Memory And Critical Avalanches In The Brain

Physicists at Indiana University are extending their study of the relation between observed patterns of neuron activity and memory storage in the brain. First came experimental work with slices of rat brain. Later the researchers performed simulations to try to emulate the data. Activity in the actual samples displayed two fascinating features: (1) the ensemble of neurons firing varies in size very much like "avalanche" phenomena such as occur in sandpiles and forest fires; and (2) there are stable activity patterns that resemble memory sequences measured in lab studies of rats in a maze. Every time a rat runs a particular route the same sequence of neural firings occur. At night the same sequence might be replayed as a rat "dream." If the rat's dream is interrupted, his ability to run the same route the next day might be compromised. This has added evidence to the notion that sleeping and dreaming help to consolidate memories from the previous day's activities. Stable activity patterns also appear in artificial neural networks as a way of storing information.

The Indiana physicists take a fundamental look at those patterns. They used a 60-electrode array to look at firings in a thin slice of rat brain tissue. The cells in the slice, supplied with oxygen and nutrients, go on behaving as if they were part of a living brain. The general ensemble firing of cells is classified as subcritical (one cell firing leads, on the average, to less one additional cell firing), critical (one firing leads to another firing), or supercritical (a firing leads to two or more cells firing). In this regard, neural cells triggering each other are somewhat like chain reactions among uranium-235 atoms in a nuclear reactor.

The subcritical case is uninteresting. The supercritical situation often leads to the case in which all the cells in the sample end up firing, which is also uninteresting. The critical case has the most to offer: neural ensembles of all sizes ensue. If you plot (with logarithmic rulings) the number of firing events versus the size of the firing ensemble, you get a straight line, indicative of classic "power law spectrum" behavior. In other words, the likelihood of an event (earthquake, sand avalanche, hurricane) of size E drops off according to E raised to a negative exponent.

Now, in the simulation work, the notion that the most interesting outcomes occur when the brain system is maintained right at criticality is reinforced. The simulations, which do roughly match the observed behavior, are surprising and even counterintuitive. This is because precisely amid conditions which favor the greatest number of avalanches the largest number of stable neural activity patterns also occurs. One of the researchers, John M. Beggs, says that the work is meant to explore how avalanches in brain cells might be used to store information. (Haldeman and Beggs, Physical Review Letters, 11 February 2005,jmbeggs@indiana.edu, 812-855-7359; lab website, http://biocomplexity.indiana.edu/research/info/beggs.php )