Fig 1.
Ganglion cell spike trains during random bar motion.
A: Position of the bar as a function of time. B: Example of one stimulus frame; motion is perpendicular to the bar (red arrow). Ellipse fitted to the spatial receptive field profile of one representative ganglion cell (pink). C: Spiking activity of 180 cells in the guinea-pig retina in response to a bar moving randomly with the trajectory shown in A. Each line corresponds to one cell, and the points represent spikes. The order of the cells along the y-axis is arbitrary. D: Spiking activity of 123 cells in the salamander retina responding to a bar moving randomly. Same convention as C.
Fig 2.
High-accuracy reconstruction of the bar’s trajectory.
A: Schematic of the linear decoding method, here for 4 cells. A temporal filter is associated with each cell. Each time the cell spikes, its filter is added to the ongoing reconstruction at the time of the spike. The filters are optimized on part of the data to have the lowest reconstruction error and then tested on the rest of the data. B: Top: prediction of the bar’s position (black) from the activity of 123 cells in the salamander retina versus the real trajectory (red). Bottom: prediction of the bar’s position (black) from the activity of 178 cells in the guinea pig retina versus the real trajectory (red). C, D: Decoding performance plotted against the number of cells in the salamander (C) and guinea pig (D). Gray points correspond to random subsets of cells, black to the average performance. E: Histogram of the average decoding performance across all experiments using either causal (black) or acausal (white) decoding filters; results shown for the entire recorded neural population in both the salamander (left) and guinea pig (right).
Fig 3.
The prediction error is in the hyperacuity range for salamander retina.
A: Squared error as a function of frequency (red), compared to the power spectrum of the trajectory (blue). B: Root mean-squared error as a function of position (red) with the point of minimal error (pink circle). C: Error spectrum as a function of frequency (pink), for the position labeled in B. Power spectrum of the full trajectory (blue); error spectrum corresponding to the spacing between cone photoreceptors (dashed line).
Fig 4.
Decoding based on the neural image in the salamander retina.
A: Neural image in response to the moving bar. Color plot: neural image of the ganglion cell’s population activity at each point in time. White points: most likely position of the bar inferred from the peak in the neural image. Real trajectory in red. B: Population firing rate summed over all the cells as a function of time, for the same time window than A. C: Prediction of the bar’s trajectory using the linear decoding (black); real trajectory (red).
Fig 5.
Coding for motion in the receptive field surround in the salamander retina.
A: Schematic of the experiment: the bar is randomly moved at three different average locations relative to the array. B, C, D: Prediction of the bar’s trajectory using the linear decoding (black); real trajectory (red), for the three average locations above. E: Performance of linear decoding (blue) for individual ganglion cells (dots) plotted as a function of the distance between their receptive field center coordinate and the average bar position; probability distribution of bar position (black). Blue line: average decoding performance as a function of the distance. F: Performance of linear decoding (blue) for individual ganglion cells (blue dots) plotted as a function of the normalized distance between their receptive field and the average bar position (see text). Blue line: average decoding performance as a function of the normalized distance.
Fig 6.
Responses to motion in the receptive field surround in salamander retina.
Responses are shown for two example cells, cell 1 (first column) and cell 2 (second column). For each cell, PSTH of the response to the same trajectory is plotted when the trajectory is displayed over the receptive field center (blue, A and B); in the near surround (green, C and D); and in the far surround (red, E and F). G, H: average time course of the bar speed before a spike for different positions of the trajectory (color code matches that in A-F). I, J: Decoding filter of the same cells for different positions of the trajectory (same color code)
Fig 7.
Redundancy of the retinal code.
A, B: Information rate obtained when decoding with different subsets of ganglion cells (dots) plotted against the number of cells (A) or the sum of the individual information rate of each cell of the subset (B); color scale indicates the number of cells in a subset. C, D: Plots analogous to panels A, B but for the guinea pig retina. E, F: Information rate for the best (blue) and worst (red) subsets of salamander ganglion cells (see text) plotted against the number of cells (E) or the sum of the individual information rate of each cell of the subset (F).
Fig 8.
A: Comparison of the norm of the filters of each cell for the linear decoding (blue) against the regularized L1 decoding (red). The dashed line separate the 10 cells with the highest regularized norm. B: Decoding performance of groups of 10 cells sorted by their L1 norm, from the highest to the lowest. C: Number of cells in each subset that could reach a decoding performance of 0.85.
Fig 9.
Redundancy of cells of the same type.
A: Ellipses fitted to the spatial receptive field profile for the biphasic OFF cells. Scale bar: 100 microns. B: Temporal profiles of the fast biphasic OFF cell type. C: Plots analogous to Fig 7B. Information rate obtained when decoding with different subsets of ganglion cells (dots) plotted against the sum of the individual information rate of each cell of the subset; subsets are either picked randomly (blue), or from the biphasic OFF cells (red). D: Average information rate plotted against the sum of the individual information rate of each cell of the subset; subsets are either picked randomly (blue), or from the biphasic OFF cells (red).