One of the primary missions of neuroscience is to make connections between particular neurons in the brain and specific behaviors. Now a team of researchers has used computer-vision and machine-learning techniques in fruit flies to create behavior anatomy maps that will help us understand how specific brain circuits generate Drosophila aggression, wing extension, or grooming. The data are being published July 13 in the journal Cell as a resource for other investigators.
"The ultimate goal of this work is to assign behaviors to neurons in the fly brain," says senior author Kristin Branson, a computational biologist at the Howard Hughes Medical Institute Janelia Research Campus. "Past studies have used functional MRI to connect certain behaviors to regions of the brain, but we're doing this at a cellular level with much higher resolution. In addition, our research uses freely moving flies, enabling us to catalog a broader set of behaviors–both movements and social interactions. We tried to quantify everything we could see the flies doing."
To collect the information, the flies were placed 20 at a time into a walking arena–actually a dish about five inches in diameter. The ceiling of the dish was low, only about two fly heights, preventing the flies from walking over the top of each other, which would make their actions harder to monitor. Each session lasted 15 minutes, with eight cameras trained on the flies at the same time. The activities measured were both locomotion behaviors, like walking, turning, and jumping, and social and courtship behaviors, like chasing and wing extension.
"One of the advances that enabled this research was tools developed at Janelia to genetically target specific populations of neurons," Branson explains. "These different lines of flies allow us to manipulate the activity of specific neurons and measure their effects on behavior." For example, one set of neural changes may cause flies to jump. Another set may induce male flies to court females more frequently. In total, the researchers studied about 400,000 flies from about 2,200 different genotypes over a period of 18 months.
Another essential element was computer-vision and machine-learning analysis required to conduct this research in an automated manner. To train the machine-learning classifier to detect a behavior, the movements were manually labeled in a small number of sample frames. Eventually, the classifiers could be applied to new videos, using the computer to categorize and define the flies' movements in a high-throughput way. In total, the behavior dataset is derived from about half a petabyte of video. "It was a huge undertaking to put these large datasets together and get scientifically meaningful insight," Branson says.
"We believe these databases will be useful in two ways," notes Alice Robie, a research scientist in Branson's lab and the study's first author. "Fly neurobiologists will be able to look at a region of the brain that they're interested in and find all the behaviors that are correlated with it. On the other side, researchers who study behavior will be able to link those activities with regions of the brain, which then can be further studied."
The researchers add that the mechanisms that underlie how fly neurons operate and communicate are likely to be the same mechanisms that operate in the human brain. "This is the first step to studying these circuits, so that you can move toward an understanding of how all brains generate behavior," Robie says.
Cell, Robie et al.: "Mapping the Neural Substrates of Behavior" http://www.cell.com/cell/fulltext/S0092-8674(17)30716-X
Cell (@CellCellPress), the flagship journal of Cell Press, is a bimonthly journal that publishes findings of unusual significance in any area of experimental biology, including but not limited to cell biology, molecular biology, neuroscience, immunology, virology and microbiology, cancer, human genetics, systems biology, signaling, and disease mechanisms and therapeutics. Visit http://www.cell.com/cell. To receive Cell Press media alerts, contact [email protected]
Related Journal Article