Learning takes brain acrobatics
When neural areas more easily switch communication partners, learning improves
Peer inside the brain of someone learning. You might be lucky enough to spy a synapse pop into existence. That physical bridge between two nerve cells seals new knowledge into the brain. As new information arrives, synapses form and strengthen, while others weaken, making way for new connections.
You might see more subtle changes, too, like fluctuations in the levels of signaling molecules, or even slight boosts in nerve cell activity. Over the last few decades, scientists have zoomed in on these microscopic changes that happen as the brain learns. And while that detailed scrutiny has revealed a lot about the synapses that wire our brains, it isn’t enough. Neuroscientists still lack a complete picture of how the brain learns.
They may have been looking too closely. When it comes to the neuroscience of learning, zeroing in on synapse action misses the forest for the trees.
A new, zoomed-out approach attempts to make sense of the large-scale changes that enable learning. By studying the shifting interactions between many different brain regions over time, scientists are beginning to grasp how the brain takes in new information and holds onto it.
These kinds of studies rely on powerful math. Brain scientists are co-opting approaches developed in other network-based sciences, borrowing tools that reveal in precise, numerical terms the shape and function of the neural pathways that shift as human brains learn.
“When you’re learning, it doesn’t just require a change in activity in a single region,” says Danielle Bassett, a network neuroscientist at the University of Pennsylvania. “It really requires many different regions to be involved.” Her holistic approach asks, “what’s actually happening in your brain while you’re learning?” Bassett is charging ahead to both define this new field of “network neuroscience” and push its boundaries.
“This line of work is very promising,” says neuroscientist Olaf Sporns of Indiana University Bloomington. Bassett’s research, he says, has great potential to bridge gaps between brain-imaging studies and scientists’ understanding of how learning happens. “I think she’s very much on the right track.”
As a person learns, connections between brain regions can change. Some neural partners connect, then split apart (as in the blue and yellow groups); others form new partnerships (as in the red and yellow groups). A measure of this swapping, called flexibility, may relate to how well a person learns, new research suggests.
Already, Bassett and others have found tantalizing hints that the brains that learn best have networks that are flexible, able to rejigger connections on the fly to allow new knowledge in. Some brain regions always communicate with the same neural partners, rarely switching to others. But brain regions that exhibit the most flexibility quickly swap who they’re talking with, like a parent who sends a birthday party invite to the preschool e-mail list, then moments later, shoots off a work memo to colleagues.
In a few studies, researchers have witnessed this flexibility in action, watching networks reconfigure as people learn something while inside a brain scanner. Network flexibility may help several types of learning, though too much flexibility may be linked to disorders such as schizophrenia, studies suggest.
Not surprisingly, some researchers are rushing to apply this new information, testing ways to boost brain flexibility for those of us who may be too rigid in our neural connections.
“These are pretty new ideas,” says cognitive neuroscientist Raphael Gerraty of Columbia University. The mathematical and computational tools required for this type of research didn’t exist until recently, he says. So people just weren’t thinking about learning from a large-scale network perspective. “In some ways, it was a pretty boring mathematical, computational roadblock,” Gerraty says. But now the road is clear, opening “this conceptual avenue … that people can now explore.”
It takes a neural village
That conceptual avenue is more of a map, made of countless neural roads. Even when a person learns something very simple, large swaths of the brain jump in to help. Learning an easy sequence of movements, like tapping out a brief tune on a keyboard, prompts activity in the part of the brain that directs finger movements. The action also calls in brain areas involved in vision, decision making, memory and planning. And finger taps are a pretty basic type of learning. In many situations, learning calls up even more brain areas, integrating information from multiple sources, Gerraty says.
He and colleagues caught glimpses of some of these interactions by scanning the brains of people who had learned associations between two faces. Only one of the faces was then paired with a reward. In later experiments, the researchers tested whether people could figure out that the halo of good fortune associated with the one face also extended to the face it had been partnered with earlier. This process, called “transfer of learning,” is something that people do all the time in daily life, such as when you’re wary of the salad at a restaurant that recently served tainted cheese.
Study participants who were good at applying knowledge about one thing — in this case, a face — to a separate thing showed particular brain signatures, Gerraty and colleagues reported in 2014 in the Journal of Neuroscience. Connections between the hippocampus, a brain structure important for memory, and the ventromedial prefrontal cortex, involved in self-control and decision making, were weaker in good learners than in people who struggled to learn. The scans, performed several days after the learning task, revealed inherent differences between brains, the researchers say. The experiment also turned up other neural network differences among these regions and larger-scale networks that span the brain.
