Simultaneous recording in the cat's retina and lateral geniculate nucleus (LGN) was used to find excitatory inputs to LGN cells. These recordings, correlated with measurements of LGN cell receptive-field properties, suggested new functional subdivisions of LGN cells. Distinctions between lagged and nonlagged cells were described before (Mastronarde, 1987a,b; Mastronarde et al., 1991), classification of nonlagged cells is examined here. The X(S)-type relay cells described before (Mastronarde, 1987a, b) each had detectable excitatory input from only one retinal X cell. Cells that received significant input from more than one retinal X cell were of three kinds: relay cells with pure X input (X(M)); relay cells with mixed X and Y input (X/Y); and cells that could not be antidromically activated from visual cortex (X(I)). In the series of relay cells, X(S)-X(M)-X/Y-Y, cells had progressively larger receptive-field centers, lower spatial resolution, and faster and more Y-like responses to various stimuli. X(I) cells resembled X(M) and X/Y cells in some respects but tended to have higher maintained firing rates, more sustained responses, and weaker surround suppression of the center response. The distinctness of X(S), X(M), X/Y, X(I), and Y from each other was examined with a modification of discriminant analysis that allowed cells to lack measurements for some parameters. Any given pair of categories could be distinguished reliably with only three parameters, although less so for X/Y-Y. In particular, X(I) cells were distinguishable from relay cells by properties other than the results of cortical stimulation, thus supporting the identity of X(I) cells as a separate class of X interneurons. Two discontinuities in the behavior of retinal input suggest that X(M) cells are a separate class from X(S) and X/Y cells: (1) LGN X cells received either no detectable input from any of the retinal X cells adjacent to their main input, or an easily detectable amount from several such cells; and (2) cells received either no Y input or a certain minimum amount. No such discontinuity in input underlies the distinction between X/Y and Y cells. LGN Y cells were also heterogeneous. Those with substantial input from more than one retinal Y cell had larger receptive fields and a greater preference for fast-moving stimuli than did Y cells dominated by a single input. Three Y cells could not be antidromically activated. They tended to differ from Y relay cells and resemble X interneurons in several ways. These shared properties, and the general reliability of cortical stimulation for nonlagged cells, indicate that the cells were Y interneurons. The strength of excitatory input extrapolated to zero at a separation between LGN and ganglion cell receptive fields equivalent to the radius of a retinal X axonal arbor for X input to X(M), X(I), and X/Y cells, or to the radius of a Y arbor for Y input to X/Y and Y cells. Thus, a retinal axon appears to be selective in providing input primarily to cells with somata within its arbor, rather than to all cells with overlapping dendrites. Coverage, the number of receptive-field centers overlapping a single point, was estimated for each kind of LGN cell described here. Each had a coverage of at least 6, comparable to that of retinal Y cells; most kinds had coverages of 15-35. These estimates support the idea that these subdivisions of LGN cells are functionally significant. X(M) and X/Y cells fill in the functional gap that is present between retinal X and Y cells and make the distribution of spatial properties more continuous, while multiple-input Y cells broaden the range of spatial properties. One role of LGN circuitry might thus be to provide a substrate for the correspondingly broad and continuous range of spatial-frequency tuning in the visual cortex.
