I. INTRODUCTION (HISTORICAL BACKGROUND)

Top Previous Next References

Studies of the cerebellum have a long history well documented in the monographs of Jansen and Brodal (247) and Moruzzi and Dow (127). The elaborate structural organization of the cerebellum analyzed by Jansen and Brodal (247) encouraged researchers toward exploration of its functional meanings. The enormous data of classic lesion experiments compiled by Dow and Moruzzi (127) suggested characteristic functional features of the cerebellum as enabling us to learn to move smoothly and accurately even at high speeds and without visual feedback. These earlier studies paved the way for the modern cerebellar research to proceed in the directions to uncover the elaborate neuronal mechanism of the cerebellum and to define its roles in the entire nervous system function. Brief history presented here overviews the progress of cerebellar research made during the past four decades toward these directions.

Until around 1960, synaptic plasticity was a concept of only theoretical importance, since there was no experimental evidence for it. Hebb (195) postulated that, in an assembly of mutually interconnected neurons, synapses activated synchronously, both presynaptically and postsynaptically, are strengthened in their transmission efficacy. Eccles (133), investigating posttetanic potentiation in the spinal cord, assumed that presynaptic activation alone is sufficient to potentiate synaptic transmission. Based on a set theory that includes possible combinations of presynaptic and postsynaptic events, Brindley (70) proposed the presence of 10 different types of synaptic plasticity in the central nervous system, including the Hebb and Eccles types. It is noteworthy that Brindley included long-term depression (LTD) under the designation "habituation," giving equal weight to potentiation and habituation. He also predicted synaptic plasticity in inhibitory synapses, which was not known to exist until very recently (see sects. VC5 and VC8). To demonstrate the computational capability of a neuronal assembly, Rosenblatt (454) constructed a three-layered artificial neuronal network capable of learning, called the "Simple Perceptron," which incorporated synaptic plasticity in one layer.

In the 1960s, the neuronal circuit of the cerebellum, involving five types of neurons [Purkinje cells (PCs), basket, stellate, Golgi and granule cells], was dissected in detail by morphological and microelectrode techniques, as summarized by Eccles et al. (134). PCs were revealed to receive two distinct excitatory inputs: climbing fibers (CFs) and axons of granule cells (GAs) (135-137) (Fig. 1). CFs originate from the inferior olive, while GAs relay mossy fibers (MFs) originating from the brain stem and spinal cord. Basket, stellate, and Golgi cells were identified as inhibitory neurons. The PCs providing the sole output pathway of the cerebellar cortex were defined as exclusively inhibitory upon their target neurons in the vestibular and cerebellar nuclei (244-246).

View larger version (26K): [in this window] [in a new window]   Fig. 1. Connections in the cerebellar cortex drawn by Ramon y Cajal. A part of Figure 104 of Cajal (74) is shown with the right, left axis reversed to match Fig. 6. A, mossy fiber; B, Purkinje cell axon; a, granule cell; b, parallel fiber; c, Purkinje cell; d, climbing fiber. Arrows indicate the supposed directions of signal flow. [Modified from Cajal (74).]

Convergence of numerous GAs and a single CF to each PC is a characteristic unique feature of neuronal circuitry in the cerebellum (Fig. 1) well known since Cajal (74). Brindley (69) was the first who viewed this structure as representing the Hebbian type of synaptic plasticity. Because CF signals via the numerous junctions between a CF and a PC (see sect. IIA1) could regularly excite PCs, synchronous activation of the CF and GAs results in a conjunction of presynaptic and postsynaptic excitation, which might in turn lead to long-term potentiation (LTP) of the GA-PC synapses. Marr (359) adopted this view in developing his epochal network theory of the cerebellar cortex. In view of the stable operation of the system, however, Albus (15) suggested that the conjunction results in a depression instead of potentiation and conceived the neuronal circuit in the cerebellar cortex as a pattern separator like the Simple Perceptron.

In the 1970s, two lines of evidence suggested the presence of a synaptic plasticity in the cerebellum. We observed that in the flocculus, an evolutionarily old part of the cerebellum, MF pathways arising from the vestibular organs and CF pathways arising from the retinas converge onto PCs (157, 224, 351). This unique pattern of convergence suggests that a synaptic plasticity occurs in the flocculus as causal to the remarkable adaptability of the vestibuloocular reflex (VOR), and the observation made in PCs of rabbits flocculus favored Albus' hypothesis (128, 166, 167, 230). Gilbert and Thach (170) found that, in a monkey adapting to compensate for a sudden change in arm load, PCs changed their discharge patterns in the manner predicted from Albus' synaptic plasticity assumption (sect. VIA5).

Nevertheless, earlier efforts devoted to demonstrating the occurrence of synaptic plasticity by conjunctively stimulating GAs and CFs failed to reveal a significant change in GA-PC transmission, which may be attributed to the following two major reasons. The first reason appears to be the small extracellular field potentials representing GA-PC transmission. LTP had successfully been detected in the hippocampus by recording synaptic events extracellularly (57); however, because the field potentials in the cerebellar cortex are ~10 times smaller than those in the hippocampus, minor changes could have been undetected. A later study using an averaging technique revealed a depression (by ~25%) in the field potentials of the in vivo cerebellum (234) and cerebellar slices (88). The second reason could be the vulnerability of LTD to various factors that might block its induction (see sects. III and IV).

