S.M. Smirnakis, D.K. Warland, M.J. Berry, and M. Meister
Physics Dept. and Medical School, Harvard University
MCB Dept., Harvard University
How much does the temporal fine structure within a spike train contribute to the neural code? We stimulated isolated tiger salamander retinae with diffuse flicker and recorded ganglion cell spike trains extracellularly with a multi-electrode array. The light intensity was chosen randomly every 30 ms from a Gaussian distribution with photopic mean intensity and 35% root-mean-square contrast. Ganglion cells tended to fire tight bursts with a variable number of spikes (1-6 under our stimulation conditions). Within a burst, action potentials were separated by a few milliseconds, little more than the cell's refractory period. By contrast, different bursts were separated by several hundred milliseconds. By correlating the stimulus with bursts of different sizes, we found that bursts had qualitatively different visual response properties from those of single spikes. Assigning different messages to bursts of different sizes in the context of a linear decoding scheme (Bialek et al., Science 252, 1854), results on average, in a 16% increase in the information encoded by the ganglion cell spike trains (stdev = 12%). This suggests that it might be advantageous for higher visual areas to be able to distinguish between bursts of different sizes.
Note: see related abstract at NIC 97 by Berry et al. Supported by a Markey Scholarship and a grant from HFSP to MM and by an HST Research Assistanship to SS.