Fig 1.
Step-by-step analysis illustrated.
A, Subjects wake and sleep volumes where organized and submitted to group-level spatial independent component analysis (ICA). B, ICA was performed using the GIFT toolbox implemented in MATLAB; 42 neurophysiologically plausible sources were selected and sorted into functional families. GICA 1 back-reconstruction was used to estimate the time courses (Ri) and spatial maps (Si) for each subject. C, All time courses (TCs) were post processed by removing subject motion variance, despiking, and filtering. D, TCs were used to estimate mean FC in Wake and NREM2 by computing pairwise correlations between all ICs. E, Dynamic FC was estimated using a sliding window approach (window width = 15 TRs, time step = 1 TR) resulting in windowed correlation matrices. Correlation matrices were vectorized and concatenated into a large data matrix of all IC-to-IC pairwise correlation values over time. The concatenated data matrix was submitted to k-means clustering (F). k-means clustering was performed to extract recurrent features of the data. A k-7 solution was selected which resulted in a matrix with rows equivalent to the number of connectivity states (7) and columns equivalent to the number of unique IC-IC correlations (861). Each row representing a connectivity states was back-reconstructed into a matrix format to visualize coupling relationships between ICs. An IDX vector, a window state label vector, unique for each subject in Wake and NREM2 was computed revealing window classification under one of the 7 connectivity states.
Fig 2.
42 neurophysiologically plausible independent components (ICs).
42 neurophysiologically plausible independent components (ICs). Were divided into groups and arranged based on their spatial and functional properties. A total of seven functional families were identified including the auditory, somatomotor, visual, default mode, cognitive control (including the dorsal and ventral attention networks), subcortical, and cerebellar networks. ICs are displayed on sagittal, coronal, and horizontal slices on a cortical surface implemented in MANGO.
Fig 3.
Mean functional connectivity in Wake and NREM2.
Mean functional connectivity in Wake (A), and NREM2 (B). Within defined boundaries around the diagonal, positive correlations indicate strong coupling relationships between ICs that were classified within the same functional family. Off-diagonally, weaker connectivity can be found between functional families.
Fig 4.
Difference matrix between wake and NREM2 FC matrices.
Stationary Mean FC in Wake (B) subtracted from stationary mean FC in NREM2 (A) to produce a Difference Matrix (C). t tests were performed with the null hypothesis of zero correlation on the Difference Matrix (D). To correct for multiple comparisons, the false discovery rate (FDR) method was used with a P value of .01. t-tests confirmed that mean FC in Wake and NREM2 were similar with five correlation differences that were greater than 0.
Fig 5.
Mean number of transitions (NT) and inter-transition interval (ITI).
A, Mean number of transitions (NT) in Wake and NREM2. Participants expressed more transitions in Wake (M = .18; SD = .036) than in NREM2 (M = .15; SD = .033) (p < .05). B, The inter-transition interval was significantly higher in NREM2 (M = 7.08; SD = 1.96) than in Wake (M = 5.52; SD = 1.10) (p < .01).
Fig 6.
Cluster analysis, mean frequency, and mean dwell time.
Clustering analysis revealed 7 connectivity states in Wake and NREM2 (A). B, mean frequency of state expression in Wake (yellow) and NREM2 (blue). The frequency of connectivity state-1 and 6 were significantly greater in NREM2 (*, p < .05). The frequency of connectivity state-5 was significantly greater in Wake (*, p < .05); the frequency of connectivity state-4 expression in wake was marginally significant when compared to NREM2 (#, p = .053). C, Mean dwell time for connectivity state-1 was significantly greater in NREM2 than in Wake (*, p < .05).