In the early 1980s, an analysis of the firing index of PCs in in vivo cerebella of rabbits revealed the presence of LTD (138, 241). When the stimulus for MFs or GAs was critically set at a threshold for exciting a PC, LTD was sensitively reflected in the lowering of the probability of firing in the PC. Later, LTD was also detected by recording GA-induced events in PCs either extracellularly (234) or intracellularly (459) and was eventually established as a distinct type of synaptic plasticity near the end of the 1980s (228). Different types of LTD were also reported to occur in the hippocampus and cerebral neocortex (65; see also Ref. 26).

Although neuronal network theories developed by Marr (359), Albus (15), and others help us to understand computational mechanisms of elaborate neuronal networks in the cerebellum, control system theories help to understand how the elaborate networks of the cerebellum are utilized for controlling movements. Considering the unique way of involvement of the cerebellum in control of VOR, the author suggested that the cerebellum makes feed-forward control possible by replacing the feedback loop and that this replacement is effected due to a learning mechanism referring to "control error" signals conveyed by CFs (223, 225). That CF signals represent errors was also suggested from two other viewpoints. Miler and Oscarsson (375) proposed that the inferior olive acts as a comparator for command signals from higher centers and the activity these signals evoke at lower levels. Albus (15) postulated that CFs convey teacher's signal informing PCs about their misperformance in conducting pattern recognition like the Simple Perceptron. Current evidence for the error representation by CFs is reviewed in section VIA.

Through the 1970s and 1980s, various simple forms of motor learning involving the cerebellum such as adaptation in ocular movements, eye-blink conditioning, locomotion, and hand/arm movements were established as model systems for studying mechanisms of motor learning. Thach (515) applied microelectrode techniques to the cerebellum of awake animals and collected signals of individual PCs correlated to these motor tasks that the animals performs. A morphological study of the cerebellum by Voogd and colleagues (178, 179) led to a remarkable finding that, in addition to the classic lobular structure, the cerebellum consists of a longitudinal compartmental structure including seven major zones (A, B, C1, C2, C3, D1, and D2). Oscarsson (422) electrophysiologically defined a small, functionally uniform longitudinal area, called microzone, as a module of the cerebellar cortex.

Efforts were devoted earlier to understand global functional meanings of the unique anatomical architecture of the cerebellum such as the interconnection between the cerebellum and the cerebral cortex via thalamus, red nucleus, and pontine nucleus (17, 145, 223, 496). The unique cerebrocerebellar communication loop was interpreted as representing an internal model simulating the limb and its spinal motor centers (223, 227) (see sect. VE). These initial efforts are rewarded by the remarkable development of computational approach to the cerebellum in the 1990s (275, 276, 372, 558).

In the 1990s, owing to marked advancements in cellular physiology, biochemistry, and molecular biology, the complex signal transduction mechanisms of LTD have become a major research theme in neuroscience, as reviewed in sections III and IV. The advancement in knowledge of LTD signal transduction mechanisms provides new tools for investigating functional roles of LTD in various forms of cerebellar learning, as reviewed in section VI. In the 1990s, a new development in cerebellar studies was the expansion of cerebellar roles to cognitive functions. Such roles were proposed based on anatomical connections of the cerebellum with the cerebral association cortex (309, 310), and the author supported this view by analogies in movement and thought from a control system viewpoint (229). Experimental evidence for the involvement of the cerebellum in certain mental functions now accumulates in studies applying noninvasive measurements and advanced clinical examinations to a variety of human mental activities including language (146), attention (16), cognitive affective syndromes (466), fear and anxiety caused by threats of pain (436), thirst sensation and fear for air hunger (426), and motor relearning (221).

This review focuses on the major results of investigations conducted during the past two decades on the characterization of cerebellar LTD, its signal transduction mechanisms, and roles in cerebellar functions. Based on analyses of the available data, the review aims to clarify the targets of LTD studies in the forthcoming decades, in particular, the relationship of LTD with permanent memory and roles of LTD in not only physical but also mental activities.

	II. CHARACTERIZATION OF CEREBELLAR LONG-TERM DEPRESSION

Top Previous Next References

A.  Induction of LTD

1.  GA and CF synapses on PCs

GAs consist of segments ascending from the granular layer to the molecular layer and the bifurcating parallel fiber (PF) branches that run along the molecular layer (Fig. 1). A recent study revealed that PFs extend for 2-3 mm to each side of the bifurcation point, much longer than previously thought. Functional implication of this long PF projection is discussed in section VB2. Previously, PFs were assumed to provide the sole GA inputs to PCs, but Llinás (334) proposed a substantial contribution of the ascending axons to GA-PC transmission. Functional implication of the asending segments of GAs is discussed in section VB1.

While each PC was reported to receive 60,000-80,000 synapses on their dendritic spines, generally, one synapse from one PF (424), more recent data in rats indicate that each PC receives as many as 175,000 PF synapses (405, 406). In contrast, each PC comes into contact with only one CF via numerous discrete synaptic junctions formed on stubby dendritic spines (424). Multiple innervation of a PC by CFs occurs during early development or in abnormal conditions such as genetic deficiency (see sect. VIB4). The number of junctions formed between a CF and a PC has been calculated as ~300 in the frog cerebellum (335). In the rat cerebellum, the number of synaptic junctions formed per 100-µm length of PC dendrites is 11.45 for GAs and 1.7 for CFs (412). This may suggest that as many as 26,000 synaptic junctions are formed between a CF and a PC in rat. Since, however, CFs do not reach peripheral portions of PC dendrites (424), this would give an overestimate.

