Redwood Neuroscience
Title: The Role of Dominant Feedforward
Inhibition in Cortical Layer 4
Kenneth
Miller
Neuroscience
and Physiology
Abstract:
We
study the
circuitry underlying the response properties of simple cells in layer 4 of cat
V1. Rectification of LGN firing rates at
0 induces a nonspecific component of LGN input to these cells: a nonspecific
input will activate some of a simple cell's inputs, which can greatly raise
their firing rates, while suppressing other inputs, which can only lower their
firing rates to zero. The result is a
net positive input. In the case of
response to a drifting grating, this nonspecific input is the mean rate of LGN
input, which is untuned for orientation and grows
with contrast. To suppress this input
and achieve sharp tuning, an orientation-untuned
component of feedforward inhibition is required. Several studies have shown that the net
inhibition received by a cat V1 layer 4 simple cell is
tuned for orientation, with similar tuning to the excitation received, but is
strongest at the opposite phase (light or dark) to the phase that best drives
excitation. We therefore initially
suggested that inhibitory simple cells connected to excitory
cells of roughly opposite phase would respond to all orientations, but our
strongest prediction was that a set of inhibitory cells would exist in layer 4
that responds to all orientations in a contrast-dependent manner.
These
predicted inhibitory cells have now been found (Hirsch et al, Nature
Neuroscience, 2003). About half of layer
4 inhibitory cells were simple cells, and only a minority of these showed
response to all our orientations. But
the remaining layer 4 inhibitory cells were complex cells, with mixed ON and
OFF responses throughout their receptive fields, and these were essentially untuned for orientation. Thus we now predict that these
complex cells eliminate the nonspecific LGN input, while simple cells that are
predominantly well tuned for orientation provide the antiphase
and orientation-tuned inhibition known to be received by layer 4 simple cells.
We
show that this basic idea can account for cortical temporal frequency as well
as orientation tuning, provided the complex cells have temporal tuning that
follows that of LGN and a reasonable fraction (~35%) of geniculocortical
excitation is carried by NMDA receptors.
The basic idea common to both orientation and temporal frequency tuning
is that the mean feedforward inhibition dominates the
mean feedforward excitation, so that only modulations
of excitation that exceed this mean can excite the postsynaptic cell. As one moves in orientation away from the preferred
orientation, the LGN input to a simple cell loses its modulation because its
LGN inputs become driven at different phases.
As one moves in temporal frequency toward higher frequencies, the LGN
input lose some of its modulation because of the NMDA-mediated component, which
is long-lasting and acts like a low-pass filter.
The
problem of nonspecific input induced by input rectification is a general one --
inputs to any piece of cortical layer 4 have low spontaneous rates, and can be
driven to greatly increase their firing rates whereas rates can decrease only
to zero. Therefore there is a general
need for a nonspecific component of feedforward
inhibition to maintain selectivity to feedforward
excitation -- to respond only to patterns of input activity that are
well-matched to a cell's particular pattern of received inputs, regardless of
the magnitude of input activation. Thus
such dominant, nonspecific feedforward inhibition may
be a general property of layer 4 processing, designed to yield magnitude-invariant
form recognition. Evidence of this has
also been seen in rat whisker barrel cortex.