The Cerebellum: A Neuronal Learning Machine? Jennifer L. Raymond, Stephen G. Lisberger, Michael D. Mauk' Comparison of two seemingly quite different behaviors yields a surprisingly consistent picture of the role of the cerebellum in motor learning. Behavioral and physiological data about classical conditioning of the eyelid response and motor learning in the vestibulo- ocular reflex suggest that (i) plasticity is distributed between the cerebellar cortex and the deep cerebellar nuclei; (ii) the cerebellar cortex plays a special role in learning the timing of movement; and (iii) the cerebellar cortex guides learning in the deep nuclei, which may allow learning to be transferred from the cortex to the deep nuclei. Because many of the similarities in the data from the two systems typify general features of cerebellar orga- nization, the cerebellar mechanisms of learning in these two systems may represent principles that apply to many motor systems. The work of Brindley, Marr, Albus, and Itu (I, 2) made the idea that the cerebellum is a primary site of motor learning intc one of the most appealing hypotheses of crrehellar function. Their general hypothesis has been supported by lesion, electrical stimulation, and recording studies in a variety of move- ment systrms (3-5). However, thc rxact function of the cerebellum in movanmt and its specific role in Icarning have re- mained controversial. To determine whethcr and haw the cer- ebellum participates in motor learning, it is necessary to establish cause-and-effect rela- tions between laming of motor responses and changes in the responses of cerebellar and extraccrehellar neurons. Much prugrcss toward this goal has been made for two forms of motor learning that require an intact cerebellum: classical conditioning of the eyelid rcsponse and motor learning in the vestibulo-ocular reflex (VOR). Here, on the basis of physiological and behaviriral data from these twc movement systems, wc outline a set of unifying principlcs of cere- bellum-dependent learning. Basic Cerebellar Circuitry The anatomy and physiology of the cerebel- lum are remarkably tegrilar over different cer- ebellar regions and ate highly conserved across species. These Facts suggcst that the cerebellum performs the 5dme general compu- tation for rnany diffcrcnt motor (and perhaps nonmotor) tasks. Because the anatoniy and physiology of the cerebellum are so central to J L Raymond IS in the Department ot Physiolopy and the W. M Keck Foundation Center tor 1ntegrati;e Newm science, University of Calitornia, San Francisco. CA 94143, USA. S. G. Lisbeiaer IS in the Deoaltment of Physiology. the W. M. KeciFoundation Center tor lnte~ grative Neuroscience, and the Sloan Center for Theow ical Neurobiology, University ot Calitornia, San Francisco. CA 94143. USA. M. D. Mauk is in the Deualtment of thought about sites of plasticity and mecha- nisms of motor learning, we begin with a brief review of cerebellar organization (6). The entire cerebellum shares a common architecture (Fig. 1). The two major ana- tomical compartments of the cerchellum are thc ccxtrx and thc deep nuclei. Purkinje cells, the only outputs from the cerebellar cortex, project through inhibitory connec- tions to the deep cerebellar nuclei, which provide the outputs to other brain regions. Inputs are transmitted to the cerebellum over climbing fibers and mossy fibcrs, two pathways with fundamentally different physiulogy and anatomy. The climbing-fibcr input to thc cercbcl- lum arises from the inferior olivary nuclei. In the cerebellar cottex, each Purkinje cell receives monosynaptic inputs from just one climbing fiber, and each climbing fiber projects to about 10 Purkinje cells. The climbing fibcrs cause Purkinje cclls to emit complex spikcs, which occur at ratcs of jut one or a few per second. The infrequent Fig. 1. Schematic of the ba- sic cerebellar circuit, which is iterated throughout the structure. A mossy fiber - granule cell + parallel fiber input (red), an inferior olive -climbing fiber input (blue), and some, but not all. of the intricate connections of the inhibitory interneurons (gray) are shown. Climbin! occurrence of the coinpl~x spikrs is n compatible with traditional rdfe codes for information transfer and has led to the sug- gcstion that climbing fibers are involved in guiding motor learning and in keeping time for movement coordination (1, 2, 5, 7) The mossy-fiber inputs arise from a va- riety of brainstem nu& as well as from the spinal cord, and they influrnce Pur- kinje cell firing through a web of inter- neurons in the cerebellar cortex. Mossy fibers synapse on granule cells, which in turn form parallel fibers and make excita- tory contacts on numerous Purkinje cells as well as on inhibitory interneurons. The inhibitory interneurons synapse on Pur- kinje cells and also provide inhihitory feedback to the granule cells. Because of the massive convergence and divergence in the connections from granule cells and inhibitory interneutons onto Purkinje cells, the mossy fibers affect Purkinje cell firing through pathways that offer many upportunities fur both spatial and tcmpo- ral integration. In contrast to the complex spikes caused by the clirnbing-fihcr inputs to Purkinje cells, the simple spikes drivcn by the mossy-fiber inputs firc at ratcs as high as 100 per second and prohahly use a frequency code to transinit information. The best recognized actions of climbing fibers and inossy fibers are their inputs tO the cerebellar cortex, hut axon collaterals from both inputs alsrr project to the deep ccrchcllar nuclci. Thus, thc ccrchcllum contains parallcl pathways for affcrcnt in- formation, one through the cerehellar cor- A Cerebellar cortex 'arallel fiber Granule cell L Excitation + Inhibition nucleus 1126 SCIENCE * VOL. 272 24 MAY 1996rex and one directly through the deep nu- clei. The signals transmitted through the deep nuclei and the signals transmitted through the cerebellar cortex are trans- formed by different intervening neural net- works, which likely perform different com- putations and mediate different functions. Common Behavioral Properties of Eyelid Conditioning and Motor Learning in the VOR In classical conditioning of the eyelid re- sponse (4, 8), a puff of air serves as a