The perfect (test) beast

A kind of reverse neural implant.

YOUR AVERAGE neuroscientist's worldview is relentlessly mechanistic—from love to God to Pokemon, it's the neurons, babe. But even mechanists get the mind-body duality blues. "Mind" may be nothing more than our romantic description of the functions of the physical brain, but that brain is part of a body, and the way it functions is fundamentally interconnected with the rest of that body. Which is why research is a bitch when you've got nothing more to study than a bunch of cells in a dish.

It's a problem that plagues researchers: Put simply, you can study neurons in living subjects for only a short time before the cells die. Nearly all current research and monitoring techniques are invasive. The few that aren't, such as magnetic resonance imaging, can take only brief, blurred snapshots of neuronal activity. Worse, as Caltech scientist Dr. Steve Potter puts it, "the cells are stacked on top of each other inside a skull." (And if you find that description disheartening, don't get him started on the nature of consciousness.)

Currently scientists can only study clusters of living, growing neurons in great detail over time by observing their activities in a dish. In fact, one of the classic dues-paying chores of neuroscience graduate students is spending the night in the lab watching for signs of mold on cultured rat neurons. Potter, who's logged plenty of time looking for mold, has an idea for bridging the gap: Make your own virtual animal in the lab.

His work is an extension of the concept of "animats"—artificial animals, either software simulations or actual robots. What distinguishes animats from more abstract entities such as neural nets is that animats have virtual bodies and environments, so that physical concepts like motor control can be studied.

Still, animats aren't ideal research subjects. Without the context of the body there's no way to study how, for example, we develop procedural memory. Procedural memory is what dancers call "muscular memory"—the ability to do physical tasks, like riding bicycles or performing pirouettes, without consciously thinking through the steps involved. You can watch a live rat learn a maze, or you can watch cells respond to electrical stimuli, but you can't watch a rat's neurons grow and change as it learns the maze.

What distinguishes Potter's animats from everybody else's is that his have an actual organic component. Potter combines software animats with multi-electrode arrays (MEAs)—dishes lined with electrodes. Neurons are then grown on top of the electrodes. Unlike earlier methods that often damaged cells, this technique allows cells to flourish for weeks.

Potter connects the neuron-lined MEA to a computer programmed to simulate the behavior of an animal body. The body exists in software, but the brain external to the body is wetware—or at least a cybernetic hybrid of living neurons and electrodes. As Potter puts it, "it's an input device that happens to have brain cells growing on it." The pun is unavoidable: He's created the first true computer mouse.

"We will create something that behaves and learns. This has never been done by anybody who studied cells in culture," says Potter. "They've never been able to say 'These cells in a dish here are learning.' All they've been able to say is, 'I've probed them and prodded them and electrocuted them, and look, I've made a change.'"

THE ANIMAT "LIVES" in Potter's lab. To a layperson, it looks rather Rube Goldbergian—electrodes stick out from a petri dish like multicolored plastic quills, their sensations directed and recorded by an adjacent computer and viewed through a two-photon microscope. The result: a real-time visualization of neurons firing and growing. The images are striking when you realize you're looking at a kind of cybernetic video game: The ponglike movements (three-dimensional graphs) are being produced by the stimulation, or simple random firing, of living neurons. As you watch, the patterns increase in complexity as the cells themselves begin to grow and branch.

"[Neurons] get bigger and branchier as they get older," Potter explains. "That process happens in culture for about a month, and so the animat must gain more intelligence since its capacity to process information has something to do with how many connections it has to other cells. When we stimulate the animat, we can literally watch those connections being formed." The visualization tools are already in place: Dr. Tom DeMarse, an expert in animal learning, is now working out the basics of how best to stimulate the neurons, while graduate student Daniel Wagenaar is analyzing the neural data for patterns significant enough to be recognized as behaviors.

DeMarse explains that what they are looking for is "association among stimuli," that is, trying to see how the subject comes to understand that stimulus X is associated with result Y. It doesn't necessarily require a Pavlovian reward/punishment system since the object is to see what goes on when the brain thinks "a big rock is located to the right of a tree and behind that tree is a stream"—a series of associations, one of which happens to be an obstacle that the physical body must react to.

Potter cautions that his animat must be understood as a model brain. He explains, "Most scientists eventually deal with 'models'—some version of the interesting thing, with a lot thrown out to help simplify matters. It's a hard job deciding what to keep in and what to throw out of the model. I hope this system will help us answer what components are the minimum necessary set of neural circuits to study behavior and learning."

Potter hopes that studying "physical" stimulus to the virtual bodies of his animats will lead to conclusions about procedural memory. He sums up his team's work as "learning the basic language of how neurons talk to each other, using dishes that learn." Beyond that, he hopes to have created a general tool for all neuroscientists. His animats won't cure human ills, but they may become essential research instruments for the scientists who do. At the very least, through a sealed-culture system, Potter has devised a way to get fewer moldy neurons.

Of course, Caltech isn't terribly far from Hollywood. Brains in dishes controlling computer-simulated bodies? He may also just have created the plot for the next Keanu Reeves movie.

For more information on the work of Drs. Potter and DeMarse, see www.caltech.edu/~pinelab/PotterGroup.htm

 
comments powered by Disqus