Naturalistic global fMRI activity has an underlying signal which flips between two anti-correlated activity states.

(a) Distinct bands appear in the ROI averaged whole-brain time series, implying discrete states. (b) The temporal correlation matrix of the timeseries shows how similar the activity pattern at one timepoint is to all others. It shows prolonged states in which the neural activity patterns are similar over time, which are anti-correlated with the subsequent state. (c) The meta-temporal correlation matrix shows how similar neural patterns at two timepoints are in terms of their relationship to all other timepoints. It shows that the majority of timepoints are either strongly correlated or strongly anti-correlated to each other. (d) Left: The rows of the meta temporal correlation matrix (the relationship of one timepoint to all other timepoints in panel c) are distributed either unimodally or bimodally. Middle: We can thus define three distinct and discrete whole-brain states: Two anti-correlated states and an orthogonal transition state. These three states occupy roughly equal time during naturalistic activity. Right: Transition matrix shows the temporal structure of discrete states. Two anti-correlated states rarely transition to and from one another. A unimodal state mediates the transition from one to the other. (e) Whole-brain timeseries and corresponding brain states. We repeated analyses we conducted on the whole-brain activity for each local ROI (voxel by time) activity timeseries. (f) Local ROIs have fewer unimodal (green) neural states. Each column shows how much of the whole timeseries is spent at each state for an ROI. (g) The mean BOLD activity of one of the bimodal states is consistently higher than its anti-correlational counterpart. Therefore we can call both ends of the local anti-correlated patterns the up and down states. Histogram shows the distribution of t-values resulting from Welch’s t-test comparing the mean activities of the two states across all ROIs. Brown bins show significant t-values and blue bin shows t-values which fall below the significance threshold.

Anti-Correlated signals across spatial scales are related.

(a) Global and local anti-correlated patterns mapped on the whole cortex. Top: Global anti-correlated pattern which differentiates the Default Mode State (blue) and Task Positive State (red) visualised on the cortex. Bottom: Local anti-correlated patches of voxels visualised on the cortex. (b) Top: Global whole-brain (ROI averaged multivariate timeseries), and Bottom: an example ROI’s local (within ROI multivariate voxel timeseries) timeseries is plotted above the anti-correlated fluctuations of the template shown in line plots. (c) Left: Ten example voxel timeseries from the left parietal inferior lobule which were selected at the extreme ends of the template patterns to illustrate the presence of locally anti-correlated voxels, and Right: time by time correlation matrix of left parietal inferior lobule. (d) Global - Local anti-correlated state switches are most strongly aligned in the Dorsal Attention Network (DAN) areas, specifically FEF and SPL

Stimuli boundaries are followed by a global transition to the TPS, and coincide with transitions to localized up states

(a) Increase in probability of being in the up state spreads across the cortex after event boundaries. (b) Probability of being in the TPS increases after event boundaries. (c) Increase in probability of being in the up state compared to baseline spreads across the cortex after shot changes. (d) Probability of being in the TPS increases after shot boundaries. (e) Increase in probability of being in the up state spreads across the cortex after MFCC changes. (f) Probability of being in the TPS increases after MFCC boundaries. (g) Stimuli changes lead to a cascade of local up state transitions across the cortex and then dissipate. Results are visualized as state probability changes around boundaries, with significant timepoints (p<0.05 after FDR correction) marked with asterisks

Key brain areas and networks annotated with the global anti-correlated templates.

a) The annotated cortical map (same map in 2a) shows which ROIs participate in which of the global anti-correlated states. Blue shows the default mode state and red shows task positive state. b) Correlation between the each of the functional brain networks defined by Thomas Yeo et al. (2011) and global anti-correlated state for each dataset. Brown is default mode state and blue is task positive state.

Local template patterns in more detail.

(a) Voxel-wise Global Template. Each row shows the same data but with different colour map ranges to highlight the contrast within and between local and global scales. (b) Two example voxels from a visual ROI shows how neighbouring voxels show anti-correlated activities. (c) Local template patterns computed from non-hyperaligned data.

Anti-correlated voxel timeseries and time by time correlation matrices for example ROIs.

Here, we show data from three example ROIs. For each, 10 voxels were selected from extreme ends of the template pattern to illustrate the local anti-correlation. These three specific ROIs were selected to illustrate the range of the degree of anti-correlation across ROIs. Although most ROIs contain clearly anti-correlated groups of voxels, the degree of anti-correlation is much less prominent for some ROIs.

Relationship between various stimuli boundaries and local and global states for CamCAN

(a) Increase in probability of being in the up state spreads across the cortex after cause boundaries. (b) Probability of being in the TPS increases after cause boundaries. (c) Increase in probability of being in the up state compared to baseline spreads across the cortex after goal changes. (d) Probability of being in the TPS increases after goal boundaries. (e) Increase in probability of being in the up state spreads across the cortex after time changes. (f) Probability of being in the TPS increases after time boundaries. Results are visualized as state probability changes around boundaries, with significant timepoints (p<0.05 after FDR correction) marked with asterisks. Empty silhouettes show that no statistically significant ROIs are found.

Relationship between various stimuli boundaries and local and global states for Narrattention

(a) Increase in probability of being in the up state spreads across the cortex after event boundaries. (b) Probability of being in the TPS increases after event boundaries. Results are visualized as state probability changes around boundaries, with significant timepoints (p<0.05 after FDR correction) marked with asterisks

Global-Local state boundary overlaps (replication of 2) for all datasets.

(a) Cortical maps showing the alignment between local ROI state transitions and the global brain state transitions. (b) Bar plots showing how the global-local alignment is distributed across functional networks.

Comparing QPPs and anti-correlated templates.

The anti-correlated patterns we uncover are almost identical to the QPPs defined by Xu et al. (2025). QPPs constitute one cycle of the anti-correlated template we uncover (see both a and b, right). a) Left: ROI by time whole-brain activity. Middle: The anti-correlated template pattern. Right: The QPP, with the similarity of each timepoint of the QPP to our anti-correlated template. Similarity is quantified as Pearson r. b) Left: voxel by time local ROI activity. Middle: The anti-correlated template pattern. Right: The QPP, with the similarity of each timepoint of the QPP to our anti-correlated template. Similarity is quantified as Pearson r.