Children who have difficulty learning math, when scanned, also show unexpected brain connectivity, according to research by neuroscientist Vinod Menon of Stanford University and colleagues. Compared with kids without disabilities, children with developmental dyscalculia who were scanned while doing math problems had more connections, particularly among regions involved in solving math problems. That overconnectivity, described in 2015 in Developmental Science, was a surprise, Menon says, since earlier work had suggested that these math-related networks were too weak. But it may be that too many links create a system that can’t accommodate new information. “The idea is that if you have a hyperconnected system, it’s not going to be as responsive,” he says.
Flex when needed
As people learned a new task — pairing a butterfly with a particular flower — brain flexibility changed in a recent study. Flexibility across the whole brain rose then dropped as training continued. Reasons for the decline are unclear, but information flow in the brain may change as people master a task. Each colored line represents one person’s flexibility score over time (the black line is an average).
There’s a balance to be struck, Menon says. Neural pathways that are too weak can’t carry necessary information, and pathways that are too connected won’t allow new information to move in. But the problem isn’t as simple as that. “It’s not that everything is changing everywhere,” he says. “There is a specificity to it.” Some connections are more important than others, depending on the task.
Neural networks need to shuttle information around quickly and fluidly. To really get a sense of this movement as opposed to snapshots frozen in time, scientists need to watch the brain as it learns. “The next stage is to figure out how the networks actually shift,” Menon says. “That’s where the studies from Dani Bassett and others will be very useful.”
Flexing in real time
Bassett and colleagues have captured these changing networks as people learn. Volunteers were given simple sequences to tap out on a keyboard while undergoing a functional MRI scan. During six weeks of scanning as people learned the task, neural networks in their brains shifted around. Some connections grew stronger and some grew weaker, Bassett and her team reported in Nature Neuroscience in 2015.
People who quickly learned to tap the correct sequence of keys showed an interesting neural trait: As they learned, they shed certain connections between their frontal cortex, the outermost layer of the brain toward the front of the head, and the cingulate, which sits toward the middle of the brain. This connection has been implicated in directing attention, setting goals and making plans, skills that may be important for the early stages of learning but not for later stages, Bassett and colleagues suspect. Compared with slow learners, fast learners were more likely to have shunted these connections, a process that may have made their brains more efficient.
Flexibility seems to be important for other kinds of learning too. Reinforcement learning, in which right answers get a thumbs up and wrong answers are called out, also taps into brain flexibility, Gerraty, Bassett and others reported online May 30 at bioRxiv.org. This network comprises many points on the cortex, the brain’s outer layer, and a deeper structure known as the striatum. Other work on language comprehension, published by Bassett and colleagues last year in Cerebral Cortex, found some brain regions that were able to quickly form and break connections.
These studies captured brains in the process of learning, revealing “a much more interesting network structure than what we previously thought when we were only looking at static snapshots,” Gerraty says. The learning brain is incredibly dynamic, he says, with modules breaking off from partners and finding new ones.
While the details of those dynamics differ from study to study, there is an underlying commonality: “It seems that part of learning about the world is having parts of your brain become more flexible, and more able to communicate with different areas,” Gerraty says. In other words, the act of learning takes flexibility.
But too much of a good thing may be bad. While performing a recall task in a scanner, people with schizophrenia had higher flexibility among neural networks across the brain than did healthy people, Bassett and colleagues reported last year in the Proceedings of the National Academy of Sciences. “That suggests to me that while flexibility is good for healthy people, there is perhaps such a thing as too much flexibility,” Bassett says.
Too much of a good thing
Compared with healthy people (left column), people with schizophrenia (right column) and their close relatives (middle) showed signs of more flexibility across their brains. The greater the flexibility, the larger the spheres and redder the color.
Just how this flexibility arises, and what controls it, is unknown. Andrea Stocco, a cognitive neuroscientist at the University of Washington in Seattle, suspects that a group of brain structures called the basal ganglia, deep within the brain, has an important role in controlling flexibility. He compares this region, which includes the striatum, to an air traffic controller who shunts information to where it’s most needed. One of the basal ganglia’s jobs seems to be shutting things down. “Most of the time, the basal ganglia is blocking something,” he says. Other researchers have found evidence that crucial “hubs” in the cortex help control flexibility.
Push for more
Researchers don’t yet know how measures of flexibility in brain regions relate to the microscopic changes that accompany learning. For now, the macro and the micro views of learning are separate worlds. Despite that missing middle ground, researchers are charging ahead, looking for signs that neural flexibility might offer a way to boost learning aptitude.
