Cerebellar motor learning is suggested to be caused by long-term plasticity of excitatory parallel fiber-Purkinje cell (PF-PC) synapses associated with changes in the number of synaptic AMPA-type glutamate receptors (AMPARs). However, whether the AMPARs decrease or increase in individual PF-PC synapses occurs in physiological motor learning and accounts for memory that lasts over days remains elusive. We combined quantitative SDS-digested freeze-fracture replica labeling for AMPAR and physical dissector electron microscopy with a simple model of cerebellar motor learning, adaptation of horizontal optokinetic response (HOKR) in mouse. After 1-h training of HOKR, short-term adaptation (STA) was accompanied with transient decrease in AMPARs by 28% in target PF-PC synapses. STA was well correlated with AMPAR decrease in individual animals and both STA and AMPAR decrease recovered to basal levels within 24 h. Surprisingly, long-term adaptation (LTA) after five consecutive daily trainings of 1-h HOKR did not alter the number of AMPARs in PF-PC synapses but caused gradual and persistent synapse elimination by 45%, with corresponding PC spine loss by the fifth training day. Furthermore, recovery of LTA after 2 wk was well correlated with increase of PF-PC synapses to the control level. Our findings indicate that the AMPARs decrease in PF-PC synapses and the elimination of these synapses are in vivo engrams in short-and long-term motor learning, respectively, showing a unique type of synaptic plasticity that may contribute to memory consolidation.long-term depression | high-voltage electron microscope | Golgi staining I mage stabilization in the visual field via the vestibulo-ocular reflex and optokinetic response requires accurate extraocular muscle synergies that rely on long-term plastic calibrations in the cerebellar flocculus (FL) and its downstream target vestibular nuclei (VN) (1-8). Long-term depression (LTD) in parallel fiber-Purkinje cell (PF-PC) synapses has been postulated as a possible mechanism for this plastic calibration based on many lines of mutant mice that lack both LTD and learning (9-12). However, LTD's role in motor learning has been recently questioned by a few mutant mice lines (13) and mice with pharmacological treatments (14) that showed lack of LTD but no impairment of learning. Furthermore, long-term potentiation in PF-PC synapses has been also shown to be involved in the motor learning (15). Recent evidence indicates that various forms of synaptic plasticity works synergistically and can compensate each other when one is missing in cerebellar motor learning (16). Despite the apparently contradictory results, no direct evidence for the decrease or increase of synaptic AMPA receptors (AMPARs) has been shown in physiological motor learning. To elucidate in vivo neuronal substrates for motor learning in wildtype mouse, we examined individual PF-PC synapses using quantitative SDS-digested freeze-fracture replica labeling (SDS-FRL) (17) combined with morphometric EM analysis after adaptation of ho...