4. Selforganization in the Brain
4.2 Development
Now we
will try
to find
evidence for
the hypothesis
that selforganization
plays a role in the development of the brain. At what level can we
expect to find signs of this role? The prerequisites are a large number of
similar, interacting neurons and positive feedback mechanisms. The cortex
combines a high synapse density with a relatively homogeneous structure.
Positive feedback (mutual inhibition and recurrent excitation) is to be
expected due to the abundance of interneurons. Which type of structures must
we look for? The examples given in Part 2 can
provide us some clues. Most of
them are characterized by a spatial
periodicity in at least one dimension: mosaic patterns, rings, regular
stripes. A number of different
labeling techniques
have demonstrated the
presence of
spatially periodic
structures or activity patterns in the mammalian visual
cortex:
1.
A
few weeks after injection of
3H-proline
in one eye, autoradiography of the visual cortex showed the presence of
ocular dominance columns (Hubel,
1982), an extremely regular stripe-like pattern which is periodic in one
dimension. The pattern corresponds to a regular alternation of projections
of both eyes.
2.
Experiments using 14C-deoxyglucose to measure cerebral glucose
utilization
showed
the existence of ocular dominance
columns (DesRosiers et al., 1978)
after occlusion of one eye,
orientation columns
(Hubel, Wiesel and Stryker, 1978) after stimulation
of both eyes with stripes of only one orientation, and
spatial frequency columns
(Tootel, Silverman and DeValois,1981) after stimulation of both eyes with
stripes of all orientations but only one spatial frequency. All these
structures are spatially periodic 'activity patterns' which arise after
special stimulation.
3.
Cytochrome-oxidase (CO) staining of primate visual cortex showed a 'regular
patchy distribution' (Horton and Hubel, 1981; Hubel, 1982, Wong-Riley and
Carrol, 1984) which is very similar to the static mosaic patterns found in
the BZ-reaction. This periodic structure does not require any special visual
stimulation.
4.
Immunocytochemical localization of the GABA synthesizing enzyme GAD showed a
columnar organization into patches, matching the cytochrome oxidase
pattern (Hendrickson
et
al., 1981; Hunt, 1982)
5.
A
monoclonal antibody
against an
uncharacterized antigen
on the
surface of certain neurons
(CAT 301) demonstrated a regularly patchy distribution of these neurons
(Hendry et
al., 1984).
6.
After treatment of monkey striate cortex with voltage-sensitive dyes,
spatially periodic activity patterns corresponding to orientation and ocular
dominance columns, were seen (Blasdel and Salam, 1986).
7.
Periodic intrinsic connections have been demonstrated in the visual cortex
of the tree shrew, by local injections of HRP (Rockland and Lund, 1982).
These results are consistent with intra-cellular HRP injections in single
neurons, showing a periodic arborization pattern (Gilbert and Wiesel, 1980;
Wiesel, 1982).
All these periodic structures are interrelated in a complex way, while
experience often has an effect on their development (reviews: Hubel, 1982;
Wiesel, 1982; Swindale, 1982). Swindale (1981a) has explained the
spontaneous formation of ocular dominance patterns out of an initially
unsegregated uniform distribution of projections of both eyes, by a few
simple competition rules.
Both 14C-deoxyglucose and cytochrome-oxidase staining, are
methods for measuring metabolic activity which is coupled to electrical
activity (Torbati et al., 1983;
Sokoloff, 1977). Still, anatomical structures are thought to underlie these
'activity patterns'. The
periodicity seen
in orientation
and frequency columns can be thought of as a direct consequence of
the regularity of the visual stimuli necessary to produce them, but this
relationship is not at all self-evident. Retinotopic mapping of these images
in unconstrained animals during at least 30 minutes will probably produce a
quite homogeneous average stimulation of the visual cortex. Therefore, the
observed regularity of the activity patterns is not imposed by stimulation.
The formation of these periodic activity patterns can be explained by
selforganization. The following mechanism is proposed. Neurons in the visual
cortex are tuned to various clues (orientation, size, contours, spatial
frequency). Prolonged
visual stimulation which is rich in one of these clues will produce
an overall increase of excitation. Inhibitory interneurons, which are
abundant in the cortex, will also become activated, and raise the level of
inhibition. A simultaneous increase of excitation and inhibition is
incompatible with a spatially homogeneous distribution of activity. As a
consequence, the initially homogeneous distribution of activity can become
unstable and a spatially periodic activity
pattern may result,
analogous to the
formation of Benard cells and the static mosaic pattern in the
BZ reaction. In such a
spatially periodic activity structure both the excitatory and inhibitory
neurons can be simultaneously activated.
The other patterns (CO, GAD, CAT 301) do not
require special visual input, they reflect a periodic organization of
the cortex, which is more directly demonstrated by the HRP experiments. The
clear relationship of these periodic structures with energy metabolism and
functional activity, together with their plasticity suggests the following
hypothesis for their development and maintenance. In the initially
homogeneous cortical layers, spatially periodic activity patterns develop
due to the high level of energy dissipation by a selforganizing process as
defined in Parts 2 and 3. The
developing mammalian cortex
differs
from the adult one in
a number of aspects which could favor selforganization. Initially,
the cortex is relatively "self-contained":
sensory projections are not yet established and cortical activity is
not disturbed by unpredictable external sensory input. Interaction between
cells is relatively strong due to
the low number of glial cells and the initial over-production of
synapses. Due to the high
level of plasticity in
this early stage of development, activity patterns can be 'imprinted' in the
wiring diagram. 'Hebbian' synapses (cf. Viana di Prisco, 1984) play an
essential role here. The
periodic intrinsic connections which result, promote organized
activity structures, thus providing a positive feedback mechanism for their
maintenance. Although sensory input
is a priori not a necessary
prerequisite, it may in a later stage help to
keep the level of activity (degree of energy dissipation) at a
sufficiently high level. In this way, sensory input can have an important
influence on the type of structures formed. In some cases (ocular dominance
columns) deprivation of sensory input leads to the absence of structures
normally formed (Swindale,
1981b).
Time-independent spatially periodic activity patterns were found by Kaczmarek and Babloyantz (1977), studying the behaviour of a neuronal network model. Although they failed to see the physiological significance of such solutions, these findings seem to be very relevant in the light of the present discussion. Ermentrout and Cowan (1979) showed the existence of stable spatially periodic activity patterns (both mosaic patterns and 'rolls') formed spontaneously out of an initially homogeneous situation, in a two-dimensional neural network model. Both studies show that selforganized spatial activity patterns can arise in neuronal networks, without making very special assumptions about connectivity and interactions.