Mind's eye: Saccade-related evoked potentials support visual encoding in humans.

In active vision, the brain receives and encodes discontinuous streams of visual information gated by saccadic eye movements. Saccadic modulation of neural activity is hypothesized to evolve to support perception and memory; however, it remains unclear whether this mechanism exist in humans, and potential functional roles have not been determined. We used eye tracking and intracranial local field potentials recorded from invasively monitored epilepsy patients when they performed visual encoding tasks, and observed consistent evoked potentials following saccades (saccade-related evoked potentials, SREPs) across the cerebrum. These SREPs were not attributable to ocular muscle activity or retinal input. Their magnitudes were not explained by spatial proximity to eye muscles or saccade eccentricity, and their polarity bore no relationship to saccade direction. Instead, the phase of pre-saccadic oscillation aligned with saccade timing and dissociated SREPs with positive polarity from those with negative polarity. Spatiotemporal profiling revealed that SREPs emerged earliest and with the greatest magnitude in the temporal lobes. We developed a saccade-related neural dynamic (SRND) model that characterized pre-saccadic oscillatory activity and SREP at each electrode contact location using finite features. Random forest models trained with these features achieved 62.6% balanced accuracy for predicting next-day recognition (long-term memory). Using the Shapley value, a framework for explaining machine learning models, we identified an SREP profile, characterized by earlier latency and larger magnitude, which was associated with successful visual encoding. In contrast, predicting saccade direction using the same SRND model performed at chance level, indicating that the observed SREP is less likely a corollary discharge signaling saccadic motor copy. These findings demonstrate SREPs as a neural mechanism of saccadic modulation with a role in human visual encoding.