Columnar coding of neuronal populations in primary visual cortex

Roger Heriskstad(1), Jonathan Baker(2), Charles M. Gray(2), Shih-Cheng Yen(1)

(1) Department of Electrical and Computer Engineering, National University of Singapore
(2) Center for Computational Biology, Montana State University

    In this study, we used a 54-channel columnar probe to record from populations of neurons within a cortical column in the primary visual cortex of the anesthetized cat. To efficiently sort the data, we used the specific geometry of the recording probe to assign compound signals to one or more spike trains. We applied this method to recordings in which natural movies and drifting grating stimuli were presented.

    The first step in our algorithm was to separate a signal into channel complexes. We defined a channel complex as a continuous range of channels for which the activity exceeded the background noise level. Through simulations, we found that complexes originating from a single cell exhibited a distinct alternating and decaying pattern of spike amplitudes across channels caused by the specific arrangement of the channels on the probe. This allowed us to label complexes as single-cell or multi-cell complexes based on the channel profile. Similar single-cell complexes were then merged together based on the number of common channels, as well as the similarity between mean waveforms, to form template clusters. The multi-cell complexes were then matched against individual templates or pairs of templates using the correlation coefficient computed across the active channels of the multi-cell complex. We also made sure that adding the multi-cell complex did not create inter-spike intervals smaller than 1 ms.

    The above approach allowed us to identify a total of 393 cells across five sites in one experimental animal, with a median of 91 cells per site. Using the responses to drifting gratings, we defined a cell as tuned if the response distribution at the stimulus direction containing the highest firing rate was significantly different from the response distribution at the perpendicular orientation. By this criteria, 149 cells were defined as tuned. We also computed the maximum firing rate for all cells across 40 ms bins, which was the duration of a movie frame in our movie stimuli. The grating stimuli evoked significantly stronger responses (2-sample KS-test (KS-test2), p < 0.01), with median and quartiles of 20,10 and 30 Hz for the movie responses, and 25, 20 and 40 Hz for the grating responses. We also calculated the lifetime and population sparseness of the cells. Surprisingly, we found that the grating stimuli evoked higher lifetime and population sparseness (KS-test2, p<0.01 for both), with median and quartiles of 67.8, 44.9, 98.7 (lifetime) and 0.49, 0.35 and 0.66 (population) for the movie responses, and 99.8, 99.8 and 99.9 (lifetime) and 0.93, 0.87 and_0.97 (population) for the grating responses. An analysis of the correlation between the PSTH of all possible pairs of cells revealed that the responses to movie and grating were quite heterogenous, with median and quartiles of 0.016, 0.0017 and 0.069 for movies, and 0.0, 0.0, and 0.0038 for gratings. The grating responses were again significantly more heterogenous (KS-test2, p < 0.001).

    Our preliminary results suggest that the responses of neuronal populations in a cortical column exhibit significantly different responses under grating and movie stimulation.