When recorded extracellularly in vivo, PCs generate two different types of spikes. Simple spikes discharge at a rate of 50-100 Hz, and complex spikes, in a form of short burst of spikes, occur at irregular, low rates around 1 Hz (515). Stimulation of GAs elicits simple spikes (Fig. 2A), whereas stimulation of CFs evokes complex spikes (Fig. 2B). When recorded intracellularly, PCs produce Na⁺ spikes and Ca²⁺ spikes (337, 338). The extracellularly recorded simple spikes are Na⁺ spikes generated in the somatic region and passively spread into the dendrites (Fig. 2C). This earlier notion has been supported by dual patch recording from somata and dendrites (500), and also using sodium-binding benzofuran isophthalate (SBFI) as a specific indicator for Na⁺ (77); the changes in intracellular sodium concentration ([Na⁺]_i) associated with antidromically or intrasomatically evoked Na⁺ spikes were confined to the cell somata. CF responses are large excitatory postsynaptic potentials (EPSPs) due to the numerous synapses made by a single CF on the PC dendrites, superposed with somatic Na⁺ spikes followed by smaller Ca²⁺ spikes (337, 338) (Fig. 2D).

View larger version (16K): [in this window] [in a new window]   Fig. 2. Electrical signals of Purkinje cells. Extracellular (A and B) and intracellular (C-F) recording from Purkinje cells (PCs) are shown. A: simple spikes induced by stimulation of parallel fibers (PFs) on the pial surface of a cerebellar folium. Two sweeps are superimposed. [From Eccles et al. (135). Copyright 1996 Springer-Verlag.] B: complex spike evoked by stimulation at the inferior olive. [From Eccles et al. (137).] C: Na+ spike evoked by stimulation of granule cells (GAs). [From Eccles et al. (136). Copyright 1996 Springer-Verlag.] D: climbing fiber responses composed of an excitatory postsynaptic potential (EPSP) and Na+ and Ca2+ spikes. [From Eccles et al. (137).] E: AMPA-EPSPs evoked by stimulation of GAs. [From Karachot et al. (265). Copyright 1994 Elsevier Science.] F: mGluR-EPSPs evoked by repetitive stimulation of GAs in the presence of an AMPA antagonist. [From Miyata et al. (381). Copyright 1999 Cell Press.]

GA impulses evoke two pharmacologically distinct types of synaptic potentials in PCs. One is mediated by DL--amino-3-hydroxy-5-methyl-4-isoxazolone-propionate (AMPA)-selective glutamate receptors (see sect. IIIB1) (Fig. 2E) and the other by metabotropic glutamate receptors (mGluRs) (sect. IIIB3) (Fig. 2F). AMPA-EPSPs are fast and distinctly evoked by individual GA impulses, whereas mGluR-EPSPs are slow and observable after a brief tetanus of GAs (8 pulses at 50 Hz) in the presence of an AMPA receptor antagonist (39, 40). The size of the AMPA-mediated excitatory postsynaptic currents (AMPA-EPSC) generated by a single GA-PC synapse was estimated in cerebellar slices to be 2-60 pA (34).

LTD in GA-PC transmission is indicated by persistent reduction of the firing index of a PC in response to GA stimulation in extracellular recording, the initial rising slopes of GA-evoked AMPA-EPSPs in intracellular recording (Fig. 3), or the size of GA-evoked AMPA-EPSCs (Fig. 4A) or spontaneously arising miniature EPSCs (mEPSCs) (Fig. 4C), in whole cell clamping.

View larger version (25K): [in this window] [in a new window]   Fig. 3. Long-term depression (LTD) observed in PCs in cerebellar slices. Inset: EPSPs induced by stimulation of GAs. Six traces of EPSPs recorded before (control) and 10-50 min after conjunctive CF/GA stimulation are superimposed. Each trace indicates EPSPs averaged for 5 sweeps repeated every 5 s. The graph plots the rising slopes of the EPSPs against time. Shaded column indicates conjunctive climbing fiber/GA stimulation. [Modified from Karachot et al. (267).]

View larger version (23K): [in this window] [in a new window]   Fig. 4. Reduced form of LTD in cultured PCs. Inset: diagram shows arrangement of two pipettes, one for recording (M) and the other for iontophoresis of glutamate (I) on a PC. A and B: glutamate-induced potentials in cultured PCs. A: 3 cases of glutamate-induced currents recorded before (1), 10 min after (2), and 40 min after (3) glutamate/depolarization conjunction, showing no depression (a), short-term depression (STD) (b), or LTD (c). B: plotting the amplitudes of glutamate-induced potentials against time before and after glutamate/depolarization conjunction. Perfusion with phospholipase (PL) A2 inhibitors changed LTD to STD. U73122, inhibitor of PLC. [A and B from Linden (322). Copyright 1995 Cell Press.] C and D: miniature excitatory postsynaptic currents (mEPSCs) recorded from cultured PCs. C: specimen records of mEPSCs before (a), 2 h after (b), and 168 h after (c) 5-min conditioning treatment with 50 mM K+ and 100 µM glutamate. D: changes of mEPSC amplitudes induced by 5-min treatments twice with a 30-min interfval. Each point represents data from 9-35 cells. [C and D from Murashima and Hirano (395). Copyright 1999 Society for Neuroscience.]

GA-PC synapses also exhibit homosynaptic LTD when a relatively large set of GAs is repetitively stimulated without involving CFs (188). However, as long as GAs are moderately stimulated, LTD occurs only after GA-CF conjunctive stimulation (265, 381). Homosynaptic LTD also occurs in CF-PC transmission during stimulation of CFs at 5 Hz for 30 s (184), but not during 1-Hz CF/GA stimulation used for inducing conjunctive LTD (265). Functional meanings of these types of homosynaptic LTD is discussed in section v, C1 and C4. When conjunctive or homosynaptic LTD occurs in a set of GA-PC synapses, LTD is also induced in the neighboring GA-PC synapses (445, 539). Functional meanings of this heterosynaptic type of LTD are discussed in section VC2.

