Cortical Neuroprosthesis Merges Visible and Invisible Light Without Impairing Native Sensory Function

Discriminating IR light with visual cortex. A, Learning curve for IR-discrimination task using V1 (n = 6 animals). Note animals typically surpassed criterion (85% correct) within the first day. B, Direct comparison of learning curve for the same task with V1 implants (blue) and S1 implants (green; n = 15 for S1 implants). C, Same information as in B, but with learning curve superimposed when an animal has just a single IR sensor on its head (single-sensor learning data are from a previous study [Thomson et al. 2013]; n = 4 animals). D, Trial-based moving average analysis of performance in IR discrimination task. The curve shows percentage correct as a function of trial number (moving average of 20 trials) from same data shown in A. E, Same analysis as in C, but with S1-implanted animals shown in green for direct comparison. F, Performance in IR-discrimination task as a function of angle θ between the ports (see Fig. 1A). Includes data from V1 and S1 implanted animals with four IR sensors, as well as data from S1-implanted animals with a single IR sensor for comparison (blue, green, and red lines respectively with mean ± SEM percentage correct). G, Performance of naive V1- and S1-implanted animals (mean ± SEM percentage correct), that were not pretrained on a visual discrimination task (solid lines). Data from S1-implanted animals that were pretrained on a visual discrimination task are included for comparison. There is no significant difference, among any of the three groups, in the number of sessions it takes to reach 85% correct in the task (p > 0.05, two-tailed t test).

Neuronal responses to intracortical microstimulation in V1. A, Example of reduction of four-element full representation of electrical stimulation vector (four-element vector where each number represents frequency of microstimulation at a different location) down to two-dimensional population vector. There are five examples from five different time points T1–T5. B, Geometric representation of the population vectors from A. C, Stimulus population vectors from all trials in a session, with mean of first and last three substimuli overlaid in red. The black diamond outline is the convex hull of stimulus population vectors, showing the perimeter of the set of all possible vectors. Note that only a fraction of this set is actually presented in a given behavioral session. See text for more details. D, Neuronal data from four representative units. PSTHs show mean response to first three and last three microstimulation pulses, averaged over all trials in the session (note there are different axis limits for the different PSTHs). Associated with each PSTH is an IR-RF, a contour plot that shows the mean number of spikes in response to each stimulus population vector, with hot colors representing high firing rates. E, Distribution of all 240 IR-RF centers from all units recorded: note they tend to concentrate at the two medial corners of the stimulus space, although these are fairly rare events (C). F, Polar count histogram illustrates the medial, rather than anterior, concentration of IR-RF angular distribution. G, H, Same as E and F, but from data from S1 from a previous study (Hartmann et al. 2016). Note the anterior distribution of the IR-RF centers, much closer to the distribution of the stimuli.

Visual-IR integration task. A, Schematic of the task. In a cylindrical chamber with the same ports shown in Fig. 1A, four stimuli are randomly assigned to the ports. The target stimulus is the one with both visual and IR lights on. The IR distractor has only an IR light on, the visual distractor only a visual light, and there is one port with no light activated. B, Learning curve from three animals trained on the visual-IR integration task (implants in V1). The discontinuous right portion of the curve shows the last four sessions using an optimized frequency (see text). C, Comparison of learning curves between V1- and S1-implanted rats in the visual-IR integration task (n = 3 animals for each group). D, Same data as in C, but with trial-based moving average (moving window average taken with 20 trials).

Effects of S1 prosthesis on whisker discrimination. A, Aperture-width discrimination task. After the movable bars making up aperture move to the desired position, the rat enters the stimulus chamber, sweeps its whiskers across the aperture (movable bars controlled by the computer), and activates the central nose poke (CNP). The animal then retreats into the stimulus chamber and pokes in the right or left reward port depending on the aperture width. B, Psychometric curve from a single session in the 14-width aperture-width discrimination task. The points are the mean proportion of left responses at each width; the red line is the best minimum least-squares fit to the points, fitted using a Weibull function. C, As in B, but in an animal with worse performance (lower maximum slope of the curve and lower peak performance at each end of the spectrum). D, Performance of stimulated and matched control animal in the whisker discrimination task during different stages of the experiment. Before stimulation (Baseline), while learning the IR discrimination task via ICMS (Stimulation), and after whiskers were clipped (Clipped). See text for details. E, Mean ± SEM change in performance after the two transitions illustrated in D. ΔStim, transition from baseline to stimulation; ΔClip, transition from normal stimulation to stimulation with whiskers clipped. This analysis is restricted only to 54/78-mm widths, and includes eight animals (four stimulated/four control for the Δstim transition) and seven animals (four stimulated/three control) for the Δclip transition (one control animal stopped performing the task after whisker clipping). F, Same as E, but for the full multiple-aperture with discrimination task (either 12 or 14 widths). This includes six animals for the first transition and five animals for the second (as before, one control animal stopped behaving after whisker clipping). Percentage correct is calculated as the integral under the best Weibull fit to the psychometric curve (see Methods). G, Change in behavioral sensitivity, or the measure of the maximum slope of the best fit curves to the psychometric data, with each transition in the task. Lower slope means less behavioral sensitivity.