Re: Holloway/Morgan ...the BAH....

HARRY R. ERWIN (herwin@osf1.gmu.edu)
3 Aug 1995 12:33:21 GMT

Ralph L Holloway (rlh2@columbia.edu) wrote:
: On 31 Jul 1995, HARRY R. ERWIN wrote:

: > ---trimmed---
: Somewhere there are some calculations regarding how many synaptic
: connections there might be on a cortical neuron (visual cortex?) and the
: number was somewhere in the order of about 10,000, which if true would
: mean that modification at the individual dendritic spine in one neuron
: cannot be a neuron-alone phenomena, but must also include some
: complicated relationship with the surrounding glia.

The number of synapses is generally on the order of 10^4. That means the
modifications involved in learning must occur at many synapses in the
neuropil. That doesn't mean the glia are necessarily involved; although
there is some evidence of that.

: I am here well out of
: my depth, but I recall over the past 30 years several attempts being made
: to show glial-glial electrophysiological exchange as well as between
: neurons and glia, i.e., that the relationship was more than purely one of
: biochemical sustenance. I would be grateful to Harry if he could bring us
: up to date on this.

I don't follow this that closely. There isn't anything in Arbib's Handbook.

: So, with neuron death (i.e. one neuron, say in the visual cortex)
: some 10,000 synaptic connections would be lost with what must be
: extremely minor functional consequences, underlining the high redundancy
: necessary to a holographic model ( I assume Karl still holds to that
: model ?).

That was 30 years ago. Since I've known him (about 5 years), he has been
talking about synapto-dendritic field models where the real processing
goes on in the neuropil, with the neurons sampling the field and using it
to modulate their endogenous firing activity. (Yes, this is related to the
old holographic model.) Hayashi's results on the role of metabotropic
glutamate receptors in generating the gamma wave suggest how the
endogenous firing could be generated. I suspect that when we look at this
carefully, we'll discover that the standard received model of the neuron
and Pribram's synapto-dendritic field model are formally dual in some
sense. Pribram saw Dan Alkon's WCNN '95 presentation on learning in the
dendritic arbor two weeks ago, and I think they believe they might now
have identified the actual mechanism that implements the synapto-dendritic
field. I'm in the middle of writing a dissertation proposal, and this is
one area I want to address.

: Of course, not all neurons are equal, functionally, and perhaps
: the loss of neurons in a gating structure, or one associated with fine
: tuning (e.g., striatum, cerebellum) will lead to more "serious"
: functional consequences over time.

Yes especially in the nuclei, although again it's loss of numbers that
matters (Parkinson's disease and perhaps Alzheimer's). There's a
threshold effect involved, which is why fetal cell implants are so
promising as a treatment for Parkinson's.

: It would be useful to have a refresher
: "course" regarding the role of plasticity in dendritic arborization as
: seen these days with regard to both selective cell death and neuritic
: growth.

The degree of neuronal growth that actually occurs during learning is a
moot point, and I've heard both sides recently at conferences.

: Our static model of so many neurons-so many connections-and
: inevitable loss of neurons through aging surely needs updating.

Yes, since much of the loss seems to be programmed cell death. Give me 6
months and I'll try to report back what I've learned.

: There was an interesting piece in Tuesday's NY Times Science Section
: regarding some MRI work done on people with severe emotion trauma in
: which it appears that these episodes can actually damage the nervous
: system (e.g., the hippocampus) and reduce the volume
: up to 25 % or more, results that seem to imply the action of cortisol in
: the brain. Here is another area in which we know just about nothing
: regarding individual variation and these processes.

The exponentially fast pattern classification process that seems to occur
in the olfactory bulb doesn't work unless the intensity of the afferent
input is multiplicatively normalized at some stage to a narrow range.
This was seen in my modeling work over the last three years
(@inProceedings{herwin:93a, author = {Harry R. Erwin}, title = {The
Application of `\uppercase{K}atchalsky \uppercase{N}etworks' to Radar
Pattern Recognition}, booktitle = {Origins: Brain and Self-Organization},
year = 1994, editor = {Karl H. Pribram}, publisher = {LEA, INNS Press},
address = {Hillsdale, New Jersey}, note = {Proceedings of the Second
Appalachian Conference on Behavioral Neurodynamics, October 1993}, } and
@unpublished{herwin:94d, author = {Harry Richard Erwin}, title = {A
Simulation Model of the Olfactory Bulb}, note = {Class project, submitted
December 1, 1994}, year = 1994, }) and is also noted by Hayashi
(@article{hayashi:93, author = {Hayashi, T. and others}, title = {Role of
a metabotrophic glutamate receptor in synaptic modulation in the accessory
olfactory bulb}, journal = {Nature}, volume = 366, pages = {687--690},
year = 1993, }). If the afferent input is too low or high, or if the
inhibitory feedback is insufficient, the system drops into a fixed point
and neither processes the input nor learns. This is apparently seen in
children when they are over or under-stimulated. Since a child's sensory
processing systems are immature, they have to rely on their parents or
other adults to control their level of sensory stimulation. This is also
apparently seen in anxiety disease and in cases of emotional trauma. Such
anxiety control drugs as Buspar and desipramine act by increasing the
release of GABA when inhibitory interneurons in these systems are
activated. Chronic over-stimulation of a sensory processing system is
likely to lead to permanent changes in the sensory systems and the
hippocampus to reduce the level of stimulation to the point where it can
be processed. I wouldn't be surprised at the result reported in the NY
Times in this context. There are similar results that have been reported
in Neural Networks with regard to Alzheimer's (unfortunately not at hand). 

--

Harry Erwin

Internet: herwin@gmu.edu 

Home Page: http://osf1.gmu.edu/~herwin (try a couple of times)

PhD student in comp neurosci: "Glitches happen" & "Meaning is emotional"