A recent toast to James Watson highlights a tolerance for bigotry many want excised from the scientific community.
Automating Drosophila behavior screens gives researchers a break from tedious observation, and enables higher-throughput, more-quantitative experiments than ever before.
January 1, 2015|
DE BIVORT LABThe dynamics inside any fruit fly room are as riveting as a reality TV show. Some Drosophila strains are bullies, while others are just out to mate; some spend more time chowing down; and some are more dedicated to grooming themselves. A decade ago, studying these complex behavioral dynamics was a tedious task, requiring hours spent watching fuzzy videos of flies being flies, jotting down their every action and the time it occurred.
“The problem was that, not only was this prohibitively time-consuming and mindless, but the behaviors were fairly subjective and people would categorize them differently,” says biologist Benjamin de Bivort of Harvard University.
Now, that’s all changing. With the plummeting cost and rising quality of high-definition cameras, sensors, and machine-learning programs, biologists are using computers or touchpads to automate the detection of fly behaviors, from grooming to mating—even detecting how often they eat. Today, such methods are so sensitive that they can reveal the individual motion of each of the six legs of a Drosophila.
“There’s a revolution happening in behavioral neuroscience that comes about because of all these cheap sensors designed for phones and personal electronics,” says de Bivort.
These new techniques are giving scientists the tools to integrate quantitative behavioral data into studies of neuroscience, aging, and even metabolism—zeroing in on the neurons responsible for different fly behaviors, for instance, and how neurodegeneration or obesity changes those neurons’ activity. “This isn’t just a tool to make experiments go faster,” says David Anderson, a neurobiologist at Caltech. “We’re trying to take a field that’s been defined by people sitting in the jungle with a notebook and make it objective and quantitative.”
WALK LIKE A FLY, GROOM LIKE A FLY
Some of the first automated screens of fly behavior used computer programs to track the movement of a fruit fly tethered atop a rotating ball. This treadmill contraption kept the animal in a camera’s field of view, and the program could detect and quantify the turning of the ball and the fly’s motion. But if the fly was standing still, these programs couldn’t discern what it was doing—was it flapping its wings? Rubbing its eyes? Cleaning its abdomen? All these unique motions of the animal’s legs occur while a fly is stationary.
At Harvard, de Bivort studies individual differences in behavior. To discern the distinctive behavioral quirks displayed by flies from the same genetic strain and raised in the same environments, de Bivort wanted to track more detailed information about a fly’s movements than just its walking patterns, so his group has modernized the rotating-ball experiment. They first attached dots of infrared-fluorescent dye to each of a fly’s legs. Then, they tethered the fly above a transparent ball. Below the ball, lasers aimed upward to excite the dots, and on each side they set up infrared sensors—scrounged from a type of computer mouse used to play high-performance video games—to detect the movements of the dots.
The team developed a computer program that could track the motion of these spots and taught the program to link the motion with movement of the fly’s legs. When the animal is rubbing its eyes clean, for instance, it moves both of its front legs. When it’s cleaning its abdomen, however, the dots that move will be associated with other combinations of legs (Nat Commun, 4:1910, 2013).
What you can learn: Although the technique is still limited to one fly at a time, the so-called LegTracker lets you collect information about fly behavior in much more detail than an individual observer can, and even detect patterns that you didn’t set out to study. “The computer can pick up very subtle things, like postural adjustments, that a person wouldn’t even notice,” explains de Bivort. Already, his team has published results about idiosyncrasies in fly patterns of walking and grooming. In general, he says, most flies spend the same total percentage of time doing certain activities, but how they transition between activities varies: one fly might always go from cleaning its eyes to grooming its abdomen, while another always follows eye cleaning with some walking around.
What it takes: The physical setup is relatively easy to replicate, de Bivort says, although it took trial and error, and lots of jury-rigging, to make it work. “The biggest challenge with the LegTracker is on the human side,” says de Bivort. “Getting the dexterity to mark the flies’ legs takes a while.” For labs that have already been using a ball to follow flies’ walking patterns, adding the fluorescent dots and infrared to track each leg can be an easy modification. To set up the LegTracker from scratch would cost about $5,000, de Bivort estimates, with the most expensive components being the laser used to fluoresce the dots and the two cameras that record experiments.
© 2014 LIM ET AL.For some researchers, tethering a fly in place too strongly constrains behaviors of interest. At Caltech, Anderson studies the neurological basis of social behaviors, and he needs to be able to quantify how a fly acts while it interacts with other flies and its environment. What causes one fly to lunge aggressively at another? What makes a fly pace in a circle around food?
Anderson, in collaboration with Caltech machine-vision expert Pietro Perona, developed an arena that is lit from below and sports a video camera to tape the interaction between flies. The technique’s real punch comes from the computer program, designed by Perona, which can be taught to recognize and quickly quantify any particular event caught on tape—how often the flies get within a certain distance of each other or how often they lunge, for instance.
“It allows us to take tens of hours of videotape, which would take a human being a week of nine-to-five work to analyze, and have the computer process it in five minutes,” says Anderson. “It’s been absolutely transformative for us.”
