ON TEMPORAL CODES AND THE SPATIOTEMPORAL RESPONSE OF NEURONS IN THE LATERAL GENICULATE-NUCLEUS

1. The present work relates recent experimental studies of the tempora l coding of visual stimuli (McClurkin, Optican, Richmond, and Gawne, S cience 253: 675, 1991) to the measurements of the spatiotemporal recep tive fields of neurons within the lateral geniculate of primate. 2. We analyze both new and previously described magnocellular and parvocell ular single units. The spatiotemporal impulse response function of the unit, defined as the time-resolved average firing rate in response to a weak stimulus flashed at a given location and time, is characterize d by the singular value decomposition. This analysis allows one to rep resent the impulse response by a small number, two to three, of spatia l and temporal modes. Both magnocellular and parvocellular units are w eakly nonseparable, with major and minor modes that account, respectiv ely, for similar to 78 and 22% of the response. The major temporal mod e for both types is essentially identical for the first 100 ms. At lat er times the response of magnocellular units changes sign and decays s lowly, whereas the response of parvocellular units decays relatively r apidly. 3. The spatiotemporal impulse response function completely det ermines the response of a unit to an arbitrary stimulus when linear re sponse theory is valid. Using the measured impulse response, combined with a rectifying neuronal input-output relation, we calculate the res ponses to a complete set of spatial luminance patterns constructed of ''Walsh'' functions. Our predicted temporal responses are in qualitati ve agreement with those reported for parvocellular units (McClurkin, O ptican, Richmond, and Gawne, J. Neurophysiol. 66: 794, 1991). Under th e additional assumptions of Poisson statistics for the probability of spiking and a plausible background firing rate, we predict the perform ance of a unit in the Walsh pattern discrimination task as quantified by mutual information. Our prediction is again consistent with the rep orted results. 4. Last, we consider the issue of temporal coding withi n linear response. For stimuli presented for fixed time intervals, the singular value decomposition provides a natural relation between the temporal modes of the neuronal response and the spatial pattern of the stimulus. Although it is tempting to interpret each temporal mode as an independent channel that encodes orthogonal features of the stimulu s, successively higher order modes are increasingly unreliable and do not significantly increase the discrimination capabilities of the unit .