Forward Dendritic Spikes - Advancing Artificial Intelligence Through Biological Process Applications - page 49

LESSONS FROM THE
COMPUTATIONAL MODELS

currents and the initiation of dendritic spikes upon
intense synchronous activation. In our opinion,
these model studies had limited application as they
were devoted to reproduce partial experimental
data, i.e., they came in support of the dominant
idea by the experimenters at the time (Hoffman et
al., 1997; Mainen et al., 1995; Rapp et al., 1996).
In general, a poor scrutiny of critical parameters
yields rigid models, which are ineffective for
reproducing any other electrical behavior than
those they were designed for. We want to rise
here a criticism to such a frequent practice, as
non-specialized readers, for whom complex
mathematical models are impenetrable jumble,
may take them as a definitive demonstration of
a certain mechanism, when it merely supports
its possibility.
As an example of how the poor reliability
of essential experimental data affects models,
we can take the reproduction of the most em-
blematic event, the somatic action potential. By
the time single-cell models became accessible
to physiologists in mid-nineties, they were us-
ing in vitro preparations to study intensely the
backpropagation of somatoaxonal action potential
into dendrites, which by the way were thought to
have too weak excitability to initiate and conduct
forward spikes. The rule was to built the soma and
the main apical dendrite with an homogeneous
low density of Na channels that enables partial
dendritic invasion of the spike but not its initiation.
The prototype somatic spike chosen for modeling
was that recorded in vitro . In fact, much higher
densities are needed to generate realistic action
potential waveforms at the soma as recorded in
vivo, where it is much faster (Varona et al., 2000).
Surely, the difference arises from the known de-
pression of Na + currents in non intact or immature
preparations (Spigelman et al., 1992), which may
have only moderate consequences at the soma,
but certainly it entails dramatic changes in local
dendritic excitability (Shen et al., 1999; Varona
et al., 2000).

The rich repertoire of computational capabili-
ties endowed by the multiple properties arising
from membrane channels constitutes the greatest
challenge for experimental studies. Some mod-
ern high-resolution techniques require excessive
tissue manipulation or dissection, while this is
strongly unadvised to study highly non-linear
systems. Too often, the experimental approach
cannot discriminate between real and fake
mechanisms unveiled by an excessive perturba-
tion of the system. Biophysical computation is at
present the best approach. Next, we will review
the computational studies on active properties of
pyramidal cell dendrites in relation to the present
issue and will use our own computational work to
describe the most relevant mechanisms.

First Models and the Reliability of
Experimental Data

The extensive knowledge of channel composi-
tion and subcellular distribution as well as the
fine anatomical details obtained in the late years
have facilitated the construction of mathematical
single cell models with a high degree of realism.
However, the single-cell models leading the way
used ad hoc channel assortments and distributions
in order to reproduce experimental data. In those,
a single axonal trigger in the axon initial segment
was assumed, and typically assembled with a huge
density of channels while dendrites were given
too few. The expected outcome was that dendrites
never initiated spikes. These began in the axon
and invaded dendrites in a decremental manner to
extinction. Later on, as dendritic recordings were
refined and the channels therein were identified
and quantified, the biophysical models become
more realistic and reproduced gross active prop-
erties of dendritic electrogenesis as observed in
experiments, including the reshaping of synaptic