Chemical signals in the learning brain

Sandy Hodges QXUXBTVOTSAO at spammotel.com
Wed Apr 21 12:08:55 EST 2004


The signals I have in mind are those exchanged between neurons, other
than the chemical signals involved in the firing of synapses.   I
include any chemical released by a neuron as a function of its
long-term or permanent state, which is detected by another neuron, and
which causes a synapse between the two neurons to be formed where it
would not be otherwise.  (Or not be formed when it would be otherwise,
or be inhibitory rather than excitatory, strong rather then weak,
etc.).   The point of the signals is that they allow a neuron to
behave differently, according to the activation state (which genes are
activated) of the neuron on the other side of the synapse.

    Do such signals exist?    In the fetal brain, they clearly do.  
The wiring of the brain, that is, which neurons form synapses with
which other neurons, is just about completely controlled by chemical
signals.   The forming of synapses by Hebbian learning, with cells
that fire together wiring together, does not start until the basic
wiring has been laid down.

    Certainly one mechanism used for structuring the brain, is to set
up three gradients of chemicals, so that a neuron can simply migrate
to the right position in three dimensional space.    Axons and
dendrites can go to the approximate right place, based on this system 
 But a dendrite needs to connect to the right neuron, not merely to a
neuron in roughly the right place.   Take for example the ocular
dominance columns of the lateral geniculate nucleus.    These form in
the non-abused animal, and they also form in an animal whose eyes are
covered from birth.   But any number of things can cause them to occur
in a slightly unusual pattern.   A dendrite or axon coming in from a
distant part of the brain, which needs to connect to a left-eye
neuron, can't rely solely on the location of the neuron.    If the
patterning of the columns was just a little off, then the wiring would
end up completely wrong, if the incoming dendrites relied on the
neurons being in their blueprint locations.

    Several families of signaling chemicals are now known; the role of
chemical signals in the initial structuring of the fetal brain is
becoming clear.   It is not so clear that neurons continue to be in
different classes, and to give off chemical signals of their class,
once the fetal stage is over and learning from experience begins.  
But it seems overwhelmingly likely that they do - certainly it makes
no sense to merely assume that all the neurons in some region of the
brain are the same at birth.

     Neural net modeling does exactly this, starting with a population
of identical starting neurons.    As a result of learning, neurons
become associated with input conditions.   For example, a neuron may
respond to a particular person's face.    But neural net modeling
usually assumes (on no evidence that I can see) that there is no
connection between the signification state of a neuron (what it
responds to), and its activation state (which genes are activated).   
But it is no longer possible to believe that babies are blank slates
at birth.    Which neuron, or pattern of neurons, will end up
signifying the baby's mother's face, is probably not pre-determined.  
Learning leads some neurons to specialize in this frequently seen
face, while other face-recognition neurons specialize in other faces. 
 But it is very easy to implement that a gene becomes activated, only
in the neurons that respond to the most frequently seen faces.   Some
such mechanism, is the only way for the special behavior which a baby
shows for her care-givers, to be implemented.    But once a neuron has
determined that it is recognizing a care-giver's face, and activated
the appropriate genes, what are those genes going to do?

    Now consider a dendrite growing toward the face-recognition
neurons.   For the genetically determined behavior of loving one's
care-givers to occur, this dendrite needs to connect with a neuron
that responds to a face of some loved care-giver.    Those neurons are
distinguished by some particular genes, which are activated only in
them.    Synapses are cheap - the dendrite connects to many neurons,
and then keeps only the synapses it wants.   So we have a synapse
between two neurons.  One neuron knows it is a "mother's face" neuron,
and has activated the appropriate gene or pattern of genes for this.  
The other neuron knows it is a "smile" neuron, and the appropriate
genes are activated for that.   Smiling at your mother is genetically
determined.    But how does either neuron know what sort of neuron is
on the other side of the synapse?    Well, we know that neurons give
out chemical signals, and we know other neurons respond.

    In spite of all this, by the Hebbian paradigm, the only way that
the "mother's face" neuron can strengthen the synapse with the "smile"
neuron, is for them to recognize each other by their pattern of
firing.   It simply won't work.



More information about the Neur-sci mailing list