What you can learn: Anderson’s group recently used the setup to study how two male flies compete over food, showing that the presence of food increases the frequency of aggression between males (PLOS ONE, 9:e105626, 2014). Moreover, by using genetic mutants in the experiments, they found that sugar-sensing neurons are required for this food-mediated aggression. The tracking program can be used to study a broad variety of behaviors, including mating.
What it takes: The setup is much the same as what’s been used for decades by human observers. But you’ll need to design a program that’s customized to your particular arena, and “teaching” the program to recognize a specific movement can be tricky. After being fed videotape with each frame annotated—this is a fly, this is not, this is a lunging fly, this is not—the program can start to analyze fresh data. Anderson recommends that biologists collaborate with computer scientists to develop such a program for their own use. Because no two arenas look the same, a program has to be customized and retrained to work for a particular arena and lighting setup. Anderson’s setup is unique to his lab, and he says it costs around $8,000 to build—not including cameras—but some researchers are working toward programs that will be easier to customize. The Janelia Automatic Animal Behavior Annotator (JAABA), for instance—developed by scientists at Howard Hughes Medical Institute’s Janelia Farm campus—is a free, open-source machine-learning program that can be used to track some behaviors.
WATCH WHAT THEY EAT
Carlos Ribeiro, a scientist at the Champalimaud Neuroscience Programme in Lisbon, Portugal, studies how flies translate an internal state—such as nutrient depletion—to a behavior, like eating, so he wanted to quantify how often and how much the animals eat. This information isn’t just key to understanding which neurons influence feeding behavior and drive hunger, but also what causes metabolic disease, overeating, and obesity. Ribeiro turned to a touchpad like those found in tablet computers, which these days are so sensitive that when a fly walking on the touchpad sticks its proboscis into a tray of food, the pad detects it. “We wanted something which is robust and can go high-throughput and which doesn’t require expensive cameras,” says Ribeiro. “These sensors are actually very cheap now that they’re in all these consumer devices.”
With their current setup, Ribeiro’s group can connect up to 32 different feeding arenas, each monitoring one fly’s eating patterns, to a single computer. By testing flies with genetic mutations or induced metabolic diseases, they hope to uncover factors that drive eating behavior.
What you can learn: Ribeiro’s team fluorescently labeled fly food to measure precisely how much the insects eat with each extension of the proboscis. The researchers used this information to calibrate their flyPAD to quantify how often, how long, and how much a fly eats. They also used the technique to study flies’ food preferences by quantifying the different food choices made in one sitting (Nat Commun, 5:4560, 2014). And the flyPAD’s utility extends beyond feeding, Ribeiro says. It could also be adapted to other behaviors, such as egg laying, that require small interactions between an animal and the ground below it. “These kinds of behaviors are very hard to capture quantitatively with a camera,” he says.
What it takes: Ribeiro’s group is collaborating with other scientists interested in using the flyPAD. Ribeiro will ship the setup to a lab for one-time use, or help you build your own—which will cost around $1,200. Building your own requires devices that most biology labs don’t have lying around—special electrodes and capacitance converters—but might be worth the investment if you’re studying a behavior that can be tracked using the touchpad.
PAVEL ITSKOV, RIBEIRO LABOne of the trickiest behaviors to study in flies is flight. It adds an extra dimension to the animals’ movement and requires following the bending and flapping of hard-to-see wings. Barry Ganetzky, a geneticist at the University of Wisconsin–Madison, studies how the fly neuromuscular system develops and how it weakens with age or disease. He needed a quick and easy way to screen lots of flies, of different ages or with different genetic mutations, for their ability to fly. The system he and postdoc Daniel Babcock developed, called the Flight Tester, is simple: drop the flies from above into a graduated cylinder after covering its inner wall with sticky flypaper. The better a fly is at flying, the sooner after it’s dropped it will start to head sideways—straight into the flypaper. Animals that have totally lost their ability to fly will fall much further down before getting caught (possibly even hitting the bottom). The researchers can then remove the flypaper, snap a photo of it, and use photo analysis software to quantify the density of flies at different heights along the paper. (Ganetzky and Babcock used ImageJ, a free program developed by the NIH.) “Even if you have two or three hundred flies, we can make a graphic demonstration of where they’re distributed,” says Babcock.
What you can learn: In a February 2014 Journal of Visualized Experiments paper, Ganetzky and Babcock reported the use of the Flight Tester to compare wild-type flies with a strain dubbed slowpoke. While the healthy flies landed, on average, at a height of 73 cm in the meter-high graduated cylinder, the slowpoke flies fell much farther, hitting the sticky side around 44 cm. Since then, the lab has used the approach to characterize nine other mutants and has now started analyzing the genetics of these flies.
What it takes: Setting up the Flight Tester is cheap—around $250, Ganetzky estimates—and easy, requiring little more than a graduated cylinder with a funnel on top, sticky flypaper, a camera, and a piece of computer software that’s already commonly used in many biology labs. “This system is as simple as possible and has worked really well for everything we want to do right now,” he says.