Abstract
Processing in early visual areas during sleep after visual training plays an important role in facilitating visual learning (e.g., Tamaki et al., 2020, Nat Neurosci). However, the extent to which early visual areas are involved in processing sleep itself remains unclear. If the activation in an early visual area is largely related to a sleep status such as the depth of sleep, the activation in the area alone should predict the sleep depth to a significant degree. To test this hypothesis, we recorded functional magnetic resonance imaging (fMRI) activation patterns in V1 while participants were asleep (90 min) simultaneously with polysomnography in multiple sessions and examined whether sleep depth, shown as a sleep stage, is decodable from the multivoxel fMRI patterns in V1. The middle frontal gyrus (MFG) was used as a control region, since MFG is thought to originate slow waves, which are EEG waves that occur specifically during sleep, and thus to be sensitive to sleep depth. A binary sleep-stage classifier on multivoxel fMRI patterns was constructed to classify fMRI patterns from each V1 and MFG into wakefulness and NREM stage-2, a sleep stage whose onset is often regarded as a sleep onset. Subjects’ being in wakefulness and NREM stage-2 were determined by polysomnography independently from fMRI patterns. To avoid the circular analysis (Kriegeskorte et al., 2009), we trained the sleep-stage classifier based on the dataset obtained on the second (or third) day and applied the classifier to the dataset on the first (or second) day. We found that the classification accuracy from fMRI patterns in V1 was significantly higher than the 50% chance level, but not from those in MFG. These results indicate that the activation in V1 alone discriminates an important sleep stage from wakefulness, suggesting a great involvement of V1 in sleep depth.