The electrical symphony of the human brain, with billions of neurons firing at different rates, up to hundreds of times per second, likely looks like chaos to any outside observer. But there are patterns in the ongoing brain activity seen, for instance, on an EEG: slow oscillations, rhythmic coordination, and purposeful ripples of communication. The importance of this intricate harmony is best displayed when it is disrupted by an epileptic seizure, which turns the fascinating complexity of the EEG into an angry scrawl.
You don’t have to be a neurologist to see the difference between a brain’s normal behavior and a seizure, but the causes of those seizures are much less obvious. Current antiepileptic drugs have shown success in treating some forms of epilepsy, but in many cases therapeutic success or failure is poorly understood and positive results are almost accidental – doctors are not entirely sure how medications suppress seizures, but are happy when they do. But for roughly a third of patients with epilepsy, those with intractable epilepsy, there remain no such happy accidents. Understanding what sparks a seizure would provide a rational basis for scientists to develop new drugs to treat the untreatable, as well as to reduce the side-effects of the existing treatments.
“Nothing has moved in the last 20 to 25 years,” said Wim van Drongelen, professor of neurology at the University of Chicago Medical Center. “There have been a lot of new anti-convulsant medications, but that one-third of patients who do not respond to medication has remained the same. My conclusion from that is that apparently all the new medications that have been developed address more or less the same type of epilepsy. In this context, epilepsy is comparable to cancer – there’s not just one type of cancer, and there’s not just one type of epilepsy, there are multiple types.”
To understand the different ways a seizure can form, scientists need a model. Experimentalists have recorded EEGs or used higher-resolution methods such as electrophysiology to measure cellular activity in a slice of animal or human brain tissue (obtained during surgery). But to truly model the brain’s rhythms – both normal and abnormal – requires nothing less than the most powerful computers currently available, a task that van Drongelen’s lab has undertaken.
“It’s a lot easier to do an experiment with a computer model than in a real slice,” van Drongelen said. “In a real slice, you have drugs to affect a certain channel, but these drugs are dirty, they also affect other things. In a model you can really very purely see what the effects are of certain manipulations and components. An additional huge advantage is that this approach gives you simultaneous access to what the population on the whole is doing, and what the individual agents are doing.”
The laboratory’s model replicates a relatively small piece of brain neocortex, the region where many seizures begin to erupt. As many as 150,000 neurons, simplified down to the electrical activity generated by their sodium and potassium channels, can be observed simultaneously as the model runs. Researchers can watch as an aberration in one part of the neocortex spreads along millions of neuronal connections to throw off the normal rhythms of the tissue model. They can also test drugs in the computer, programming the blockade of a specific type of ion channel to simulate the effect of a particular agent.
In action, the model is beautiful and it provides the potential to explore hypotheses far beyond what any lone electrophysiologist could accomplish in a lifetime of experiments. But modeling the brain is no easy task: van Drongelen’s model required two years to build and refine, and can only run on the powerful state-of-the-art supercomputers at Argonne National Laboratory.
“There is no way that you can do these kinds of simulations on a normal PC,” van Drongelen said. “I always make a joke: If I want a 10-second multi-channel EEG, it takes 10 seconds, but with the model, if I want to have a 10-second one-channel EEG, I need several days of a supercomputer.”
But the rewards to be gained from that labor are significant. A paper from earlier this year published in the journal Neurocomputing used the model to look at a well-known neuronal phenomenon: oscillations. Literal brain waves, the oscillations can be measured by EEG or – in rare circumstances – through multi-electrode arrays. But to watch hundreds of cells simultaneously in high resolution as they oscillate is technically impossible, outside of a computer model.
The research team, led by Jennifer Dwyer, “injected” a small amount of current into their model and simply let it run. The neurons in their computerized cortex quickly fell into a pattern of oscillation that replicated what one would typically observe in an EEG recording. But the higher resolution afforded by the model allowed the team to figure out why networks of neurons fall into oscillation patterns. The answer was not a coordination of electrical spikes, as was theorized, but a more subtle fluctuation pattern of the cell’s electrical baseline: resonance.
“This gets at the idea that one of potential mechanisms that could start seizures could be resonance,” van Drongelen said. “Think of the Tacoma bridge, where resonance did lead to extreme large fluctuations and in the end the bridge was destroyed.”
The laboratory has moved on to modeling these resonance effects in brain slices, including some from mice genetically modified so that researchers can activate neurons with simple light. As van Drongelen cautioned, bringing things learned from the computational model to the biological slice is important, in order to confirm that the computer truly reflects reality.
“We are not shy about using all this technology,” van Drongelen said. “The central question is the mechanisms that will evoke epilepsy and whatever technology we can use – optogenetics, optical recordings, electrical recordings, mathematical analysis, computational tools – all these things we do to help picture what is going on around the seizure and seizure onset.”
Dwyer J, Lee H, Martell A, Stevens R, Hereld M, & van Drongelen W (2010). Oscillation in a Network Model of Neocortex. Neurocomputing, 73 (7-9), 1051-1056 PMID: 20368744