Fig 1.
The brain’s GS fluctuation (standard deviation of GS) decreases with time of day.
(A) Between-participant variation in GS fluctuation as a function of time of day in session 1. (B) Between-participant variation in GS fluctuation as a function of time of day in session 2. (C) Within-participant variation in GS fluctuation Δ as a function of time of day Δ, where Δ denotes difference between session 2 and session 1. Grey dots denote individual participants. Black dots show mean of GS fluctuation in (A, B) hourly or (C) 3-hourly time windows. Line of best fit (red) was calculated based on data from all participants in each plot. Confidence interval is shown in light red. R values denote Pearson r correlation coefficient. p-Values were derived from 100,000 permutations, while keeping family structure intact [40]. (D) GS fluctuation is elevated in run 2 compared with that in run 1 despite downward shift in GS fluctuation as a function of time of day. Bar plots denote mean GS fluctuation across participants within 3-hourly time windows for each run. Error bars denote standard error of the mean. Two opposing effects are observable: a fast increase in GS fluctuation on the scale of minutes—i.e., run effect (green arrows)—superimposed on a downward drift of GS fluctuation occurring on the scale of hours—i.e., time of day effect (violet arrow). See S1 Data for underlying data. GS, global signal.
Fig 2.
Head motion, respiratory, and cardiac measures are strongly correlated with GS fluctuation, yet only respiratory variation shows association with time of day.
(A) Between-participant correlations of 13 run-level summary metrics with GS fluctuation. (B) Between-participant correlations of 13 run-level summary metrics with time of day. Because of exclusion of individuals with poor physiological data quality, different subgroups of participants were used for analyses of head motion (session 1: N = 942, session 2: N = 869), respiratory (session 1: N = 741, session 2: N = 668), and cardiac measures (session 1: N = 273, session 2: N = 272). Correlation between GS fluctuation and time of day was repeated in each subgroup. Numbers denote Pearson correlation coefficients. Stars indicate significant correlations following FDR correction (*q < 0.05; **q < 0.01; ***q < 0.001). See S2 Data for underlying data. AD SD, standard deviation of absolute displacement; DVARS SD, standard deviation of the temporal derivative of root mean square variance over voxels; FD SD, standard deviation of framewise displacement; FDR, false discovery rate; GS, global signal; HR RMSSD, root mean of the successive differences of heart rate; HR SD, standard deviation of instantaneous heart rate; RV SD, standard deviation of respiratory variation.
Fig 3.
Negative association between time of day and GS fluctuation remains significant after controlling for respiratory variation.
(A) Between-participant variation, (B) within- participant variation, and (C) run effects of RV SD as a function of time of day. (D) Between-participant variation, (E) within-participant variation, and (F) run effects of GS fluctuation residual (after regressing RV SD) as a function of time of day. Same as in Fig 1, within-participant effects were computed by taking the difference (Δ) for each variable between session 2 and session 1. Grey dots denote individual participants. Confidence interval is shown in light blue or in light pink. Black dots denote mean of GS fluctuation in hourly (A, D) or 3-hourly (B, E) time windows. R values denote Pearson correlation coefficients. p-Values were derived from 100,000 permutations while keeping family structure intact [40]. This figure shows the results for session 1 (see S3 Fig for session 2). See S1 Data for underlying data. GS, global signal; RV SD, standard deviation of respiratory variation.
Fig 4.
(A) Four hundred–area cortical parcellation in fs_LR surface space [45]. Parcel colours correspond to 17 large-scale networks [47]. Image reproduced under a CC BY 4.0 license, credit: https://doi.org/10.6084/m9.figshare.10062482.v1 (B) Nineteen subcortical regions defined in participant-level volume space [46]. Image reproduced under a CC BY 4.0 license, credit: https://doi.org/10.6084/m9.figshare.10063016.v1. DC, diencephalon; ROI, region of interest; VentAttn, ventral attention.
Fig 5.
BOLD signal fluctuation is negatively correlated with time of day across cortical and subcortical regions during session 1 (n = 942) and session 2 (n = 869).
Between-participant correlations between time of day and BOLD signal fluctuation for (A) 400 cortical and (B) 19 subcortical regions of interest. With the exception of brainstem, all subcortical regions are bilateral and presented as left-to-right hemisphere pairs (top to bottom). Colours (cool–warm) denote brain regions with significant Pearson r coefficients (q < 0.05, FDR-corrected), whereas nonsignificant regions are shown in grey. p-Values were derived from 100,000 permutations while keeping family structure intact [40]. BOLD, blood oxygen level–dependent; ventral DC, diencephalon; FDR, false discovery rate; Hippocamp., hippocampus; S1, session 1; S2, session 2.
Fig 6.
RSFC is negatively correlated with time of day across participants, with a magnitude surpassing the strength of correlation between fluid intelligence and RSFC.
Results shown are for session 1 (see S5 Fig for session 2). (A) Correlation between time of day and RSFC across participants. (B) Correlation between fluid intelligence and RSFC across participants (with inverted colour scale to facilitate visual comparison with time of day effects). Colours in lower triangular of correlation matrix denote Pearson correlation coefficients. Colours in the upper triangular denote r values from the lower triangular averaged within network pairs. Colours on label axes denote correspondence of 419 regions to 17 large-scale cortical networks and to SC. Association between time of day and RSFC was significant in both sessions as assessed by network-based statistics (FDR-corrected at q < 0.05), whereas association between fluid intelligence and RSFC association was only significant in session 1 (FDR-corrected at q < 0.05). See Materials and methods for details of network-based statistics [51]. Fluid intelligence was chosen because it is a widely studied measure amongst resting-fMRI studies of brain-behavioural associations [50] and is one of the behavioural measures that is best predicted by resting fMRI [37,38,52]. Con, control network; DAN, dorsal attention network; DMN, default mode network; FDR, false discovery rate; fMRI, functional MRI; Lim, limbic network; RSFC, resting-state functional connectivity; Sal, salience network; SC, subcortical network; SM, somatomotor network; TP, temporal parietal network; Vis, visual network.