UC San Diego

Epiretinal prostheses are designed to restore vision to people blinded by retinal degenerations, using electrical stimulation with an array of electrodes implanted on the surface of the retina to convey artificial visual signals to the brain. Current clinical prostheses provide limited visual function, in part because the activity that they generate is different from natural retinal responses to visual stimuli. An ideal retinal prosthesis would stimulate the retinal ganglion cells (RGCs) in a way that veridically recreates the precise, cell-type dependent spatiotemporal spiking patterns present in the retina. This requires RGC activation at the native temporal and spatial resolution of the retina. We sought to better understand how electrical stimulation with high-density electrode arrays can be used to control RGC activity using an in vitro retina preparation to simultaneously stimulate and record from primate RGCs. The five numerically dominant RGC types, the ON and OFF midget, ON and OFF parasol, and small bistratified cells, could be activated to fire single, precisely-timed spikes using brief current pulses that fell within established safe charge density limits. Single-cell activation, which would prevent the simultaneous activation of different RGC types, was possible in roughly half of the tested RGCs. We further investigated whether simultaneous patterned current injection through multiple electrodes could be used to improve spatial selectivity. The majority of RGC responses to simultaneous current injection through multiple electrodes could be described by a simple model in which currents from different electrodes combine linearly to produce RGC responses. Observed nonlinearities could be effectively captured by extending the model to include multiple regions of stimulus space, in which each region contained a different linear tradeoff between currents passed through the electrodes. The usefulness of the model for improving spatial selectivity was examined by using it to predict which combinations of currents would most selectively activate a cell. The selectivity of the pattern chosen using the model was much higher than with single-electrode stimulation, and was nearly identical to that of the optimal pattern. Collectively, these findings support the feasibility of using epiretinal stimulation to control retinal activity at or near its native spatial and temporal resolution, and suggest a strategy to improve spatial resolution beyond what can be achieved using individual electrodes