The optimal condition for inducing conjunctive LTD in the cerebellum in vivo is a 100-pulse stimulation of CFs and GAs at 4 Hz with a CF-to-GA stimulus interval of 125-250 ms (139, 265). In cerebellar slices, simultaneous GA/CF stimulation at 1 Hz for 5 min (300 pulses) was optimal in the presence of a GABA_A antagonist (picrotoxin) (265). In the absence of a GABA_A antagonist, 600 pairings at 1 Hz of GA and CF stimuli within ±250-ms intervals were required for effectively inducing LTD, while 100 pairings were sufficient in the presence of the GABA_A antagonist (bicuculline) (88).

Although stimulation of GAs simultaneous with or after CF stimulation LTD is usually used for inducing LTD, whether a GA stimulus preceding a CF stimulus is effective in inducing LTD or not is often questioned. This question has been addressed in studies in which the GA/CF stimulation intervals were systematically changed (88, 139, 265). However, the results of such studies are inconclusive, because GA/CF stimulation intervals cannot be determined uniquely during repetitive GA/CF stimulation; for example, at 1-Hz stimulation, a GA stimulus preceding a CF stimulus by 100 ms follows the preceding CF stimulus with a 900-ms delay. Such a second-order effect may be reduced if the stimulus frequency is lowered. However, this strategy is not realistic because at a frequency lower than 1 Hz, conjunctive stimulation becomes less effective in inducing LTD (265). Unless an appreciable LTD can be induced by the conjunction of one GA stimulus and one CF stimulus, it is difficult to define a GA/CF time relationship.

To natural stimuli, GAs respond repetitively, and CFs often respond in the same way. Hence, there could be probabilities of overlap between a series of GA and CF impulses even when the MF- and CF-activating natural stimuli are not critically timed. It should also be kept in mind that GA and CF impulses produce complex signal transduction processes in PCs, as reviewed in sections III and IV, some of which could be sufficiently long lasting to provide a conjuction between GA-induced and CF-induced signal transduction. Schreurs et al. (468) successfully induced LTD in rabbit cerebellar slices by applying a series of 8 pulses (79 s, 100 Hz) to GAs and immediately thereafter 3 pulses (20 Hz) to CFs, 20 times at 30- to 40-s intervals. Wang et al. (539) also induced LTD in rat cerebellar slices by applying three to eight pulse stimuli to GAs followed by a single pulse stimulus to CFs within 50-200 ms. These GA-CF timing relationships are consistent with behavioral timing in motor learning (sect. VI, A2, B2, and C1). Although the reason why our previous study on rat cerebellar slices applying 15 pulses at 50 Hz to GAs and immediately thereafter 3 pulses at 50 Hz to CFs, repeated every 20 s for 10 min, failed to induce LTD (265) is unclear, the stimulus conditions for LTD induction need to be further explored in connnection with stimulus parameters and pharmacological environments, in which cerebellar tissues are placed.

Induction of GA/CF-induced LTD in the cerebellum in vivo was effectively blocked when the inhibitory postsynaptic potentials (IPSPs) were concomitantly induced in PCs by stimulation of off-beam PFs (138). This effect is apparently related to the shortening or abolition of long-lasting plateau potentials, which represent Ca²⁺ currents evoked by CF impulses. Imaging of intracellular Ca²⁺ concentration ([Ca²⁺]_i) with fura 2 demonstrated that IPSPs strongly reduced the CF-associated increase of [Ca²⁺]_i in PC dendrites (76). In cerebellar slices, this complication is usually avoided by blocking IPSPs with a GABA_A antagonist (picrotoxin or bicuculline). However, even in the absence of a GABA_A antagonist, GA stimulation with intermittent tetanic high-frequency effectively induced LTD (265, 484). In the cerebellum in vivo, LTD induction was not disturbed unless relatively large IPSPs were evoked by electrical stimulation (138). These observations refute the suggestion that the LTD induced in cerebellar slices treated with a GABA_A antagonist is not a physiological phenomenon.

B.  Features as Synaptic Plasticity

1.  Postsynaptic origin

Because the glutamate sensitivity of PCs undergoing LTD was persistently depressed after the conjunction of CF stimulation and ionotophoretic application of glutamate, GA/CF-induced LTD has been ascribed to a reduction in the efficacy of postsynaptic glutamate receptors mediating GA-PC transmission (241). That a reduced form of LTD (sect. IIB3) can be induced in cultured PCs in the absence of presynaptic elements confirms that LTD occurs entirely postsynaptically. Metabotropic glutamate receptors subtype 4 (mGluR4) are distributed in the molecular layer along the presynaptic membrane of GAs (284, 363). Mice deficient in mGluR4 exhibited impairment of the GA-induced paired-pulse facilitation and posttetanic potentiation in PCs, whereas LTD was not impaired (428). This indicates that LTD induction does not involve the mGluR4-regulated presynaptic mechanism for maintaining synaptic efficacy during repetitive activation. GAs express cannabinoid type 1 receptors (CBR1s) and their mRNAs, and selective CBR1 agonists markedly depress GA-PC transmission (313). CBR activation reduced transmitter release at GA-PC and CF-PC excitatory syanpses by modulation of presynaptic voltage-gated channels, and also at inhibitory synapses in PCs by not only presynaptic voltage-gated channel modulation but also suppression of the vesicle release mechanisms (505). CBR agonists also partially inhibited GA/Ca²⁺ spike-induced LTD (313), but how the presynaptic CBR1s are linked with LTD induction is unclear at present.

