Janelia Farm Research Campus
Dr. Leonardo is a lab head at the Janelia Farm Research Campus.
Anthony Leonardo is interested in general principles underlying neural information processing. His lab investigates the computations neural circuits implement, how robust they are, and how different circuits are linked into systems to produce behaviors.
As a lab head at the Janelia Farm Research Campus, Anthony Leonardo studies salamanders and dragonflies, efficient killers that are far more likely to be found roaming Janelia's bucolic landscape than inside the laboratory itself. But the salamander and dragonfly are Leonardo's chosen subjects in research that is building a bridge between the macroscopic world of behavior and the microscopic world of neural circuits.
Leonardo is one of the top young investigators in neuroethology, the study of the neural basis of animal behavior. He earned his Ph.D. in computation and neural systems from the California Institute of Technology in 2002 and has long been fascinated by the brain and how animals behave as they do. His undergraduate training in artificial intelligence and computer science at Carnegie Mellon University included experience modeling human cognition. But it was as a graduate student at Caltech, under the guidance of Masakazu Konishi, that Leonardo began to look inside living brains to understand the complexities of neural circuit dynamics.
That creative early research, probing the mechanisms that underlie song generation in the zebra finch, produced some surprising findings. Leonardo tested the prevailing theory that once zebra finches have learned their songs through mimicking a bird serving as a tutor, those songs then become hardwired in the bird's brain. He developed a computer-controlled system to perturb the feedback heard by singing adult finches. By replaying their recorded songs slightly out-of-sync, Leonardo found that the finches' songs deteriorated markedly. After Leonardo withdrew the distorted feedback, the zebra finches gradually recovered their original songs. "We revealed that the songs were stable not because they had become hardwired but because they are maintained dynamically," he says.
Leonardo did the second portion of his doctoral work at Bell Labs, where he collaborated with Michale Fee to help develop a miniature, motorized microdrive—a tiny device implanted onto the bird's head—that could monitor the activity of several individual neurons while the bird sang. The results of experiments done with the microdrive were surprising. "We had assumed that when the bird produced two similar sounds at different times in his song, the same patterns of neurons would fire in the same sequence. Instead, we found that the bird uses entirely different sets of neurons to encode the same sound," Leonardo says. In essence, the bird has played a clever trick—because it sings only a single song, it represents each moment in the song with a different set of neurons. "We hypothesize that this makes learning much faster and more robust because correcting errors in one portion of the song does not affect sounds produced at later times," Leonardo explains. Elegant theoretical work in the laboratory of Sebastian Seung at MIT has tested this idea and shown that the structure of neuronal firing patterns can indeed arise from song-learning rules that minimize learning time.
What relevance can lessons from a bird brain have for understanding the human mind? Leonardo's hope is that, just as single neurons in different species operate according to the same physical principles, the same logic applies to networks of neurons. "We often have the sense that beneath the complexity of the countless connections linking neurons into a circuit, there lies some relatively simple computation," Leonardo says. "My goal is to develop an understanding of those computations. Model systems like the songbird are useful because they have a complex yet stereotyped behavior that is generated by a small number of brain areas. These types of systems present well-defined and tractable problems. We should be able to solve them in their entirety, from the earliest stages of sensory processing to the final moments of motor control."
Enter the salamander, an amphibian that excels at snagging flying insects with its "ballistically launched" tongue. While a postdoctoral fellow in the lab of Markus Meister at Harvard, Leonardo studied how salamanders capture their rapidly moving prey. He investigated a circuit in the retina that might allow the salamander to predict the motion of its prey a few moments into the future. Without such a circuit, the salamander's tongue would hit where its prey was moments ago, rather than its real position. Leonardo showed that the circuit that produces this computation does so only for a small range of target sizes and speeds that are matched to the prey captured by the salamander. So signatures of the salamanders' specialized prey preferences may actually begin in the retina.
At Janelia, Leonardo and postdoctoral researcher Bart Borghuis have demonstrated that the behavior of the salamander is broadly consistent with the underlying retinal position prediction. Just as the retina computes the future position of the target, the salamander launches its tongue toward the position the prey will attain in a few hundred milliseconds. This time compensates for neural delays in the salamander’s brain as well as the time it takes to align its head to the prey’s position. Leonardo and Borghuis are now working to determine whether the computations in the salamander’s retina actually cause its behavior. “To do this,” he says, “we need to show that as the accuracy of position estimation in the retina begins to fail, the way in which the computation breaks down is mirrored exactly in the behavior. If we can do that, then we will begin to believe we have the seeds of a mechanistic description of how salamander prey capture works.”
In addition to salamanders, he has begun a second line of research to study the neural circuitry underlying prey capture in the dragonfly. One project will attempt to implant miniature electrodes into dragonfly target tracking neurons and record from them wirelessly while the dragonflies fly and catch prey. "Of tremendous importance to our work," he says, "is looking at neural circuits in different brain areas and different animals. What we would really like to know is: are there fundamental principles that govern how neurons solve behavioral problems?"