Post by Elisa Guma
What's the science?
The human brain is a complex system that must integrate different cognitive and behavioural processes in response to changing environmental demands. It is theorized that in order to do so, the brain has a ‘dynamic core’ set of regions that together act as a master controller. This ‘dynamic core’ would guide the flow of activity within the brain to facilitate cognition, and integrate more specialized brain regions depending on the task at hand. Analyzing these complex brain networks can give us some clues as to how the brain is organized, however, little is known about how these networks adapt as a function of cognition. This week in Nature Neuroscience, Shine, Breakspear and colleagues aimed to investigate the spatial, dynamic, and molecular signatures of neural activity across a range of cognitive tasks using whole-brain functional magnetic resonance imaging (fMRI) data.
How did they do it?
The authors used fMRI data from the Human Connectome Project to examine blood-oxygen-level-dependent (BOLD) activity from 200 unrelated individuals while they performed seven cognitive tasks, each of which engages distinct cognitive functions (emotional processing, gambling, mathematical calculation, language processing, motor execution, working memory performance, relational matching, and social inference). The analyses were first developed in 100 participants (discovery data) and replicated in another 100 participants (replication cohort) to ensure reproducibility. BOLD time series data was extracted from 375 different cortical and subcortical regions or ‘parcels’ in the brain, and concatenated across all seven fMRI tasks, and all subjects. The authors then applied a data dimensionality reduction technique (principal component analysis) to the multitask BOLD time series data to reorganize the regional BOLD data into a set of smaller principal components (i.e. distinct patterns of activity within the data).In other words, instead of analyzing 375 different BOLD signals over time in 375 different brain regions, they grouped brain regions together into several different groups or ‘principal components’, such that, using only those components, they could still explain the variance in the signal. Next, the authors created a time series for each principal component (tPC) by weighting the original BOLD time series from the replication data, with the parcel loading from the discovery data set at each time point in the experiment. In other words, the authors were able to use their understanding of how each brain region contributed to each principal component in the discovery data set to explore the data weighted in the same way in the replication data set. To further investigate the first tPC (tPC1), they divided its time course into phase segments based on the shape of tPC1’s signal (i.e. whether it was in a trough, increasing, in a plateau, or decreasing), and tried to determine whether parts of the signal were more associated with certain tasks. The authors also assessed spatial overlap of these networks (principal components) in the brain with other resting state networks. In order to investigate the relationship between cognition and the brain’s dynamics, they used NeuroSynth to gather meta-analytic data from 5,809 fMRI studies in which individuals performed a variety of tasks, and identified four topic families (motor, cognitive, language, memory) They then weighted original BOLD data with the spatial activation pattern of each topic family. Finally, to determine the biological relevance of the brain dynamic patterns associated, they used Allen Brain Micro-Array Atlas to identify the spatial coverage of two groups of receptors in the brain; one known to facilitate and another to inhibit cognitive functions.
What did they find?
The authors found tPC1, which explained 38.1% of variance in the BOLD signal, was strongly correlated with all seven tasks, potentially pointing to a task invariant brain network that might represent the ‘dynamic core’ of the brain. The other tPCs were associated with more specific cognitive tasks. tPC1 also had spatial overlap with various known resting state networks such as the dorsal attention, frontoparietal, and visual networks, along with stratum, thalamus, and lateral cerebellum. Interestingly, they also found that greater recruitment of this network was associated with fluid intelligence (based on scores on Raven’s progressive matrices task — a common intelligence test), providing further support for the ‘dynamic core’ functions of the system. In their investigation of the different phases of tPC1, the authors found phases with greater frequencies to be associated with task blocks, and lower frequency phases to be associated with rest blocks between tasks. Interestingly, they found that the brain network phase and type of cognition task were segregated based on their cognitive topics analysis; motor and cognitive topics occupied higher phases of the networks (i.e. BOLD signal higher), whereas memory and language were associated with low phases (i.e. BOLD signal lower).
Finally, when the authors assessed the relationship between spatial maps of neurotransmitter receptors and the principal components, they found somewhat positive associations between the facilitatory receptor group and tPC1, but somewhat negative associations with the inhibitory group. Interestingly, they found that the second tPC (tPC2) differentiated the receptor families better versus tPC1; a spatial map of tPC2 brain regions mapped positively onto a spatial map of monoaminergic receptors, and negatively onto a map of serotonergic and cholinergic receptors. This suggests that global brain state dynamics may be controlled by recruitment of distinct neurotransmitter classes.
What's the impact?
These results present a novel view of brain function based on the coordinated dynamics of functional brain networks over time. The authors were able to identify a set of brain regions that co-activated across many different cognitive tasks. Interestingly, the authors suggest that this dynamic brain network may be shaped by different families of neurotransmitter systems, providing plausible biological support underlying brain dynamics. Finally, future work could investigate whether these networks are present in other species, such as non-human primates, to understand whether this dynamic network is distinctive of human cognition.
Shine, Breakspear et al. Human cognition involves the dynamic integration of neural activity and neuromodulatory systems. Nature Neuroscience (2019). Access the original scientific publication here.