It’s possible that external brain stimulation may enhance flexibility. After receiving brain stimulation carefully aimed at a known memory circuit, people were better able to recall lists of words, scientists reported May 8 in Current Biology. If stimulation can boost memory, some argue, the technique could enhance flexibility and perhaps learning too.
When you’re happy and you know it
A team tracked one researcher’s mood and brain flexibility for a year. When he was happiest, his brain was most flexible. Regions in which high flexibility was most closely linked with positive mood are red (top). When he was surprised, flexibility was low overall. Regions with the strongest link between surprise and low flexibility are dark blue (bottom).
Certain drugs show promise. DXM, found in some cough medicines, blocks proteins that help regulate nerve cell chatter. Compared with a placebo, the compound made some brain regions more flexible and able to rapidly switch partners in healthy people, Bassett and colleagues reported last year in the Proceedings of the National Academy of Sciences. She is also studying whether neurofeedback — a process in which people try to change their brain patterns to become more flexible with real-time monitoring — can help.
Something even simpler might work for boosting flexibility. On March 31 in Scientific Reports, Bassett and colleagues described their network analyses of an unusual subject. For a project called MyConnectome, neuroscientist Russ Poldrack, then at the University of Texas at Austin, had three brain scans a week for a year while assiduously tracking measures that included mood. Bassett and her team applied their mathematical tools to Poldrack’s data to get measurements of his neural flexibility on any given scan day. The team then looked for associations with mood. The standout result: When Poldrack was happiest, his brain was most flexible, for reasons that aren’t yet clear. (Flexibility was lowest when he was surprised.)
Those results are from a single person, so it’s unknown how well they would generalize to others. What’s more, the study identifies only a link, not that happiness causes more flexibility or vice versa. But the idea is intriguing, if not obvious, Bassett says. “Of course, no teacher is really going to say we’re doing rocket science if we tell them we should make the kids happier and then they’ll learn better.” But finding out exactly how happiness relates to learning is important, she says.
The research is just getting started. But already, insights on learning are coming quickly from the small group of researchers viewing the brain as a matrix of nodes and links that deftly shift, swap and rearrange themselves. Zoomed out, network science brings to the brain “a whole new set of hypotheses and new ways of testing them,” Bassett says.
This story appears in the September 16, 2017, issue of Science News with the headline, “Flex time: The brain’s ability to shift connections might ease learning.”
D. S. Bassett and M. G. Mattar. A network neuroscience of human learning: Potential to inform quantitative theories of brain and behavior. Trends in Cognitive Sciences. April 2017. doi: 10.1016/j.tics.2017.01.010
L. R. Chai et al. Functional network dynamics of the language system. Cerebral Cortex. November 2016, p. 4148. doi: 10.1093/cercor/bhw238
D. S. Bassett et al. Dynamic reconfiguration of human brain networks during learning. Proceedings of the National Academy of Sciences. Vol. 188, May 3, 2011. doi: 10.1073/pnas.1018985108
D. S. Bassett et al. Learning-induced autonomy of sensorimotor systems. Nature Neuroscience. Vol. 18, May, 2015. doi:10.1038/nn.3993
R. F. Betzel et al. Positive affect, surprise, and fatigue are correlates of network flexibility. Scientific Reports. Published online March 31, 2017. doi: 10.1038/s41598-017-00425-z
U. Braun et al. Dynamic brain network reconfiguration as a potential schizophrenia genetic risk mechanism modulated by NMDA receptor function. Proceedings of the National Academy of Sciences. Vol. 113, November 1, 2016. doi: 10.1073/pnas.1608819113
Y. Ezzyat et al. Direct brain stimulation modulates encoding states and memory performance in humans. Current Biology. Vol. 27, May 8, 2017, p. 1251. doi: 10.1016/j.cub.2017.03.028
R. T. Gerraty et al. Transfer of learning relates to intrinsic connectivity between hippocampus, ventral prefrontal cortex and large-scale networks. The Journal of Neuroscience. Vol. 34, August 20, 2014, p. 11297. doi: 10.1523/JNEUROSCI.0185-14.2014
R. T. Gerraty et al. Dynamic flexibility in striatal-cortical circuits supports reinforcement learning. http://www.BioRXiv.org, posted May 30, 2017. doi: 10.1101/094383
M. Rosenberg-Lee et al. Brain hyper-connectivity and operation-specific deficits during arithmetic problem solving in children with developmental dyscalculia. Developmental Science. Vol. 18, May, 2015. doi: 10.1111/desc.12216