The overall
goal of our research is to understand how the eye conveys
visual information to the brain via the electrical signals
of the optic nerve fibers. Our research proposal set
two specific goals: to identify the key features of
this electrical activity, and what messages they carry
about the visual scene: this amounts to "breaking the
code" used by the retina; and to understand how this
code is generated by the neural circuits of the retina.
Concerted
signaling by the retina
We are continuing
our investi-gations of concerted firing among optic
nerve fibers. As reported previously, nearby retinal
ganglion cells of the salamander tend to fire synchronously
(within ca 10 ms of each other) much more frequently
than expected by chance. Such strong correlations are
mostly found between cells of the same functional type.
The effect decreases exponentially with distance between
the two cells, with a length constant of about 200 mm.
A more detailed analysis showed that ganglion cells
are coupled not only in a pairwise fashion; rather,
the phenomenon involves larger groups of cells firing
in synchrony, up to 7 neurons at a time in our recordings.
It also became apparent that the concerted firing patterns
persisted under a broad range of visual stimuli, accounting
for about 50% of all action potentials recorded from
the ganglion cells. This suggested that they play an
important role in visual signaling, and two extreme
hypotheses could be formulated: (a) If two ganglion
cells carry the same spike trains most of the time,
then their signals are obviously redundant; thus the
population of ganglion cells would collectively convey
much less information than one might estimate on the
basis of the classical single-cell recordings from these
neurons; (b) Alternatively, each cell might participate
in several different firing patterns, and each such
pattern of synchronous spikes might carry a very specific
visual message; in this case, the range of messages
that the retina conveys to the brain might in fact be
much richer than expected from the single-cell studies.
Recently,
we have obtained evidence for the latter view. Statistical
analysis showed that an individual ganglion cell does,
in fact, contribute to several different firing patterns,
each of which typically accounts for only a small fraction
of its activity. Thus one finds many more multi-neuron
firing patterns than there are ganglion cells. We then
measured the visual receptive field of each firing pattern
by reverse correlation to a random flicker stimulus
(Meister et al. J. Neurosci. Methods 51, 95-106, 1994).
As a rule, the receptive field of a synchronous pair
of spikes from two cells was distinct from and smaller
than the receptive fields of each cell firing alone;
it generally fell into the overlap region between the
two individual receptive fields. The receptive fields
of firing patterns involving more than two neurons were
often even more sharply defined in space. We suggest
that the synchronous firing patterns are caused by shared
excitatory input from a presynaptic neuron, such as
an amacrine cell; spikes in this amacrine cell lead
to synchronous firing among all the ganglion cells it
feeds. In this view, each distinct firing pattern identifies
the activity of a distinct amacrine cell, and its receptive
field is simply the receptive field of that amacrine
cell.
Although
these neurons do not project fibers through the optic
nerve, the neural circuits receiving the optic nerve
signals could identify an amacrine cell's activity by
detecting its characteristic pattern of synchronous
firing among ganglion cells. In a sense, the signals
of amacrine cells might be "multiplexed" on top of the
ganglion cell signals. In this way, the brain could
obtain a representation of the visual scene of greater
spatial resolution than expected from the classical
single-neuron analysis of visual signaling.
To test this
idea quantitatively, we measured the information conveyed
by a small group of ganglion cells about a randomly
modulated visual stimulus (Warland and Meister, 1995)
. For this purpose we used a simple decoding algorithm,
based on linear filtering of the spike trains, to reconstruct
the visual scene from the ganglion cell responses. Pairs
of cells that engaged in coincident firing often conveyed
independent information about the visual stimulus, even
though their receptive fields overlapped to a great
extent. It was found that the optimal decoder of these
spike trains assigned a different visual message to
coincident spikes over single spikes, thus improving
the reconstruction significantly. We conclude that the
full meaning of the message the retina sends to the
brain can only be recovered by considering the concerted
firing patterns among ganglion cells. In particular,
multi-neuronal firing patterns can encode spatial information
distinct from that conveyed by individual neurons.
Adaptation
of the retinal code
Work over
the coming year will focus on the dynamic alterations
in retinal function in response to the visual environment.
For example, it is well known that the retina adapts
to changes in the mean intensity of the visual scene.
Clearly this is advantageous because of the large daily
changes in overall illumination. On the other hand,
we also encounter visual environments that differ strongly
in other image statistics, such as the mean contrast
(difference between dark and bright regions), or the
spatial correlation (the size of a typical blob of uniform
intensity). Does the retina adapt its function dynamically
to these image statistics? These questions are being
pursued vigorously by Stelios Smirnakis and Michael
Berry, a post-doc who recently joined the lab. They
have already yielded intriguing results (Smirnakis et
al., 1995a, 1995b) , and will further our understanding
of plasticity in neural circuits.
Warland,
D. K., and M. Meister. 1995. "Multi-neuronal firing
patterns among retinal ganglion cells encode spatial
information". Invest. Ophthalm. Vis. Sci. (Suppl.).
36: 932.
Smirnakis,
S. M., D. K. Warland, W. Bialek, and M. Meister. 1995a.
"Tiger salamander retina adapts to temporal contrast
modulation to improve coding efficiency". Invest.
Ophthalm. Vis. Sci. (Suppl.) 36: 624.
Smirnakis,
S. M., M. J. Berry, D. K. Warland, W. Bialek, and M.
Meister. 1995b. "Dynamics of Adaptation to Changing
Spatial Structure in the Tiger Salamander Retina". Soc.
Neurosci. Abstr. 1995. 21:1644.
