When Johann Strauss II composed The Blue Danube in the 1860s, he could hardly have imagined that scientists in the future would watch, captivated, as a computer-generated brain waltzed to his tune.
Neuroscientist Terry Sejnowski wowed the crowd doing just that during a lecture last year at the Salk Institute for Biological Studies, in La Jolla, California. Sejnowski, an HHMI investigator at the Salk, presented his lab’s model: a segment of a brain floating dreamily around the screen as different parts of its cellular machinery—channels, connectors, and organelles—emerged around a single starting filament.
Sejnowski will use the model not only to entertain, but also—sans musical accompaniment—to study how the brain learns and remembers.
“This is the beginning of a whole new era,” he says. “For the first time, it gives us a three-dimensional, complete reconstruction in which we can, literally, include the impact of every single important molecule.”
Learning and memory depend on nerve cells and support cells called astrocytes. Nerves have “output” channels—long slender axons that reach out to other nerves—and “input” channels—branching dendrites that collect incoming signals. Thought and memory take shape where the axon from one nerve contacts the dendrite of another at a junction called the synapse.
At the synapse, the first nerve’s axon sends out a signal in the form of molecules called neurotransmitters. These chemicals cross the junction and are picked up by the second nerve’s dendrites. Astrocytes then mop up excess neurotransmitter.
The brain’s estimated one quadrillion synapses are the most numerous and complex structures involved in learning and memory, Sejnowski says. How the frequency and strength of transmissions are controlled and communicated remains a mystery. “If you don’t understand that, you’re never going to understand how the brain works.”
Axons and dendrites meet in a region called the neuropil. The walls of the Computational Neurobiology Laboratory, which Sejnowski heads, are adorned with huge black-and-white electron micrographs of cross sections of the neuropil—a dense conglomerate of bubble-like vesicles, wrinkled mitochondrial membranes, and other cellular components. “It’s all jumbled … it’s very hard to understand,” Sejnowski says.
Sejnowski and postdoctoral fellow Justin Kinney are investigating how synapses influence one another. They wondered if neurotransmitters from one synapse could spill over and activate neighboring dendrites as well. The answer, they reasoned, would depend on whether each synapse is surrounded by astrocytes, which would absorb leftover neurotransmitter. Those without astrocytes would leave neurotransmitters to diffuse into nearby synapses.
To answer their question, studying two-dimensional electron micrographs would not do; Kinney and Sejnowski needed a three-dimensional representation of the neuropil. So five years ago, they teamed with Tom Bartol, chief modeler in Sejnowski’s group, to rebuild the three-dimensional neuropil inside a computer.
The waltzing bit of brain is a precise replica of a small cube of real rat brain, five microns to a side. Collaborator Kristen Harris, of the University of Texas at Austin, sliced that cube into 100 sections and imaged them with electron microscopy. Then, the researchers identified and traced every axon, dendrite, and astrocyte in each of those 100 images—a task that took six months for Harris and her longtime collaborator, Josef Spacek of Charles University in Prague, Czech Republic, to complete.
Armed with those 100 slices, Kinney, working with Chandrajit Bajaj, also of the University of Texas at Austin, used computer technology to recreate the three-dimensional neuropil.
The model takes major computing power, Bartol notes, well beyond the video card in your Xbox. “In Hollywood and video games, it just has to look good,” he says. Game programmers cut corners by filling in large areas with simple textures. But a scientific model—in this case, one encompassing 450 synapses, 69 axons, and 77 dendritic spines—must contain fully solid structures with no space for errant neurotransmitters to leak through the membranes. “The movies aren’t cartoons, they’re actual visualizations of the simulation,” Bartol says. “It can’t just look good, it has to be good.”
Previous models simplify the biology—for example, by making all dendrites the same shape. In contrast, Sejnowski’s model mimics reality in all its glorious complexity. Experimenters will be able to test their theories in silico and head back to the bench with new insight. Among the first up: Sejnowski is collaborating with Mary Kennedy at the California Institute of Technology in Pasadena, California, to simulate how synapses reorganize during learning and memory.
As for the neurotransmitter spillover question, the model doesn’t offer a clear-cut answer: some synapses are surrounded by astrocyte membranes, but others are not.
The next step, therefore, is to add action. Kinney is now using the model to examine signaling directly by triggering digital neurotransmitter release and looking for spillover. Each signal looks like an explosion of fireworks as the computer predicts where individual neurotransmitters will end up.
And, just for fun, he also set one of these action simulations to music: the opening fanfare of Richard Strauss’s Also sprach Zarathustra, commonly known as the theme from 2001: A Space Odyssey. It’s a fitting selection for a model that the researchers say could change how scientists look at the neuropil.