12.4. Nerve, muscle, synapse[[Student version, January 17, 2003]] 479
12.4.3 Vista: Neural computation
The situation is similar with synapses between neurons. An axon terminal can release a variety of
neurotransmitters, altering the local membrane potential in another neuron’s dendritic tree. The
main difference between the neuromuscular and neuron-neuron junctions is that the former acts as
asimple relay, transmitting an impulse without fail, whereas the latter are used to perform more
subtle computations.
The effect of an arriving presynaptic action potential can either depolarize or hyperpolarize the
postsynaptic dendrite, depending on the details of what neurotransmitter is released and the nature
and state of the receiving point’s ion channels.^13 In the former case the synapse isexcitatory;in
the latter case it isinhibitory.
If the total depolarization in the soma near the axon (the “axon hillock”) exceeds a threshold,
the neuron will “fire,” that is, generate an action potential (Section 12.2.5 sketched how threshold
behavior can arise in the context of the axon). In many neurons, the arrival of a single action
potential at a dendrite is not enough to make the cell fire. Instead, each incoming presynaptic
impulse generates a localized, temporary disturbance to the membrane potential, similar to elec-
trotonus (Section 12.1.3 on page 449). If enough of these disturbances arrive close enough in space
and in time, however, they can add up to an above-threshold stimulus. With thisintegrate-and-
fire modelof neuron activity we can begin to understand how a neuron can perform some simple
computations:
- Adding up those disturbances that overlap in time lets a cell measure the rate of incom-
ing action potentials at a particular synapse. Thus, although all action potentials
along a given axon are stereotyped (identical), nevertheless your nervous system can
encode analog (quantitative) signals asratesof action-potential firing, a “rate-coding
scheme.” - Adding up those disturbances that overlap inspace,that is, those arriving in the same
neighborhood of the dendritic tree, lets a cell determine whether two different signals
arrive together.
One simple model of neural computation is to suppose that the cell sums its input signals with par-
ticular weights, corresponding to the excitatory or inhibitory character of each component synapse.
Then it fires if the sum exceeds a threshold, like the one we found in our simple model (Figure 12.9b).
Acrucial aspect of the scenario just sketched is that neurons can adjust their synaptic couplings,
for example altering the numbers of ligand-gated channels at a dendritic spine, and so alter the
computation they perform. Neurons can also modulate their connections by adjusting the amount
of neurotransmitter released in response to an action potential, and in other ways. Taken together,
such reconfigurations allow a network of neurons to “learn” new behavior.
Connecting even a few dozen of such simple computational devices can yield a system with
sufficiently complex behavior to operate a simple organism, like a mollusk. Connecting a hundred
billion of them, as your body has done, can lead to very complex behavior indeed.
(^13) T 2 It’s also possible for a neurotransmitter to have an indirect effect on the postsynaptic membrane; for
example, it can alter a voltage-dependent conductance that is not currently active, modulate the response to other
synaptic inputs, and so on.