Sleep is required for improving reaction times after training on a procedural visuo-motor task.

Sleep has been found to enhance consolidation of many different forms of memory. However in most procedural tasks, a sleep-independent, fast learning component interacts with slow, sleep-dependent improvements. Here, we show that in humans a visuo-motor saccade learning task shows no improvements during training, but only during a delayed recall testing after a period of sleep. Subjects were trained in a prosaccade task (saccade to a visual target). Performance was tested in the prosaccade and the antisaccade task (saccade to opposite direction of the target) before training, after a night of sleep or sleep deprivation, after a night of recovery sleep, and finally in a follow-up test 4 weeks later. We found no immediate improvement in saccadic reaction time (SRT) during training, but a delayed reduction in SRT, indicating a slow-learning process. This reduction occurred only after a period of sleep, i.e. after the first night in the sleep group and after recovery sleep in the sleep deprivation group. This improvement was stable during the 4-week follow-up. Saccadic training can thus induce covert changes in the saccade generation pathway. During the following sleep period, these changes in turn bring about overt performance improvements, presuming a learning effect based on synaptic tagging.