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

Abstract

Models of cognitive function typically focus on the cerebral cortex and hence overlook functional links to subcortical structures. This view does not consider the role of the highly-conserved ascending arousal system's role and the computational capacities it provides the brain. We test the hypothesis that the ascending arousal system modulates cortical neural gain to alter the low-dimensional energy landscape of cortical dynamics. Here we use spontaneous functional magnetic resonance imaging data to study phasic bursts in both locus coeruleus and basal forebrain, demonstrating precise time-locked relationships between brainstem activity, low-dimensional energy landscapes, network topology, and spatiotemporal travelling waves. We extend our analysis to a cohort of experienced meditators and demonstrate locus coeruleus-mediated network dynamics were associated with internal shifts in conscious awareness. Together, these results present a view of brain organization that highlights the ascending arousal system's role in shaping both the dynamics of the cerebral cortex and conscious awareness.

Fig. 1. Sympathetic activity precedes network-level integration.

a Regional time series were extracted from the subcortical locus coeruleus (red), which is thought to alter multiplicative gain, and the basal nucleus of Meynert (green), which is thought to alter response gain, and they were compared to cortical BOLD signal and topological signatures during the resting state. b We observed an anterior-to-posterior travelling wave (velocity ~ 0.13 m s⁻¹) following peaks in τ_LC-BNM, which are shown on both the left (LH) and right (RH) hemispheres of a cortical flat map (and v.v. following peaks in τ_BNM-LC). c Lagged cross-correlation between ${\tau }_{\mathrm{LC}-\mathrm{BNM}}$ and PC for each parcel (faint line) and mean PC (solid line); dotted line depicts the zero-lag correlation, and the black lines depict the upper (lower) bounds of the block-resampled null model 95% CI. d Mean cortical participation coefficient (PC) preceding (left) and following (right) the zero-lagged ${\tau }_{\mathrm{LC}-\mathrm{BNM}}$ value, only the right hemisphere is shown, which is mirrored for Post. e Mean participation coefficient following peak ${\tau }_{\mathrm{LC}-\mathrm{BNM}}$ was higher in the right (red) vs. the left (blue) hemisphere (error bars represent mean± SEM, n = 200 parcels within a hemisphere, p < 0.001 green bar one-sided permutation test 5000 random permutations).

Fig. 2. LC and BNM mediated shifts in the brain-state energy landscape.

a An example energy landscape, which defines the energy required to move between different brain states: by increasing multiplicative gain, the LC should flatten the energy landscape (red); by increasing response gain, the BNM should accentuate the energy wells (green). b The topography of the energy landscape can be conceptualized as similar to the activation energy (E_A) that must be overcome in order to convert one chemical to another. c Empirical BOLD energy landscape as a function of mean-squared displacement (MSD) and TR of the baseline activity (E_A, black) and after phasic bursts in LC (E_LC, red) and BNM (E_BNM, green). d Empirical activation energy as a function of MSD averaged across lags $t=\left[10:15\right]$ TR during base baseline activity (E_A, Left) and following phasic bursts in LC (E_LC, red) and BNM (E_BNM, green). Relative to the baseline energy landscape, phasic bursts in LC lead to a flattening or reduction of the energy landscape, whereas peaks in BNM lead to an accentuation of the well by raising the energy landscape.

Fig. 3. Awareness of intrinsic state changes.

a Participants performing breath-awareness meditation (focus; blue) were trained to respond with a button press (orange) when they became aware (purple) that they had become distracted (i.e. their attention had wandered from their breath) and to then re-focus their attention (blue) on their breath. b We observed a peak in ${\tau }_{\mathrm{LC}-\mathrm{BNM}}$ (red; LC relative to BNM BOLD activity) ~4 s before the button press, which then returned to low levels in the 2–4 s following the button press. c The mean-squared displacement (MSD; dark orange) of TR-to-TR BOLD signal was increased above null values around the peak in ${\tau }_{\mathrm{LC}-\mathrm{BNM}}$, as well as following the re-establishment of attentional focus (in (b, c) grey shading depicts 95% CI of block-resampled null distribution). d We observed a peak in mean participation coefficient (PC) ~4 s (2 TRs) prior to the button press during the task. a–d Grey shading depicts mean-centred 97.5th and 2.5th percentile of block-resampled null distribution 5000 permutations, i.e. outside grey shading indicates a value different than null (p < 0.05); and (d) red shading represents mean ± SEM across (n = 400 parcels). Source data are provided as a Source Data file.