Multiplexing of visual information by spike rate and latency encoding in retinal ganglion cells
Jutta Kretzberg1, Insa Winzenborg1, Imke C.G. Reimer1, Leon M. Juarez Paz1, Martin Greschner2 and Andreas Thiel1
1University of Oldenburg, Oldenburg, Germany 2The Salk Institute for Biological Studies, La Jolla, CA, USA

The activity of retinal ganglion cells (RGCs) is the only source of information about the visual environment for the brain. Hence, the output signals of these cells must simultaneously transmit information about many dimensions of visual stimuli like light intensity, velocity etc. Moreover, several stimulus properties are changing on fast time-scales in natural environments, demanding for fast mechanisms of encoding.

Since the timing of spikes in response to stimulus changes depends strongly on the stimulus, response latency could provide a fast coding mechanism. However, unlike the experimenter, the nervous system cannot determine latency by using the information at which time the light stimulus changed. In this study, we show that transients in RGC population activity could be used as reference points to determine response latency, allowing the decoding of certain stimulus features.

We performed multi-electrode recordings from RGCs in the turtle retina. The data was analyzed with two complementary approaches, Bayesian stimulus reconstruction [1] and metric-based clustering [2].

We used two types of stimuli, i.e. slow full-field intensity modulation flicker and a moving spatial pattern, which pseudo-randomly changed speed and direction. In both situations, instantaneous stimulus changes induced transient peaks in the RGC population activity. When applying Bayesian reconstruction, several stimulus properties could be reconstructed better and faster based on response latencies of individual RGCs relative to transients in population activity than they could when spike rates were used. While relative latencies allowed better reconstruction of light intensities, temporal contrasts and changes in motion velocity, spike rates were found to be superior to estimate constant velocities.

Metric-based clustering supported these findings. This method also suggested the encoding of constant motion velocity by spike rates, because all time windows longer than 100 ms were equally well suited to estimate this property. In contrast, the estimation of velocity changes showed a clear optimum for shorter time windows, hinting towards temporal encoding.

Moreover, these findings agree with our previous result of discriminant analysis, that the combination of spike rates and relative latency improves stimulus estimation significantly [3].

In summary, we conclude that RGCs could use multiplexing of spike rates and response latencies to transmit information about several different properties of visual stimuli.

[1] Zhang et al. J Neurophysiol 79:1017-1044 (1998).
[2] Victor & Purpura, Network: Computat Neural Syst 8:127-164 (1997).
[3] Greschner et al., J Neurophysiol 96:2845-2856 (2006).