Post by Shireen Parimoo
What's the science?
Brain networks are groups of brain regions that are co-activated while someone is doing a certain task, or while they are resting. For example, the frontoparietal network is active during tasks involving planning and problem solving (i.e. executive function). If an unexpected stimulus is encountered, a different network - the ventral attention network - will become active to redirect attention to that stimulus. Previous studies have found that being able to flexibly switch from one network to another is associated with motor learning and performance on tasks involving executive function. However, it remains unclear how network switching is related to the spatial and temporal dynamics of brain activity, and how the transition between network states is related to behavior. This week in PNAS, Pedersen and colleagues used functional magnetic resonance imaging (fMRI) data to investigate the network switching dynamics of brain regions, and their relation to behavioral outcomes.
How did they do it?
The authors used the resting-state fMRI data of over 1000 participants from the Human Connectome Project. First, they performed an independent component analysis (a technique that isolates statistically independent sources of brain activity from noise) to identify 25 distinct brain regions that were active in all participants. They then used a correlation-based sliding-window analysis of the fMRI data, which estimates the functional connectivity (i.e. synchrony) of those regions over time. Higher correlation coefficients indicate greater functionally connectivity between regions. The functional connectivity data was used as input into a multilayer modularity algorithm, in order to identify separate modules or networks in the brain. This algorithm estimates two parameters: (i) the temporal coupling strength of brain regions between adjacent time points, and (ii) the spatial resolution of networks (i.e. an estimate of the number of networks). These parameters were used to determine the rate of network switching based on how often a brain region’s network assignment changed. The authors then examined the relationship between network switching and spatiotemporal connectivity. The standard deviation of the sliding-window correlations was used as a measure of dynamic spatial connectivity of each brain region, with higher values reflecting more dynamic connectivity (meaning more variable fluctuations in connectivity over time). The sample entropy of the five nodes with the greatest and the lowest rate of network switching were calculated to obtain the temporal complexity of fMRI signals; higher entropy reflects random signals, and lower entropy values reflect structured and predictable signals. Finally, they used elastic net regression in order to determine whether whole-brain network switching predicted behavioral outcomes. This is a data-driven statistical technique that automatically selects variables that are predicted by network switching and penalizes variables not predicted by network switching. Behavioral variables included cognitive, social, personality, physical, and sleep-related outcomes for each participant.
What did they find?
Greater network switching was associated with reduced dynamic functional connectivity between brain regions and occurred more frequently when global network connectivity was low. Brain regions that had high rates of network switching also exhibited greater sample entropy in their fMRI signal, including the superior parietal lobule, the frontoparietal lobe, and the intraparietal sulcus, which are part of the frontoparietal network. This means that fMRI connectivity in brain regions that switch between networks frequently is more random and less predictable across time, compared to regions with lower network switching rates, which have relatively structured and stable temporal signals.
The multilayer modularity algorithm revealed five brain networks across all participants, which included the frontoparietal network, the dorsal and ventral attention networks, the default mode network, and sensory network. The authors found that regions comprising the default mode, frontoparietal, and the attention networks had greater switching rates than sensory networks. Although the brain switched between the sensory network states less often, sensory regions were more connected to each other across time than other regions (i.e. greater dynamic connectivity). Average network switching rates across the brain predicted performance on tasks relying on working memory, planning, and reasoning.
What's the impact?
This study is the first to demonstrate the relationship between spatiotemporal connectivity dynamics in the brain and the rate at which the brain switches from one network state to another. High rates of network switching – particularly in frontoparietal regions – was also linked to better performance on cognitive tasks. These findings provide deeper insight into the network activity dynamics that support brain function and cognition.
Pederson et al. Multilayer network switching rate predicts brain performance. Proceedings of the National Academy of Sciences of the United States of America. (2019). Access the original scientific publication here.