When the vestibular nerve on one side was stimulated in conjunction with stimulations of CFs at the inferior olive, LTD was observed in the flocculus PCs only in their responses to stimulation of that nerve, but not of the other vestibular nerve (241). When two PF beams were differentially stimulated at a distance of 100-200 µm in in vivo cerebellum (138, 256) or 300 µm in cerebellar slices (88) in the direction perpendicular to the PF beams, LTD was induced only in the PF beam involved in conjunctive stimulation with CFs. In cultured PCs, a reduced form of LTD (see below) occurred only on the part of the dendrite exposed to stimuli in the absence of presynaptic elements (324). These observations are indicative of the input specificity of LTD in PCs, but according to the recent studies introduced below, the input specificity of LTD does not strictly hold for those synapses located within ~100 µm from stimulated GA-PC synapses.

When conjunctive LTD was induced by stimulating a PF beam, test stimulation of a second PF beam at a distance of 36-109 µm from the first beam revealed spread of LTD to neighboring synapses despite the fact that they were not involved in the conjunctive stimulation (445). The spread still occurred even when the stimulation of the first PF beam was lowered to involve only ~20 PF fibers. Spread of LTD to neighboring synapses has also been demonstrated by locally applying glutamate to PC dendrites (539). Glutamate was released by uncaging with 3- to 5-µm diameter ultraviolet (UV) light spot. It was thus found that glutamate sensitivity of a PC dendrite was reduced for 100 µm from the activated synapses. The spread of LTD could be mediated by a chemical signal moving from the conjunctively activated synapses to their neighborhood, either extracellularly or intradendritically. Such a signal may be provided by one of the complex signal transduction processes underlying LTD such as nitric oxide (NO) (see sect. IIIA2). Functional implication of the reduced input specificity of conjunctive LTD is discussed in section VC2.

In the cerebellum in vivo, iontophoresis of glutamate or quisqualate, but not kainate or aspartate, to PCs, replacing GA stimulation, effectively induced a persistent reduction in sensitivity of PCs to glutamate, when combined with CF stimulation (241, 256). In cerebellar slices, replacement of CF stimuli by application of depolarizing pulses, which cause the entry of Ca²⁺ into PCs through voltage-gated channels, induced LTD when combined with GA stimulation (GA/depolarization-pairing at 1 Hz for 15 min) (104, 105). To obtain LTD with GA/depolarization-pairing, however, GA stimuli stronger than with GA/CF conjunction are required, suggesting an additional contribution of CF stimulation to Ca²⁺ entry (445). A protocol using 20 pairs of a brief GA tetanus combined with a subsequent 100-ms depolarization is also effective in inducing LTD (151).

Various chemical stimuli, related to signal transduction for LTD as reviewed in sections III and IV, induce LTD when combined with GA stimulation. For example, NO donors (484), membrane-soluble cGMP analogs (485), and protein phosphatase inhibitors (12) are effective. The simplest procedure for inducing LTD is to apply quisqualate (574) or to increase the extracellular K⁺ concentration (101). The grease-gap method was used for determining the chemical sensitivity of PCs in a cerebellar slice trimmed into a wedgelike form (39, 235, 236, 237, 574). Persistent reduction in AMPA-induced potentials of PCs was induced by AMPA application following perfusion of 8-bromo-cGMP, trans-1-aminocyclopentyl-1,3-dicarboxylate (trans-ACPD), a mGluR agonist, or sodium nitroprusside, a NO donor (236).

In cultured PCs devoid of both GAs and CFs, a reduced form of LTD is induced by a combination of glutamate (or quisqualate) pulses and membrane depolarization (328, 329) (Fig. 4B). The LTD then is detected as a reduced sensitivity of PCs to glutamate or AMPA. This form of LTD occurs in very simplified preparations devoid of dendritic spines, indicating that LTD induction does not require such morphological specialization (407). In PCs cocultured with granule cells, mEPSCs can also show LTD (395) (Fig. 4C).

GA/CF-induced LTD in cerebellar slices usually develops progressively over an hour (265, 267) (Fig. 3). In contrast, glutamate/depolarization-induced LTD in cultured (322) or freshly isolated (407) PCs or GA/depolarization-induced LTD in PCs cocultured with granule cells (542) reaches a steady peak within a few minutes (Fig. 4B). LTD in cultured PCs occurs in two phases. The short-term depression (STD) declines after the peak at ~5 min and recovers in 30 min and normally is replaced by the late-phase depression. Inhibitors of phospholipase (PL) A₂ (see sect. IIIE2) abolished only the late phase and left the STD in isolation (322). Apparently, STD is lacking in GA/CF-induced LTD, which would correspond to the late phase of glutamate/depolarization-induced LTD. STD might arise from the use of large depolarizations to replace CF stimuli.

