The ascending arousal system shapes neural dynamics to mediate awareness of cognitive states

The study investigates how the ascending arousal system (notably the noradrenergic locus coeruleus, LC, and the cholinergic basal nucleus of Meynert, BNM) modulates large-scale cortical dynamics to support flexible cognition and conscious awareness. Despite relatively fixed structural connectivity, the brain exhibits rapid functional reconfigurations during cognition. The authors hypothesize that neuromodulatory systems modulate neural gain to shift the brain’s low-dimensional energy landscape, balancing integration and segregation. Given the few arousal neurons with widespread projections, the authors propose that arousal imposes low-dimensional control over cortical dynamics, with LC increasing multiplicative gain (promoting integration) and BNM increasing response gain (promoting segregation), thereby reshaping the probability and energetics of brain-state transitions.

The paper builds on anatomical and theoretical work showing neuromodulatory nuclei provide diffuse (LC) versus targeted (BNM) cortical projections and implement distinct gain-control computations (multiplicative vs divisive normalization). Prior frameworks link structural connectivity to integration/segregation but lack mechanisms for rapid functional flexibility; neuromodulation is posited to fill this role. Energy landscape approaches have described neural dynamics in spiking activity, fMRI, and MEG, often with binarization; the present work advances a continuous, displacement-based approach (mean-squared displacement, MSD) that avoids thresholding. Previous empirical and modeling studies suggest LC activity promotes network integration and traveling waves; cholinergic activity supports attentional precision and stability. The study also connects to theories of LC-mediated network reset, and microcircuit findings showing noradrenergic and cholinergic modulation of pyramidal dendritic excitability relevant to conscious processing.

Datasets: (1) 7T resting-state fMRI (59 healthy adults; TR=586 ms; 2 mm isotropic; ~10 min). (2) Meditation task fMRI at 3T (14 experienced meditators performing focused-breath attention; button press upon awareness of mind wandering). Preprocessing included realignment, skull-stripping, coregistration, segmentation, DARTEL normalization, nuisance regression (motion and derivatives, white matter, CSF, aCompCor), temporal band-pass filtering (0.01–0.15 Hz), and exclusion for excessive motion. Brain parcellation used the Schaefer 400-parcel cortical atlas. LC and BNM (Ch4) ROIs were defined via probabilistic anatomical atlases; mean BOLD from each was extracted. To mitigate non-neural contamination, signals were compared to nuisance regressors (CSF, white matter, frame-wise displacement, fourth ventricle sphere) and residualized for fourth ventricle and nearby pontine signals; results were robust to these controls. Phasic events: Identified phasic bursts via the second derivative of LC and BNM time series meeting thresholds: acceleration ≥ 2 SD; original series ≥ 2 SD above mean within following 10 TR; and not within first/last 20 TRs. Events categorized: LC relative to BNM (T_LC-BNM; n=148), BNM relative to LC (T_BNM-LC; n=130), and simultaneous LC+BNM (n=316). Traveling waves: Computed lagged cross-correlations between arousal signals and cortical parcels to extract time-to-peak within 10 TR after events; mapped timing to cortex and estimated wave velocity using parcel MNI coordinates. Time-varying connectivity: Used multiplication of temporal derivatives (MTD) with a 20 TR window to construct dynamic connectivity matrices. Community structure via Louvain modularity (γ=1.0) with consensus partitions across 500 runs per epoch. Computed participation coefficient (PC) per parcel as between-module connectivity measure. Energy landscape: Quantified brain-state displacement using BOLD mean-squared displacement (MSD) across parcels relative to event onset t0 over lags t=1–15 TR. Estimated P(MSD,t) via Gaussian kernel density; defined energy E = ln(1/P). Constructed baseline energy landscape from periods outside phasic events and compared to landscapes following LC, BNM, and LC+BNM phasic bursts. Evaluated changes around 10–15 TR (~6–9 s) where effects peaked. For LC+BNM, assessed nonlinearity versus linear superposition and fit dominance coefficients (α, β). Meditation analysis: Built a finite impulse response model around button presses (±5 TR) to assess LC>BNM activity, MSD, and network PC dynamics. Statistics: Non-parametric block-resampling null models and permutation tests (typically 5000 permutations) controlled for spatial/temporal autocorrelation; significance at p<0.05.

