INFORMATION RATES CONVEYED BY GROUPS OF INDIVIDUAL NEURONS IN PRIMARY VISUAL CORTEX

Daniel S. Reich^1,2

Ferenc Mechler²

Jonathan D. Victor²

¹The Rockefeller University, New York

²Weill Medical College of Cornell University,

Department of Neurology & Neuroscience, New York

We extend the “direct method” (Strong et al., 1998) to estimate information rates in the responses of groups of simultaneously isolated, nearby single neurons. The approach is limited—by the feasibility of data collection—to the analysis of typical information rates in short time windows. We present the methods by which we correct for biases that arise because real data sets are limited in size. We make tetrode recordings of the activity of multiple single neurons in primary visual cortices (V1) of anesthetized macaque monkeys. We analyze the responses to three types of stimulus: high-contrast flickering checkerboards, drifting gratings, and flashed (stationary) gratings. We compare two different codes for information transmission by multiple neurons. The first—a “summed-population code”—considers only the average spike count across groups of neurons in each time bin. The second—a “labeled-line code”—keeps track of the neuron of origin of each spike. For each code, we estimate two quantities: the rate of information transmission and a “redundancy index,” which is a normalized measure of the degree to which nearby neurons convey independent information. Since nearby neurons in V1 are tuned to similar stimulus features, we might expect to find significant redundancy. Instead, we find that the information rates are nearly independent when analyzed by the labeled-line code, but more redundant when analyzed by the summed-population code. This is the case for all three types of stimulus. The difference between the information rates for the two codes increases with the number of neurons being analyzed. Thus, reading out the activity of many neurons with a summed population code is likely to underestimate the total information content.