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 has been observed in nonhuman primates and is hypothesized to support perception and memory; however, its functional role in humans remains unclear. In our study, we tested the hypothesis that saccadic modulation of neural activity in humans predicts successful visual encoding. For this purpose, we measured eye gaze and local field potentials in intracranially-monitored patients while they performed visual encoding tasks. We observed consistent brain-wide evoked potentials following saccades (saccade-related evoked potentials, SREPs). 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 oscillations 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 intracranial recording site using a finite set of 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, that 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 that SREPs are a direct result of saccadic modulation of neural activity and that they contribute to human visual encoding.