Consequences of stochastic release of neurotransmitters for network computation in the central nervous system.

Neuronal membrane potentials vary continuously due largely to background synaptic noise produced by ongoing discharges in their presynaptic afferents and shaped by probabilistic factors of transmitter release. We investigated how the random activity of an identified population of interneurons with known release properties influences the performance of central cells. In stochastic models such as thermodynamic ones, the probabilistic input-output function of a formal neuron is sigmoid, having its maximal slope inversely related to a variable called "temperature." Our results indicate that, for a biological neuron, the probability that given excitatory input signals reach threshold is also sigmoid, allowing definition of a temperature that is proportional to the mean number of quanta comprising noise and can be modified by activity in the presynaptic network, a notion which could be included in neural models. By introducing uncertainty to the input-output relation of central neurons, synaptic noise could be a critical determinant of neuronal computational systems, allowing assemblies of cells to undergo continuous transitions between states.