‘Network Neuroscience’ Offers a Better Understanding of Brain Function

Neuroscientists know how to generate a variety of “pictures” of the brain—structural and metabolic pictures, brain-wide and fine-scaled pictures, even maps of the brain’s major interconnections—but one they traditionally have lacked is a useful picture of the brain as a functional network.

Engineers routinely make such network diagrams to design or illuminate the workings of computers and other machines. Often they’d be lost without such models. Now neuroscience is developing its own network-based approach to understanding brain function.

“There has really been a sea change in the field,” said Indiana University researcher Olaf Sporns. “We’re starting to look at brains as systems with interconnections, where those interconnections matter a lot—in other words, we’re not just looking at one part of the system at a time, but at the dynamics and the architecture of the whole.”

A common network-neuroscience approach is to use neural imaging data to make a map of the major connections among regions, and then to turn that map into a more abstract network diagram in which distinct brain regions are “nodes” and their interconnections (“edges”) have differing strengths. Applying network theory to such a diagram yields a model of brain function and associated behavior, which can then be optimized by checking it against real-world brain function and behavior.

The hope is not only to get a better handle on how the brain normally works, but also to find better ways to understand, diagnose, and correct brain dysfunctions due to disease or injury.

“Our approach stems from the recognition that the human brain may respond to stimuli with principles akin to man-made systems, such as power systems and robotic networks,” said Fabio Pasqualetti, a researcher at the University of California, Riverside.

A flock of thoughts

In one recent study a group including Pasqualetti, and led by senior investigator Danielle S. Bassett of the University of Pennsylvania, used network neuroscience concepts to suggest how activity in various individual brain regions contributes to the overall dynamics of human cognition. Such information may help us understand, for example, how the brain switches from one task to another.

It may also help us design more efficient therapies, such as those using transcranial magnetic stimulation (TMS). In the context of TMS and similar therapies, said Pasqualetti, “there is a critical need for biologically informed computational models and theory for predicting the impact of focal neurostimulation on distributed brain networks.”

For the study, Bassett and her colleagues used an advanced MRI technique called diffusion spectrum imaging to make maps of the major nerve-bundle connections among 234 separate brain regions in human volunteers. The team then used the connection data, along with advanced network theory, to model the brain as a network with nodes and edges. The goal was to estimate the relative ability of different brain regions to influence the dynamics of the wider system.

The analysis suggested a few things. First, several of the modeled regions appear well suited to making small changes in the system—pushing it, with minimal effort, towards other brain states that are relatively easy to reach. These are physically central and highly interconnected regions that include the precuneus, known to be involved in executive functions, working memory, and self-awareness; the posterior cingulate, an adjacent region involved in memory, awareness, and cognitive control; and the superior frontal gyrus, a frontal lobe region also implicated in self-awareness.

Most of these regions belong to what neuroscientists call the “default mode network,” a network in the brain that tends to be active when a person is awake but unfocused—daydreaming or introspecting or otherwise not engaged in outward, goal-directed activity. (For reasons that aren’t yet fully understood, the default mode network is affected early in Alzheimer’s dementia.)

Bassett and her colleagues suggest that these central, highly interconnected regions are adapted to embody the resting, or baseline, state of the brain—from which it can reach many other commonplace states of activity.

The researchers also used their model to predict the regions that are well suited for pushing the brain into harder-to-reach states. These turned out to be more lightly connected regions including the postcentral gyrus, the supramarginal gyrus, the inferior parietal lobule, the pars orbitalis, and the medial orbitofrontal and rostral  middle frontal cortices. Many of the latter regions are known to be involved in executive functions—also called “cognitive control” functions—by which the brain can adapt to rapidly changing situations, maintain long-term goals, and inhibit behavior that would get in the way of its reaching those goals.

Prior studies of cognitive control have often focused on the competitive dynamics of distinct prefrontal regions. By contrast, this analysis hints that cognitive control emerges from the more complex coordination of a much broader set of regions. “People have previously thought about cognitive control as being specific to the frontal cortex, and we’re saying actually it’s more about the dynamics of a distributed network,” Bassett said.

The brain is so complex that it doesn’t always offer an intuitive view of its workings. But Bassett, based on this study, has likened its network dynamics to the dynamics of a flock of birds: Those in the crowded center of the flock can cause only small shifts in the flock’s course, whereas those on the outside can trigger much larger changes in direction.

Another way of looking at is with a flow metaphor: The more connections run from a given region, the more its influence is dissipated. “If you stimulate a particular area of the brain that’s highly connected, you really can’t move the dynamics of the system that much, because that information gets diffused broadly across the network,” Bassett said. “It just sort of changes the average activity rate, but it doesn’t really change the pattern.”

By this logic, less densely connected regions would be able to exert more concentrated influence.

A better way to understand the connectome

“I think [that study] is a very nice example of the application of network thinking to understanding control [in the brain]—really from an engineering perspective thinking of control as a process by which you exert influence on a system and move it from a given ground state to a desired state,” said Sporns.

The illumination of a hub of central, highly interconnected regions also accords with findings in studies of other animals’ brains. “There seems to be a select group of those [regions] in every nervous system that we and others have looked at over the years, in organisms as simple as C. elegans all the way up to humans,” Sporns said.

Bassett and her colleagues now have a software-based model of brain dynamics that they are using to predict the effects of TMS when delivered to distinct brain regions. They’re also using actual TMS on people to validate the model. “We’re stimulating areas of the brain that are predicted to be a certain type of controller [of the broader network], and then asking whether or not the effects on brain dynamics are consistent with these theoretical predictions,” Bassett said.

One initial goal is to find better ways to employ TMS to hasten recovery in stroke patients. “We’re particularly interested in seeing whether we can stimulate certain areas of the brain to enhance recovery in aphasia [the loss of language skills],” she said.

The new network-based approach could end up making an even greater impact on basic neuroscience, which in recent years has been generating thousands of terabytes of structural and connection data from imaging studies. “We really have no other way to get our arms around those enormous datasets than by applying network thinking and network methodologies,” said Sporns. “We need to reduce those enormous numbers of bytes down to something we can interpret, and network neuroscience is a key framework for doing this.”

In this sense, he adds, neuroscience is merely catching up to other technical fields that have long taken a network-based approach: “It’s already very much in use in other fields and disciplines—in social science, in information technology, for example. Every time you do a Google search, you’re using a [network theory] algorithm to find what you’re looking for within the vast hyperlink architecture of the web. So it’s already surrounding us, it’s pervasive. And I think it will become pervasive in neuroscience as well, in part because it’s driven by the availability of these connection data and the need to understand them.”