next up previous
Next: Peter E. Latham Up: No Title Previous: Alexei Koulakov and Dmitri

Gabriel Kreiman (1), Fabrizio Gabbiani (1), Walter Metzner (2), and Christof Koch (1)

tex2html_wrap_inline239CNS, Caltech, tex2html_wrap_inline241Biology, UC Riverside

Robustness, variability and modeling of amplitude modulation encoding by P-receptor afferent spike trains of weakly electric fish

P-type tuberous receptors in weakly electric gymnotiform fish encode changes in electric field amplitude. Single receptor afferents transmit accurate information about random amplitude modulated stimuli in a behaviorally relevant range(<100 Hz) of AM frequencies (Wessel R., Koch C., Gabbiani F. 1996, J. Neurophys. 75:2280-2293).

Random amplitude modulated electric fields were presented to curarized fish while recording extracellularly from P-receptor afferents. Reconstruction techniques (Bialek W., Steveninck R., Warland D. 1991, Science 252:1854-1857) were used to estimate the stimulus from the spikes. By adding temporal jitter to the spikes, we explored the timing precision required for encoding. The robustness to time jittering depended on the stimulus cut-off frequency (tex2html_wrap_inline253). The amount of jitter required for a 50% drop in coding fraction was 3 ms for high tex2html_wrap_inline253 (88Hz) but it was bigger than 20 ms for low tex2html_wrap_inline253 (less than 10 Hz). Similar results were obtained by random deletions and additions of spikes.

By repeating the same 'frozen' stimulus 10 times, we analyzed the time windows over which the spikes are reproducible using spike metrics (Victor J.D., Purpura K.P. 1996, J.Neurophys. 76:1310-1326). Maximum differences between the spike trains were observed at temporal resolutions between 3 and 40 ms. At low firing rates, the reproducibility increased with the bandwidth of the stimulus. We compare the variability and encoding with that of simple phenomenological models of the P-receptors. The experimental spike distances are close to those obtained in integrate and fire model neurons with random thresholds following a gamma distribution of order between 3 and 10.

We also computed the information gain upon averaging the responses of an afferent over several repetitions of the same frozen noise. The fraction of the stimulus encoded increased up to 40% as compared to that of individual trials and saturated when averaging was performed over more than four or five spike trains. This suggests that variability might be used by targets of P-receptor afferents to increase information about the stimulus by averaging the responses of a small number of afferent fibers.

Supported by NSF and ERC.

next up previous
Next: Peter E. Latham Up: No Title Previous: Alexei Koulakov and Dmitri

Tony Zador
Sat Mar 27 10:58:21 PST 1999