Bursting as a source for predictability in biological neural network activity

The role of bursting as a unit of neural information has received considerable support in the recent years, Experimental evidence shows that in many different neural systems, e.g. visual cortex or hippocampus, bursting is essential for coding and processing. We have recently demonstrated (Menendez de la Prida et al., 1996) the spontaneous presence of bursts in in vitro hippocampal slices from newborn animals, providing a good system to investigate bursting dynamics in physiological conditions. Here we analyze the interspike intervals (ISIs) of five intracellularly recorded cells from immature hippocampal networks. First, we test the time series against Poisson processes, typical of pure random behavior, using the Kolmogorov-Smirnov test, Only 2/5 records strongly deviate from Poisson process, Nonlinear prediction tests are then applied to compare original series with its Gaussian-scaled random phase surrogates and signs of short time predictability are observed (1/5). This predictability is originated by the intrinsic structure of bursts, in an otherwise purely random process, and can be removed completely by eliminating the bursts from the original time series. Here we introduce this method of eliminating bursts to get insight into the nonlinear dynamics of firing. Also the interburst intervals are indistinguishable from pure noise, The analysis of unstable periodicities within the bursts in the original ISIs shows that signs of nonlinearities can be statistically differentiated from their surrogate realizations (Pierson-Moss method). We discuss the computational implication of these results.