Statistical properties of information processing in neuronal networks.

Information processing and coding were analysed in dissociated hippocampal cultures, grown on multielectrode arrays. Multisite stimulation was used to activate different neurons and pathways of the network. The neural activity was binned into firing rates and the variability of the firing of individual neurons and of the whole population was analysed. In individual neurons, the timing of the first action potential (AP) was rather precise from trial to trial, whereas the timing of later APs was much more variable. Pooling APs in an ensemble of neurons reduced the variability of the response and allowed stimuli varying in intensity to be distinguished reliably in a single trial. A similar decrease of variability was observed pooling the first evoked APs in an ensemble of neurons. The size of the neuronal pool (approximately 50-100 neurons) and the time bin (approximately 20 ms) necessary to provide reproducible responses are remarkably similar to those obtained in in vivo preparations and in small nervous systems. Blockage of excitatory synaptic pathways mediated by NMDA receptors improved the mutual information between the evoked response and stimulus properties. When inhibitory GABAergic pathways were blocked by bicuculline the opposite effect was obtained. These results show how ensemble averages and an appropriate balance between inhibition and excitation allow neuronal networks to process information in a fast and reliable way.