The observation time for LTD in cerebellar slices and in the cerebellum in vivo is usually 30 min to 1 h, and occasionally for 2-3 h. In certain special cases, longer times have been reported. The quisqualate-induced persistent reduction in AMPA sensitivity of PCs, as revealed by the grease-gap method, was monitored for 12 h without recovery (235). The amplitude of mEPSCs in cultured PCs was persistently reduced after 5-min conjunctive application of 50 mM K⁺ and 100 µM glutamate, and this reduction lasted for 36 h and recovered to the original level after 48 h (395) (Fig. 4D). When the hemispheric area of the lobule simplex (HVI) of rabbit cerebellum was sliced 24 h after the rabbit had been trained for eye-blink conditioning, sequentially applied GA and CF stimuli failed to induce LTD, which occurred in slices dissected from control rabbits (469). This suggests that LTD underlying the eye-blink conditioning (sect. VIB2) persists for at least 24 h and precludes eliciting another LTD. An increased excitability of the PC dendrites in the rabbit lobule HVI was detected after eye-blink conditioning, even after 1 mo (467). This learning-specific excitability increase was presumably caused by changes in a K⁺ current, possibly mediated by an I_A-like current, but its relationship to LTD is presently unclear.

	III. SIGNAL TRANSDUCTION: INITIAL PROCESSES

Top Previous Next References

Section III deals with the initial steps of signal transduction for LTD involving first messengers, receptors, ion channels, G proteins, and phospholipases. Readers may also refer to recent review articles (53, 110, 312, 327).

In addition to glutamate that plays a major transmitter role in GA-PC synapses and probably also in CF-PC synapses, NO, corticotropin-releasing factor (CRF), and insulin-like growth factor I (IGF-I) are released from GAs or CFs and act on PCs as first messengers (Fig. 5).

View larger version (51K): [in this window] [in a new window]   Fig. 5. Signal transduction underlying LTD. The contents of sections III and IV are summarized. AMPAR, AMPA receptor; AAR, yet-unidentified amino acid receptor; ADA, arachidonic acid; 2R, 2-receptor; G-S, G-substrate; IEG, immediate early gene; RyR, ryanodine receptor. Other abbreviations are defined in the text. Broken lines indicate suggested, but not proven, relationships. [Modified from Ito (232).]

GAs contain and, upon stimulation, release glutamate (see Ref. 227). Because the specific antagonist for AMPA-selective glutamate receptors, 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX), blocks GA-PC transmission (298, 429), glutamate has been established as a transmitter from GAs. Release of glutamate from GAs in GA-PC transmission is controlled by Ca²⁺ entering PFs via multiple types of Ca²⁺ channels (377). Release of glutamate from GAs is also regulated by mGluR4s, CBR1s (sect. IIB1) and adenosine receptors coupled with adenylyl cyclase, cAMP, and protein kinase A (PKA) (sect. IVC5).

Glutamate transporters take up the glutamate released into the synaptic gap. Among the five structurally distinct types of glutamate transporters so far identified, EAAT4 is abundantly expressed in PCs and is concentrated on the extrajunctional membrane of dendritic spines in contact with the GAs (510). GLAST, on the other hand, is highly expressed in the Bergman glial processes that ensheath GAs to PC synapses (308). Since the time courses of GA-evoked EPSCs in PCs are normal in both GLAST-deficient (548) and EAAT4-deficient (512) mutant mice, these transporters do not appear to be the primary factors determining the kinetics of GA-evoked EPSPs.

Nitric oxide synthase (NOS) is of three types: neuronal (nNOS), epithelial (eNOS), and inducible (iNOS). The nNOS is abundant in the granule cells of the cerebellum (289, 291, 450); eNOS is also expressed in the granule cells at a lesser level (125). NOS immunoreactivity is also strong in basket cells, but not observed in PCs (141). The mRNAs harvested from single neurons using micropipettes encoded NOS from granule cells, which were absent in PCs (101). Nicotinamide adenine dinucleotide phosphate diaphorase (NADPH-d), which is a critical coenzyme of NOS, is abundant in the granular and molecular layers but not in PCs (494, 562). The molecular structure of nNOS, which is abundant in the brain, contains recognition sites for NADPH and calmodulin, as well as phosphorylation sites resembling those of cytochrome P-450 reductase (68). In cerebellar slices, NO is released by activation of GAs (483). The NO release from GAs is potentiated by the tetanic GA stimulation, as well as by activation of PKA in GAs (283).

The earlier assumption that NO might be released from CFs, or within PCs stimulated by CF signals, was based on the following observations. Rats treated with 3-acetylpyridine (3-AP) to lesion CFs showed a substantial reduction in electrical stimulus-induced NO release (484) or a K⁺-induced increase in cGMP concentration presumably caused via NO activation of guanylyl cyclase (sect. IIIB5) (494) in cerebellar slices. These effects are, however, transient and are not directly related to the loss of CFs; the cGMP levels in the rat cerebellum were depressed once with a peak on day 7 and thereafter returned to the control level in 14-20 days (421; see also Ref. 382).

There is substantial evidence showing that NO plays a role in LTD induction. 1) Bath application of a NO donor, sodium nitroprusside (484), or its infusion into PCs (109), effectively induced LTD. 2) The release of NO from its caged form within PCs induced LTD when combined with depolarization-induced Ca²⁺ entry (314). 3) Bath application of NOS inhibitors readily blocked LTD (103, 484). As would be expected from the location of NOS in GAs, the postsynaptic application of a NOS inhibitor did not block LTD (109). 4) Targeted disruption of the nNOS gene in mice resulted in the virtual loss of NOS activity in the cerebellum (217), and accordingly, GA/depolarization-conjunction did not induce LTD in cerebellar slices obtained from these mice (316). Photolytic uncaging of NO and cGMP inside PCs did not recover the loss of LTD, suggesting that prolonged absence of nNOS might lead to alteration of the signaling pathway downstream of cGMP. Thus the requirement of NO in LTD induction in cerebellar slices is evident, but it is to be noted that this is not the case in cultured PCs. NOS donors, scavengers, or inhibitors did not affect the reduced form of glutamate/depolarization-induced LTD in cultured PCs (326), and LTD in cultured PCs derived from nNOS-deficient mice was indistinguishable from that in cultures from wild-type mice (328).

