New research suggests that training to do a new task causes groups of brain cells to “learn” how to work together more efficiently.
While practice may not always make perfect, most tasks usually become easier with repetition. New hints about why this might be the case are emerging from experiments in which Howard Hughes Medical Institute scientists tracked neuronal activity in the brains of mice learning to perform new tasks.
The researchers found that as mice learned a simple odor discrimination task, the activity of the brain cells involved in specific aspects of the task became tightly coordinated. The researchers propose that training to do a new task causes groups of brain cells to “learn” how to work together more efficiently.
Karel Svoboda, group leader at HHMI’s Janelia Farm Research Campus in Ashburn, Virginia, and his colleagues reported on these experiments in the April 8, 2010, issue of Nature.
“Learning is thought to be accompanied by changes in the connections between neurons, but we know little about the process at the level of populations of neurons,” Svoboda says. He explained that as connections between neurons become stronger, those neurons become more likely to fire together. So to map the strength of neuronal connections, his team designed an experiment in which they could track correlations in activity within networks of neurons in the brains of living mice.
The team, led by first author and postdoctoral researcher Takaki Komiyama, used a reward of water to train mice to discriminate between two different odors. The mice were taught to lick in response to one odor, but not in response to the other. Svoboda says the mice learned the task incredibly quickly—within an hour or so.
As each mouse went through this training, the researchers watched through a small imaging window in the animal’s skull as bursts of activity indicated that movement-controlling neurons were being turned on. Using two-photon calcium imaging—a technique that Svoboda helped pioneer—the researchers could observe neural activity as it occurred. The technique takes advantage of the fact that as a neuron fires, its internal concentration of calcium rises sharply. When a calcium-sensitive fluorescent dye is infused into the brain, an increase in neuronal activity causes an increase in fluorescence. Svoboda and his colleagues used a two-photon microscope, which can see deep inside the brain, to record these activity-related changes in fluorescence within a small network of several dozen neurons.
To image the neurons that controlled the mouse’s licking behavior, Komiyama, Svoboda, and their colleagues first had to find the relevant neural networks. Somewhat surprisingly, there turned out to be two groups of lick-controlling neurons in distinct areas of the mouse motor cortex. “We speculate that one of these groups might be controlling the coordinated jaw and mouth movements, while the other group might be directly controlling the tongue muscles,” Komiyama explained.
The researchers could now record the activity of these networks of neurons as the mice learned. They found that within each group, some neurons would fire bursts of signals near the beginning of the action, some nearer the middle, and some nearer the end; but for any two neurons that normally fired around the same time, their correlation became tighter and tighter as the number of learning trials increased. The same was true of the neurons normally active during a correct no-licking inaction. In each case the increasing correlation suggested the local emergence of a more tightly connected, functionally precise network of neurons.
“We find that as the mouse learns, neurons segregate into two populations: those that show elevated activity when the mouse licks, and those that showed elevated activity when the mouse refrains from licking. With learning, the correlations within each of the neuronal populations—but not between them—increased dramatically,” Svoboda said. “The changes in these motor-related circuits could explain how neural circuits drive robust motor behavior.”
Remarkably, as learning progressed, the number of task-related neurons in these subnetworks seemed to shrink, hinting at a streamlining or simplification of the subnetworks’ activity. “We know that as we train to perform a behavior, it tends to become easier,” Komiyama said. “Using fewer neurons more efficiently might be one way to explain that phenomenon.”