Fig 1.
Brain-wide connectivity changes in W and NREM stages.
(A) FC between brain areas and BF, and FC between brain areas and LC, across sleep stages. Each point is the value for one brain area, averaged across individuals (multiple comparisons). Horizontal black lines represent significance (T-test, p < 0.0001). (B) Glass brain plots, displaying the spatial distribution of FC with BF and LC, correlation value displayed in color. (C) LC and BF position in the brain. AAL90 areas displayed in light blue (not individually labeled). (D) Functional nodal strength (sum of incoming FC values) across sleep stages.
Fig 2.
(A) Empirical priors included to construct our whole-brain model. (B) Nodal dynamics scheme, exemplifying how ACh and NA hypotheses are included. ACh is taken as a modulator of global cerebral coupling, while NA modulates the input-output slope of each node.
Fig 3.
Optimal parameters and goodness of fit across stages and modalities.
(A) Optimal parameter variation ( and
), where the euccorrelation metric between empirical and simulated FC matrices is minimized, displayed for all stages in the homogeneous and heterogeneous simulation cases. Arrows represent the direction of change from one state to the next (note that for homogeneous modulation W and N2 are in the same spot). (B) Optimal parameters for the homogeneous and heterogeneous simulation cases, displayed side by side for each stage, for ease of view. Notice the apparent “flipping" of the y axis for the coupling parameter, which arises because we are modeling a decrease in ACh as an increase in G. (C) Violin plots comparing the euccorrelation metric between the simulated and empirical FC matrices, in the optimal, across 50 seeds, for all stages (* for p < 10−3, ** for p < 10−4, *** for p < 10−22, **** for p < 10−50, see Cohen’s D interpretation in Methods).
Fig 4.
Empirical and optimal simulated matrices across stages and modalities.
Empirical and simulated optimal whole-brain FC matrices for the homogeneous and heterogeneous simulation case, displayed side by side, for all stages (rows).
Fig 5.
LC and BF effects on the integration/segregation profiles in W and NREM.
(A) Empirical nodal integration and segregation components across stages. Each dot is the value for one brain area, averaged across individuals. (B) Empirical nodal integration component versus nodal FC connectivity with the LC, and nodal segregation component versus nodal FC connectivity with the BF, in W. ρ is the Pearson’s correlation coefficient between the variables. LC-FC vs segregation () and BF-FC vs segregation (
) are shown in the small panels within. (C) Correlation between simulated and empirical integration (segregation) components, for all stages and modalities (homo, map and shuffle). Each point is the correlation between the simulated integration (segregation) component of one seed with the integration (segregation) component of the average empirical FC of the corresponding stage. A higher value implies a better fit to the empirical integration (segregation) profile.
Table 1.
Cohen’s D effect size between the empirical-simulated similarity measure (Pearson’s coefficient), measuring how much better or worse one modality of simulation is over another (for example, map-homo D = 3.88 for N1 Integration means that simulations with map have consistently higher correlation with empirical Integration distribution than homogeneous simulations, with a Cohen’s D value of 3.88).