In cerebellar tissues, sensitivity of NOS to Ca²⁺ is regulated by protein kinase C (PKC) (418). The compound 6R9-5,6,7,8-tetrahydro-L-biopterin (H₄B), a cofactor of NOS, binds to NOS to stabilize it against phosphorylation by PKC (420). Cerebellar tissues from those mice partially deficient in H₄B had significantly lower cGMP levels compared with the control (66). Yet, the increase in cGMP levels following application of a NO donor was normal, suggesting that impairment of cGMP synthesis in H₄B-deficient mice was due to decreased NO synthesis (see sect. IIIB5).

The compound NO is unique compared with the usual neurotransmitters, for it is a short-lived gas diffusing into the vicinity of its source. Calculations based on the diffusion constant suggest that NO diffuses over 14 µm in 10 ms (539). Since 1 µl of the molecular layer contains 7.24 × 10⁸ dendritic spines of PCs (406), a spherical space of the molecular layer, 14 µm in radius, would contain 4,136 spine synapses. Hence, NO released from one site in the molecular layer would influence 4,000 spine synapse by diffusion of NO in 10 ms. The mediator for heterosynaptic LTD is yet to be identified, but NO could be a candidate for it (445).

Aspartate and homocysteate were previously proposed as candidate transmitters released from CFs on the basis of the results of uptake and release experiments (534, 555), but only weak immunoreactivity to either aspartate or homocysteate was found in CFs (577). On the other hand, evidence suggests that glutamate is a transmitter from CFs. 1) CFs exhibit glutamate-like immunoreactivity at a significant level (577). 2) CF-PC transmission is effectively blocked by AMPA-specific antagonists such as CNQX (298). However, caution is needed because the excitatory action of homocysteate is also blocked by CNQX (572), and PCs express distinct aspartate receptors sensitive to glutamate antagonists (573). 3) Blocking glutamate transporters in PCs by infusion with D-aspartate through a whole cell clamp micropipette prolonged the decay of CF-evoked EPSPs (506). Because CF-EPSPs are normal in EAAT4-deficient mice, another transporter may be responsible for this effect (512). 4) A transmitter released from CFs activates both AMPA receptors and glutamate transporters expressed by the Bergman glial cells (130). Interestingly, CF transmitter at the same time presynaptically inhibited GABA-mediated inhibitory synapses on PCs (463).

Despite these lines of supportive evidence, the possibility of glutamate being a CF transmitter remains to be confirmed since the specific release of glutamate from CFs has not been proven.

CRF-like immunoreactivity has been observed in all divisions of the inferior olivary nucleus and at all levels, from the cells of origin of the olivocerebellar afferents in the inferior olivary neurons to their CF terminals on PCs (106, 425). A KCl challenge induces CRF release from rat cerebellar slices (55). CRF plays a permissive role in the LTD induction in cerebellar slices (sect. IIIB6). However, because CRF is likely to be absent in tissue cultures devoid of CFs, CRF may not be an indispensable factor for the glutamate/depolarization-induced reduced form of LTD in cultured PCs.

It is noted that besides the CFs, CRF is also distributed densely in the paraventricular nucleus of the hypothalamus, central nucleus of amygdala, and locus coeruleus, all of which are involved in stress responses. The functional meaning of the presence of CRF in CFs may be interpreted in connection with the role of CFs in motor learning (sect. VIA), since motor learning can usually be accomplished by repeated exhausting exercises. Motor learning may well be a kind of stress in a wide sense (238).

IGF-I, a basic peptide, is known to modulate cell growth and differentiation and to act as a modulator at diverse synaptic sites. Several lines of evidence indicate the presence of IGF-I in CFs. The inferior olive exhibits IGF-I immunoreactivity (4) and expresses IGF-I mRNAs (59, 551). Electrical stimulation of the inferior olive significantly increased the IGF-I levels in the cerebellar cortex (81). Chemical and surgical lesions of the olivocortical pathway produced a drastic decrease in cerebellar IGF-I levels (523). IGF-I antisense oligonucleotide injected into the inferior olive induced a significant reduction in the size of dendritic spines on PCs (412). There is now evidence that IGF-I plays a roles in LTD induction (sect. IIIB7). PCs have been reported to exhibit strong IGF-I immunoreactivity in the PC cell somata, dendrites, dendritic spines as well as axons, and in the rough endoplasmic reticulum (3, 4, 481). Relationships of the intracellularly localized IGF-I immunoreactivity to LTD induction are unclear for the present.

Activation of the four types of receptors by the first messengers released from GAs and CFs trigger diverse signaling processes in the membrane of PCs such as the activation of ion channels, G proteins, and associated phospholipases (Fig. 5). In addition, NO released from GAs diffuses into PCs and reacts with an enzyme. A unique feature of PCs, distinct from most other neurons including cerebellar granule cells, is the absence of N-methyl-D-aspartate (NMDA) receptors. Nevertheless, adult rat PCs intensely express the mRNAs for the subunit NR1, and the NR2A mRNA weakly (13). NR2D mRNAs are expressed in PCs transiently during the first week after birth. Mouse cerebellum also exhibited NR2A and NR2B immunoreactivity in PCs (519). These subunits apparently do not form NMDA receptors in adult PCs.

