Jonathan D. Victor and Keith P. Purpura
Department of Neurology and Neuroscience,
Cornell University Medical College.
We examined how the temporal pattern of spike trains produced by neurons in the visual cortex of the macaque represent simple visual stimuli. The analysis was based on ``spike time'' metrics, an approach which is motivated by the idea that neurons may act as coincidence detectors. Distances between spike trains, as determined by a ``spike time'' metric, are not Euclidean, but rather are determined by the minimal ``cost'' required to transform one spike train into another via a sequence of allowed elementary steps. The allowed elementary steps consist of inserting a spike (cost = 1), deleting a spike (cost = 1), and shifting a spike in time (cost = q/unit time). Spike trains which are close in the sense of a spike time metric have a similar number of spikes, and the times of occurrence of these spikes are small in comparison to 1/q. The value of q which maximizes stimulus-dependent clustering measures the temporal precision (but not the efficiency) of coding.
This approach was applied to single-unit and multi-unit activity recorded in the parafoveal representation of V1 and V2 in the awake monkey and the anaesthetized paralyzed monkey. Stimuli were presented transiently and varied in combinations of contrast, orientation, spatial frequency, check size, and texture. Evidence for temporal coding was widespread. The precision of temporal coding varied systematically with the attribute being encoded: temporal precision was highest for contrast (optimal q ca. 100, corresponding to a timing jitter of ca. 10 msec) and lowest for texture type (optimal q ca. 10, corresponding to a timing jitter of ca. 100 msec). For values of q within this range, multidimensional scaling of the responses, with distances derived from the spike time metric, yielded a ``response space'' in which two or three stimulus attributes defined independent, systematic trajectories.
We introduce the notion of a ``consensus response,'' which is a spike train that minimizes the distance to all of the observed responses to a particular stimulus. Consensus responses typically had fewer spikes than any of the observed responses. Nevertheless, the response space constructed from these sparser ``consensus responses'' was similar to that constructed from the full set of observed responses.
In the anaesthetized and paralyzed monkey, we also explored the extent to which temporal coding enabled representation of stimulus attributes in the face of variations of spatial phase. Simple and complex cells were able to transmit information about a single attribute (contrast, spatial frequency, or orientation) under these circumstances. However, complex cells, to a much greater extent than simple cells, could also transmit information about pairs of stimulus attributes, in a manner which was not confounded by spatial phase variations. This suggests that complex cells have a richer repertoire of temporal firing patterns, perhaps owing to a wider variety of inputs. These signalling capacities are largely developed within the first 100 msec after stimulus onset, which suggests that they are due primarily to the cortical inputs and local circuitry of V1.