• Spatiotemporal traveling waves: Following LC>BNM phasic peaks, a frontal-to-sensory traveling wave propagated across cortex with velocity ~0.13 m/s; BNM>LC peaks showed the inverse (sensory-to-frontal) propagation. - Network topology: LC phasic bursts preceded increased global integration (higher mean participation coefficient), dominated by frontoparietal cortices, with relative segregation in limbic, visual, and motor regions. Integration increased earlier in the right hemisphere than left (p<0.001). LC activity correlated positively with network integration (p<0.05, block-resampled null). - Energy landscape modulation: Relative to baseline, LC phasic bursts flattened the energy landscape (reduced energy) particularly at 10–15 TR (~6–9 s), increasing probability of large state transitions; BNM phasic bursts deepened the landscape (increased energy), favoring local, stable trajectories. - Synergistic LC+BNM effects: Simultaneous peaks produced an energy landscape not explainable by linear superposition of individual effects and showed a shift from anti- to de-correlation with the HRF slice-wise topography. Fitting LC+BNM ≈ α·LC + β·BNM yielded α≈0.16, β≈0.84, indicating BNM-dominant influence consistent with LC projections to BNM. - Conscious awareness linkage (meditation): Around awareness of mind-wandering (button press), LC>BNM signal, TR-to-TR BOLD MSD, and network-level integration (mean PC) all peaked roughly 4 s prior to the press and returned toward baseline after, with effects exceeding block-resampled nulls (all permutation p<0.05).

Findings support the hypothesis that the ascending arousal system reshapes low-dimensional cortical dynamics and network topology. Noradrenergic LC activity transiently increases neural gain, promoting integration, traveling waves from frontal to sensory regions, and flattening the energy landscape to enable transitions among brain states. Cholinergic BNM activity increases response gain, deepening energy wells to stabilize states and promote segregation. The LC+BNM co-activation produces unique, nonlinear effects with BNM-dominant contributions, suggesting coordinated neuromodulatory cascades. The observed LC-linked reconfiguration precedes moments of conscious awareness of mind wandering, aligning with theories of LC-mediated network reset and supporting a systems-level bridge between microcircuit mechanisms of neuromodulation and macroscopic network dynamics associated with awareness. The work situates an egocentric MSD-based energy framework as a practical, continuous alternative to binarized allocentric approaches, offering intuitive insights into transition energetics while acknowledging it conflates trajectories with identical MSD. The results also relate to microcircuit evidence for noradrenergic and cholinergic modulation of pyramidal dendritic excitability and to clinical contexts where arousal-system pathology may disrupt global dynamics.

The study demonstrates that phasic activity in the ascending arousal system systematically reshapes cortical network topology and the low-dimensional energy landscape, with LC flattening and BNM deepening the landscape. These neuromodulatory effects facilitate dynamic integration/segregation and are temporally linked to internal shifts in conscious awareness. Contributions include: (1) a continuous, MSD-based energy landscape framework for fMRI; (2) empirical evidence connecting LC/BNM dynamics to network integration and traveling waves; (3) demonstration of nonlinear LC+BNM interactions; and (4) linking LC-mediated dynamics to awareness in meditation. Future work should delineate precise rules mapping specific neuromodulatory receptor dynamics to landscape deformations and cognition, develop methods to disambiguate distinct trajectories with similar MSD, extend analyses to longer recordings and multimodal data (e.g., EEG/MEG), and investigate alterations in clinical populations with arousal system pathology (e.g., AD, PD, disorders of consciousness).

• Hemodynamic confounds: T2* fMRI cannot fully disentangle neural from vascular effects; noradrenergic influences on neurovascular coupling may contribute to observed signals. - Regional specificity: LC ROI may include adjacent arousal nuclei due to proximity; despite controls (fourth ventricle and pontine regression), contamination cannot be fully excluded, particularly at 3T. - Energy framework: The egocentric MSD approach does not distinguish different trajectories that yield the same displacement, potentially conflating distinct state paths. - Content of consciousness: The study infers timing of awareness via button presses during meditation but cannot access the contents of thought directly. - Sampling and generalizability: Phasic event counts per subject were modest; while robust across thresholds and nulls, generalization to tasks and populations beyond the studied cohorts requires caution.