Detecting and estimating signals over noisy and unreliable synapses :: Amit Manwani & Christof Koch, California Institute of Technology :: Neural Computation 13, pp. 1-33, (2000) http://www.mitpressjournals.org/loi/neco
Summary and review of the above paper
This paper discusses the relationship between noise and information transfer in the brain. Synaptic transmission is known to be unreliable and probabilistic. Single cortical synapses cannot transfer information reliably. Therefore, a small number of multiple synapses serves to improve information capacity. The signalling release of neurotransmitters occurs through vesicles situated in the presynaptic terminal. Each vesicle releases independently and the occurrence of release is probabilistic.
In muscles, the probability of neurotransmitter release from individual vesicles is low, but the number of release sites makes the muscle function reliable. In the brain, there are many fewer release sites compared to 1,000+ release sites in muscles. With an action potential in the brain, at most one vesicle will release neurotransmitters, and the probability of individual vesicle release is low. This is seen as the main reason for the unreliability of synaptic transmission in cortical and hippocampal neurons. It is suggested that the randomness of release is related to the location of calcium ion channels in the presynaptic membrane and their relationship to the vesicles.
Avoidance of local minima
The probability of vesicle release is not a constant; instead, it depends on whether the last action potential led to a vesicle release and the timing of the earlier presynaptic spikes. This latter sequence governs whether the tendency to release neurotransmitters increases or decreases. This form of plasticity is suggested to lie at the basis of memory and learning.
The unreliability of presynaptic transmission is compounded by variability in the amplitude of the postsynaptic response related to the original vesicle size and the number of neurotransmitter molecules. The variability of the total (pre and post) synaptic connection is indicated to be several orders of magnitude greater than those expected in a manmade engineering system. This, however, is offset by making use of multiple synapses. The unreliability of the system has been something of a puzzle for neuroscience. One suggestion is that unreliability allows greater plasticity in changing release probability. In particular, it may avoid brain systems getting stuck in a local minima, and thus allow them to learn more easily.