A new functional role for cerebellar long-term depression.

Long-term depression (LTD) of excitatory synaptic transmission has been described in several regions of the brain (Artola and Singer, 1993; Linden, 1994). In most cases, the opposite process, i.e. long-term potentiation (LTP) (Bliss and Collingridge, 1993), is present at the same synapses. Most theories on the functional role of LTD in such neurons assume that LTD acts by resetting synapses that have undergone LTP (Tsumoto, 1993; Linden, 1994). In the cerebellum, however, LTD of the parallel fiber to Purkinje cell synapse (Ito, 1989) is distinguished by the absence of LTP at this synapse (Artola and Singer, 1993). This form of LTD has been extensively investigated in the context of a specific theory of the function of the cerebellum, first proposed by Marr (Marr, 1969) and subsequently elaborated by Albus (Albus, 1971) and Ito (Ito, 1984). However, the conditions under which cerebellar LTD can be induced in vitro do not seem to be compatible with some predictions of the Marr-Albus-Ito theories (Schreurs and Alkon, 1993). I will first review and criticize the Marr-Albus-Ito theories and then describe the properties of Purkinje cells and of LTD induction that are relevant to a new hypothesis on the function of cerebellar LTD that I have proposed recently (De Schutter, 1995a). Synaptic circuitry of the cerebellar cortex A remarkable property of the cerebellum is the uniformity of its cortical architecture (Ito, 1984). Purkinje cells receive two types of excitatory inputs: climbing fibers (CF) from the inferior olive and parallel fibers (PF) from cerebellar granule cells. Each Purkinje cell receives only one CF input, which evokes a powerful voltage-gated Ca2+ spike in the dendrite (Llinás and Sugimori, 1980; Knöpfel et al., 1991; Miyakawa et al., 1992), called the complex spike (CS). In contrast, 150,000 to 200,000 PF synapses (Harvey and Napper, 1991) contact each Purkinje cell, and generate conventional dendritic EPSPs. Purkinje cells receive inhibition from basket cells and stellate cells. Both types of inhibitory neurons are excited by PF axons and therefore provide pure feedforward inhibition. Except for the sparse Purkinje axon collaterals (Bishop, 1982; Ito, 1984), there are no feedback connections between Purkinje cells and the rest of the cerebellar cortex. The feedforward excitation and inhibition in the molecular layer of the cerebellum is in sharp contrast with alloand neocortex, where pyramidal cells receive massive feedback excitation and inhibition (Shepherd, 1990). The Marr-Albus-Ito theories Since Hebb (1949) first proposed that a synaptic modification based on the co-occurrence of preand postsynaptic activity might underlie learning, it has generally been assumed that synaptic plasticity plays a fundamental role in memory. An example is the theory of motor learning by the cerebellar cortex first proposed by Marr (1969) more than 25 years ago, which is an influential example of trying to deduce neural function from structure. Marr proposed that PF synapses onto Purkinje cells are facilitated when they are activated together with the CF. In this theory the cerebellum learns motor skills by storing memory traces at the PF synapse under instruction of the olivary nucleus, which would signal correct performance. A few years later, Albus (1971) refined this theory by suggesting that the cerebellum functions as a modified perceptron (Minsky and Papert, 1969) pattern-classification device, with the complex spike as the unconditioned stimulus and mossy fiber (MF) input as the conditioned stimulus. Albus also proposed that the CF stimulus is an error signal and that the PF synapse is weakened instead of facilitated. This last hypothesis became known as the Marr-Albus theory (Fig. 1), which predicted the existence of LTD before any plasticity of synapses onto Purkinje cells was found. The theory was further extended by Ito in his flocculus hypothesis (Ito, 1984)