. They extracted band-limited envelopes in canonical frequency bands (delta to gamma), applied k-means clustering to identify recurring spatial coactivation states, and quantified how fractional occupancy and transition dynamics of these states change across baseline wakefulness, subanesthetic (psychedelic) infusion, an anesthetic bolus (loss to recovery of responsiveness), and post-recovery rest. They hypothesised that subanesthetic ketamine would increase the repertoire richness (diversity) of brain states, whereas an anesthetic dose would reduce it, with anaesthetic effects potentially fluctuating over the course of the non-stationary bolus epoch.

The researchers re-analysed previously collected high-density EEG from 15 healthy volunteers who had undergone both subanesthetic and anesthetic ketamine dosing in earlier studies. Four periods were analysed: 5 min baseline eyes-closed rest; the final 5 min of a 40-min subanesthetic infusion (0.5 mg⋅kg-1 total) assumed to be at steady state; the anesthetic period from loss to recovery of consciousness following a 1.5 mg⋅kg-1 bolus (mean duration 10.1 ± 3.4 min); and 5 min eyes-closed rest after return of consciousness. The researchers treated baseline, subanesthesia and recovery as steady states but considered the anesthetic bolus epoch non-stationary. EEG acquisition used 128-channel nets sampled at 500 Hz and downsampled to 250 Hz; after excluding face/lower-head electrodes, 109 channels were preprocessed (detrending, low-pass filtering at 50 Hz, bad-channel interpolation, average re-referencing). Noisy 1-s epochs were rejected by amplitude-based criteria separately for each period (percent rejected reported per period). Source reconstruction used a weighted minimum norm estimate (wMNE) on an FSaverage template head model with a 15,002-vertex cortical grid, constrained normal to cortex and parcelled into 100 regions based on the Yeo atlas, each mapped to one of seven resting-state networks. Supplementary analyses used alternative templates, inverse methods and parcellations. For each of five frequency bands (delta 1–4 Hz, theta 4–8 Hz, alpha 8–13 Hz, beta 13–25 Hz, gamma 25–45 Hz) the band-limited envelope per region was computed via the Hilbert transform, demeaned, normalised by regional SD, and downsampled to 40 Hz using 100 ms sliding windows with 75% overlap (supplementary tests varied window length 50–1000 ms). Envelopes from all periods and subjects were concatenated into a feature matrix (m = 500 features = 5 bands × 100 regions; Nt ≈ 838,084 time points) and clustered with k-means (correlation distance, 100 initialisations). Cluster numbers 5–15 were tested and a 10-cluster solution was chosen for balance of temporal resolution and inter-subject consistency. Each cluster centroid represents a frequency-specific spatial coactivation pattern; dominant frequency band per state was defined by the largest positive or negative envelope values. Temporal dynamics per subject and period were characterised by fractional occupancy (time fraction in each state), mean dwell time (average continuous time spent in a state), mean interval time (time between visits), Shannon entropy of the occupancy distribution (normalized 0–1), state transition probability matrices (assuming first-order Markov dynamics, with Markovianity tested), transition rate (number of distinct-state transitions per unit time), and entropy rates for transition matrices (including persistence and only transitions). Statistical inference used linear mixed models (LMM) with subject random intercepts and restricted maximum likelihood for comparisons across states and periods; post-hoc paired t-tests with Bonferroni correction and non-parametric Wilcoxon signed rank tests were used for transition analyses. Surrogate analyses (1000 permutations of state sequences) assessed significance of specific transitions. All data were tested for normality and a two-sided p < 0.05 threshold was used; several robustness checks varied smoothing windows, number of clusters, distance metric, clustering algorithm, template anatomy, parcellation and inverse methods.

