Random Connections


getmedia/0f76cece-43b3-42a6-8228-9a0103d6bdf7/Random_Connections-thumb.jpg.aspx?width=60&height=60&ext=.jpg

With its 100 million neurons per square inch, the brain is a pretty powerful processor, even if we can’t always beat computers at chess these days. But just how the circuits that make up that wondrous seat of consciousness form themselves has long been anybody’s guess.

The leading guess has been that as the brain develops, neurons grow to meet each other. The axon (the long spindly part) of one neuron, and the dendrite (the conductive coral-like branches) of another were thought to somehow chemically sniff each other out and form a synapse.

In fact, the process is much more efficient than that. It’s random.

At least that’s what researchers working on the “Blue Brain” project at the École Polytechnique Fédérale de Lausanneat say. The Blue Brain project’s goal, simply put, is to create a computer model of all the brain’s circuitry. “It forces us to ask questions we didn’t ask before,” says Sean Hill, executive director of the International Neuroinformatics Coordinating Facility and a member of the Blue Brain team. “How did they get into those positions? Are they dependent on their neighboring neurons?”
 

Neuron Growth

For 15 years, researchers at the university have been painstakingly collecting data from the live brain tissue of various animals. (The animals were dead, but the fresh slivers of their brains were “alive”). With a patch clamp on one neuron they would stimulate others to see if they were connected. When they found a connected pair, the researchers injected the neurons with a dye to reveal all the fibers of the axon and dendrite. They could see where one neuron came near another, where synapses were formed, where there were little swellings.

Neurons magnified 400x.

Eventually they had a very real map of a very real third of a cubic centimeter of a rat brain. And though the synapses seemed at first scattered haphazardly throughout, a pattern soon emerged, with a greater number of connections in certain areas.

Armed with this data, Hill’s team was able to test any model of neuron growth and organization they might dream up against the real thing. As a sort of preliminary test, Hill’s team built a model with about 10,000 neurons arranged randomly, “because we didn’t know any specific orientation to put them,” he says. And they looked at where they might bump into each other. They ran an algorithm to find all the possible places where a synapse could physically form between neurons.

We were convinced this would be wrong,” says Hill. “No way you could throw a bunch of neurons together and find synapse locations. We thought we’d just see how we’d need to adjust things to match the biology.”

In fact, they didn’t need to adjust things at all. “It matched the biology. We said, ‘Wait a minute, how’s that possible? What did we do wrong?’ It turns out neurons are not shaping themselves to fit other neurons. They form their shape and then wherever they come close to another neuron, that’s the space where they form synapses. That’s one big insight.”

Neuron Shapes  

Another was to follow. The team wanted to see how important the shape of the neurons were to the layout of the circuit. “We said, ‘Let’s be really stupid and say that each neuron type is represented by a single shape. We know that’s not true, but let’s see what the impact of that is on the innervation patterns.”

They built ten circuits, each with the same few neuron shapes, but in different arrangements. The subsequent synapse patterns were as varied as the neuron positions. But when they increased the variation of shapes, the pattern of synapse locations emerged. “As you start to grow the diversity of neurons, the distribution starts to get more and more stable,” says Hill. When the variation was increased to the point where each shape was unique, the subsequent circuit was the most robust of all. In essence it’s the shape of a neuron—its type coupled with natural variation—that allows it to make its input and output point in the statistically correct place. That is, it’s the shape of each neuron, and the multitude of shapes, that dictates the wiring of the circuit.

It’s a self-organizing principle that beats DNA for concision. If our genes had to carry that map, we’d probably need an extra chromosome or two.

“This has given us real insight into the design of the cortical circuit, into how evolution dealt with the fact that biology is so variable,” says Hill. “From my point of view, it’s a pretty cool technique for building a circuit. We don’t engineer things that way, but if you have a noisy process you’re building your circuit out of, how do you make the most of it?”

The answer? Get your cells in shape and let the neurons fall where they may.

Michael Abrams is an independent writer

We were convinced this would be wrong. No way you could throw a bunch of neurons together and find synapse locations.

Sean Hill, executive director, International Neuroinformatics Coordinating Facility

getmedia/0f76cece-43b3-42a6-8228-9a0103d6bdf7/Random_Connections-thumb.jpg.aspx?width=60&height=60&ext=.jpg

December 2012

by Michael Abrams, ASME.org