Packet information encoding in a cerebellum-like circuit.

Abstract

Packet information encoding of neural signals was proposed for vision about 50 years ago and has recently been revived as a plausible strategy generalizable to natural and artificial sensory systems. It involves discrete image segmentation controlled by feedback and the ability to store and compare packets of information. This article shows that neurons of the cerebellum-like electrosensory lobe (EL) of the electric fish Gymnotus omarorum use spike-count and spike-timing distribution as constitutive variables of packets of information that encode one-by-one the electrosensory images generated by a self-timed series of electric organ discharges (EODs). To evaluate this hypothesis, extracellular unitary activity was recorded from the centro-medial map of the EL. Units recorded in high-decerebrate preparations were classified into six types using hierarchical cluster analysis of post-EOD spiking histograms. Cross-correlation analysis indicated that each EOD strongly influences the unit firing probability within the next inter-EOD interval. Units of the same type were similarly located in the laminar organization of the EL and showed similar stimulus-specific changes in spike count and spike timing after the EOD when a metal object was moved close by, along the fish's body parallel to the skin, or when the longitudinal impedance of a static cylindrical probe placed at the center of the receptive field was incremented in a stepwise manner in repetitive trials. These last experiments showed that spike-counts and the relative entropy, expressing a comparative measure of information before and after the step, were systematically increased with respect to a control in all unit types. The post-EOD spike-timing probability distribution and the relatively independent contribution of spike-timing and number to the content of information in the transmitted packet suggest that these are the constitutive image-encoding variables of the packets. Comparative analysis suggests that packet information transmission is a general principle for processing superposition images in cerebellum-like networks.