Clustering of band-limited envelope data yielded ten recurring, frequency-specific cortical coactivation states present across subjects and periods. Key spectral–spatial characterisations were: States 1 and 2—anticorrelated posterior patterns with opposing alpha-band activity in visual cortex and posterior cingulate (Pearson r = -0.880); States 3 and 4—sensorimotor-centred activation in beta and alpha bands respectively; States 5 and 7—visual-cortex activation in beta and gamma bands; State 6—delta-band activation in anterior regions overlapping anterior DMN (prefrontal and temporal cortex); States 8–10—global patterns prominent during anesthesia, with State 8 showing global theta activation and delta deactivation, State 9 global delta activation and gamma deactivation, and State 10 the reciprocal pattern (States 9 and 10 anticorrelated r = -0.779). Anesthesia-related states (8–10) exhibited lower spatial variability as measured by standard deviation and entropy of activation values across regions. Temporal properties across subjects: each state was visited by all subjects and occupied on average between 5.6% and 14.1% of time. States specific to baseline (1 and 2) or anesthesia (8–10) had larger occupancy (p < 0.05). Mean dwell times ranged approximately 84.4–202.2 ms across states; beta- and gamma-dominant states (3, 5, 7) had significantly shorter dwell times than other states (p < 0.001). State 6 (anterior DMN-like) had a relatively long mean dwell time 196.8 ± 43.1 ms and the longest mean interval between visits (3.1 ± 0.6 s, p < 0.001 versus other states). Effects of ketamine on state occurrence: fractional occupancy of posterior alpha states (1 and 2) decreased in a dose-dependent fashion and was reduced during anesthesia; occupancy partially recovered but did not fully return to baseline (p < 0.05). State 3 showed similar decreases; State 4 showed a small decrease only during anesthesia (p = 0.006). State 5 (visual beta) occupancy increased in subanesthesia and recovery (p < 0.01) but not during anesthesia; State 7 (visual gamma) increased in a dose-dependent fashion (p < 0.001). State 6 showed no significant occupancy change. Global patterns (States 8–10) occupied the largest proportion of time during ketamine anesthesia (p < 0.001) and reduced during recovery and subanesthesia. The entropy of the occupancy distribution increased at subanesthesia relative to baseline (p = 0.019; mean change 0.065, 95% CI 0.023–0.108). During anesthesia the occupancy entropy was highly variable across time segments (fourfold greater variability than other periods); the second quarter of the anesthetic epoch showed the lowest entropy (significantly below baseline, p < 0.001; mean change -0.096, 95% CI -0.139 to -0.053) and entropy increased across subsequent segments, returning to baseline by the final segment. Recovery entropy was higher than baseline (p = 0.006; mean change 0.075, 95% CI 0.032–0.119). State transition dynamics: testing supported the first-order Markov assumption. Persistence probabilities (staying in the same state) were significantly greater than expected by chance for all states (p < 0.05). Aggregated persistence versus switching probabilities were: baseline 84.0 ± 2.8% persistence vs 16.0 ± 2.8% switching; subanesthesia 80.3 ± 3.1% vs 19.7 ± 3.1%; anesthesia 83.0 ± 3.4% vs 17.0 ± 3.4%; recovery 79.3 ± 3.0% vs 20.7 ± 3.0%. Between-state transitions were structured (non-uniform; p < 0.05 across pairs) rather than random. Ketamine-modulated transitions: in subanesthesia there were reduced probabilities of remaining in States 1 and 3 and reduced transitions between State 3 and States 1/2, together with increased probabilities of staying in States 5, 7 and 10 and increased transitions between 5↔7 and 7→10 (p < 0.05). Anesthetic dosing produced an overall decrease in persistence and transitions involving States 1–4 and increased persistence and transitions for States 7–10 (p < 0.05). At recovery, several transition probabilities did not return to baseline (p < 0.05): persistence and transition probabilities involving States 1–3 remained low, while those involving States 5, 7, 9 and 10 remained high. The transition rate (percentage time transitioning to a different state) increased at subanesthesia (p = 0.006; mean change reported as 3.7%, 95% CI 1.5–5.9%) and at recovery (statistically significant though the extracted text does not clearly report the recovery mean). During anesthesia the transition rate varied across the four segments, increasing from segment 1 to 4, with segment 4 significantly higher than baseline (p = 0.04; mean change 3.1%, 95% CI 0.9–5.2%). Markov entropy (including persistence and transitions) increased at subanesthesia (p = 0.005; mean change 0.064, 95% CI 0.027–0.100) and recovery (p < 0.001; mean change 0.076, 95% CI 0.039–0.113), but was dynamic during anesthesia with the second segment significantly lower than baseline (p = 0.01; mean change -0.057, 95% CI -0.093 to -0.021). Entropy computed from only transition probabilities showed small, non-significant increases at subanesthesia and recovery and was significantly lower than baseline during all four intra-anesthetic segments (p < 0.05). Robustness checks varying smoothing windows, cluster number, anatomy template, inverse method, distance metric and clustering algorithm yielded confirmatory, dose-dependent changes in state occurrence and transition dynamics.

The authors interpret their results as evidence that ketamine dose-dependently reconfigures the repertoire of frequency-specific cortical coactivation patterns. Subanesthetic ketamine was associated with an expanded repertoire: elevated occupancy entropy, increased transition rates and higher Markov entropy, consistent with a more diverse set of brain configurations. An anesthetic bolus induced dynamic changes: an overall shift toward globally dominated, low-spatial-variability states during anesthesia, but with entropy and transition measures that fluctuated across the anesthetic epoch and in some cases returned to baseline-like levels prior to recovery of responsiveness. Thus, ketamine can produce brain dynamics that feature characteristics of general anaesthesia (reduced repertoire/diversity), normal wakefulness (comparable diversity), and psychedelic altered states (elevated diversity), depending on dose and time. The authors relate their electrophysiological, sub-second findings to prior fMRI-based work on brain state repertoires and to EEG/MEG measures of complexity, arguing that envelope fluctuations at the time scale of a few hundred milliseconds provide a bridge between fast electrophysiological dynamics and slower haemodynamic measures. They highlight state-level findings such as the posterior alpha anticorrelated pair (States 1 and 2), an anterior DMN-like delta state (State 6), and sensorimotor alpha/beta states, noting these align in part with prior MEG and fMRI decompositions but may differ due to modality, source reconstruction, segmentation method and behavioural conditions (eyes-closed versus eyes-open). Mechanistically, the selective reduction in posterior alpha states and preserved anterior DMN-like state are discussed as compatible with ketamine producing a disconnection between prefrontal and posterior regions that may underlie preserved vivid but disconnected subjective experiences during ketamine anesthesia. The increases in visual beta/gamma states are linked to prior observations of amplified high-frequency activity in visual cortex under ketamine. The authors acknowledge several limitations: reliance on template anatomy and default electrode positions may limit source localisation accuracy; deriving states from band-limited envelopes constrains spatial separability compared with methods that ignore spectral differentiation; envelope-based metrics capture different information from phase-based measures and might miss complementary dynamics; the aperiodic 1/f background could be a confound in spectral characterisation; and k-means assumes discrete, mutually exclusive states whereas brain activity might follow continuous trajectories. They note that despite these limitations, results were robust across multiple methodological variants and that future work could address subject-specific anatomies, phase-based analyses, separation of oscillatory versus aperiodic components, and methods that permit continuous state representations. In sum, the authors conclude that frequency-specific coactivation patterns and their temporal transitions provide a tractable description of ketamine-induced changes in whole-brain dynamics, and that ketamine is a useful pharmacological probe spanning a range of brain-state repertoire configurations relevant to consciousness research.