QUOTE(Rick @ May 06, 2008, 03:07 PM)
That makes sense as a neuron can receive no new information from itself, so self-synapsing ought not to be evolutionarily advantageous, and would therefore be selected out.
Although autoreceptors are a common occurrence, which is a little bit like self-synapsing.
Autoreceptors are common on the axonal terminals of neuromodulatory neurons (serotonin and dopamine, for example). When the neuron fires, the axon's synaptic terminal releases neurotransmitter intended for a target cell. However the presynaptic terminal contains receptors for the same neurotransmitter it is releasing, called autoreceptors. These provide an inhibitory feedback signal to the sending neuron so that as neurotransmitter accumulates in the extracellular space, the axon becomes less and less likely to release more neurotransmitter.
Eventually the extracellular neurotransmitter is cleaned up via either reuptake (for reuse) or degradation. This deactivates the presynaptic autoreceptors.
In general, inhibitory self-synapsing would be okay for self-regulation, and excitatory self-synapsing would be a big problem due to self-reinforcing feedback loops (like microphone-to-speaker feedback).
When a neuron spikes, which is what causes neurotransmitter release at the synapse, news of that spike travels backwards to all dendrites (called backpropagating action potentials). So as Rick points out, there would seem to be nothing a neuron could learn from a self-synapse that it wouldn't already know via the intracellular backpropagating action potential mechanism.