Contributions of intrinsic neuronal dynamics to input-output information transfer

The human brain can be said to be the most complex system known today. It contains billions of cells called neurons. Simplified models of neurons have been shown to reproduce many of the experimental results available. Therefore it is important to understand these models well in order to gain some insight into the neural code used by nervous systems. Information theoretical techniques allow us to quantify the input-output relationship in those models. In this work we have applied information theory to neuronal models while focusing on two important issues: input signals containing naturalistic correlations and models that include intrinsic dynamics.; To determine the input-output mutual information we have used a simple and efficient algorithm, based on the direct method (Strong et al. 1998). Spike frequency adaptation, which is a feature shared by most neurons, is shown here to enhance the encoding of signals with naturalistic statistics. Two distinct aspects of adaptation were analyzed in detail, namely the reduction in the spiking rate and the decorrelation of the output. Our results indicate that adaptation provides a nerve cell with an excellent mechanism to improve its use of the energetically expensive action potentials.; More complex spiking patterns known as bursts, were also analyzed. The mutual information calculations were also applied to investigate the difference between encoding by bursts and encoding by individual spikes. We have found that isolated spikes have a much larger variability (entropy) than spikes belonging to a burst, due to the strong correlations among the spikes within a burst. However bursts were found to be clearly more robust to noise than isolated spikes. We also found that bursts, taken as single events, are capable of carrying information more efficiently than individual spikes. These facts lead us to conclude that despite the information transmission per spike being smaller in bursts, these may constitute a preferable encoding mechanism, as opposed to single spikes, when neurons have to convey sensitive information.