AMPA receptors mediate the fast GA-PC transmission and are the final target of the signal transduction for LTD (sect. IVE). AMPA receptors in mature PCs include mainly GluR2 and GluR3 and also GluR1 subunits (465). The mRNAs harvested from individual PCs encode the following five subunits: the flip and flop versions of GluR1 and GluR2 as well as GluR3 flip, with GluR2 being the most abundant (304). Receptors labeled with antibodies against GluR2/3 or GluR2 are localized in the synaptic zone of dendritic spines, with a decrease in receptor density in the outer 20% of the synapse (431). GluR1 is also present in the postsynaptic membrane of GA-PC synapses (43).

Activation of AMPA receptors appears to be one of the requirements for LTD induction, because the combination of GA/Ca²⁺ spikes failed to induce LTD when applied in the presence of CNQX (198). Nevertheless, because LTD can be induced without stimulating AMPA receptors in cases such as conjunctive activation of mGluR with membrane depolarization (108, 198) (sect. IIIB3), activation of AMPA receptors may not be an indispensable requirement for LTD induction.

Although AMPA receptors are typical ionotropic receptors associated with ion channels, evidence has been presented that AMPA receptors are also associated with chemical signal transduction like the metabotropic receptors. The GluR1 subunit in AMPA receptors of cortical neurons is associated with G_i protein (541). Activation of AMPA receptors in cerebellar neurons results in activation of protein tyrosine kinase (PTKs) (193), which also has a role in LTD induction (sect. IVC3).

The ₂-subtype of glutamate receptors is selectively expressed in PCs, while low levels of ₁-subtype are found widely in the adult brain (22, 344). Immunogold labeling revealed the presence of _1/2-subunits of glutamate receptors in GA-PC synapses with a distribution pattern similar to AMPA receptors (305, 431). In the dendritic spines of PCs, -receptors are anchored to actin filaments via spectrin, an actin-binding protein (205, 206).

A role for ₂-receptors in LTD induction is suggested by two major findings. 1) Treatment of cultured PCs with an antisense oligonucleotide against the ₂-subunit mRNA blocked LTD induction but had no appreciable effect on the basic physiological or morphological properties of PCs (208, 250). 2) PCs in cerebellar slices (269) and tissue cultures (209) derived from ₂-deficient gene-knockout mice did not exhibit LTD.

The ₂-receptors do not form functional glutamate-gated ion channels. However, when an A (alanine)-to-T (threonine) point mutation is introduced at position 654, the end of its third transmembrane segment, as occurs in lurcher mutant mice, and when a Q (glutamine)-to-R (arginine) mutation is also induced at the so-called Q/R site in the second transmembrane segment, the modified homomeric ₂-receptors display ion channel activity including a moderate Ca²⁺ permeability in the absence of ligand binding (293). A factor(s) that may activate ₂-receptors, a ligand, a receptor subunit or associated messengers have yet to be found for understanding how ₂-receptors contribute to LTD induction.

Eight subtypes of the mGluR family are classified into three groups: I (mGluR1 and -5), II (mGluR2 and -3), and III (mGluR4, -6, -7, -8) (98, 401). Group I subtypes are characterized by their stimulating phosphatidylinositol hydrolysis and Ca²⁺ release from intracellular stores, whereas the group II and III subtypes negatively regulate cAMP formation. Several of these subtypes have multiple splice variants. So far, four isoforms of mGluR1 (a-d) have been detected in PCs (49, 177). Cultured PCs exhibit immunoreactivity for mGluR1a in both somatic and dendritic regions (173, 410). In the mouse cerebellum, 80% of all PC dendritic spines express mGluR1a immunoreactivity. The expression was retained even when presynaptic GAs were lost by intoxication with methylazoxy methanol (504). In the dendritic spines of PCs, both mGluR1a and mGluR1b are localized in the perisynaptic areas outside the postsynaptic densities (346, 362, 431). While mGluR1a dominates within 60 nm from the edge of the postsynaptic densities, mGluR1b distributes more evenly to over 360 nm (362).

The involvement of mGluR1 in LTD induction has been demonstrated in the following experiments. 1) LTD induction in cerebellar slices was blocked by an antagonist of mGluR1, (RS)-a-methyl-4-carboxyphenylglycine (MCPG) (185). 2) Antibodies inactivating mGluR1 blocked the reduced form of LTD in cultured PCs (487). 3) LTD was impaired in mGluR1-deficient gene-knockout mice (6, 100). The transgenic mice whose mGluRs were rescued only in the cerebellum restored normal LTD (220). 4) Activation of mGluRs by the agonists trans-ACPD or 1S,3R-aminocyclopentyl-dicarboxylate (1S,3R-ACPD) induced LTD when combined with GA tetanus (102) or depolarization-evoked Ca²⁺ spikes, even in the presence of CNQX (108, 198).

Activation of mGluR1 evokes mGluR-EPSPs in PCs (39, 40). The mGluR-EPSPs induced by repetitive stimulation of GAs (usually 8 pulses at 50 Hz) have a slow time course when observed in the presence of an AMPA antagonist (peak time, 200-300 ms). Even though mechanisms for generating mGluR-EPSPs have not yet been identified, Tempia et al. (514) suggest that it is due to activation of a nonspecific cation channel. In fact, generation of mGluR-EPSPs is associated with an increase in intradendritic Na⁺ concentration (288). An alternative possibility is the activation of an electrogenic Ca²⁺/Na⁺ exchanger that extrudes Ca²⁺ released from intracellular stores by the activation of mGluR1s (sect. IVA2). However, because KB-R7943, a selective inhibitor of the Ca²⁺/Na⁺ exchanger, only partially depressed the mGluR agonist-evoked inward currents, mGluR-EPSPs may have only a minor contribution from the Ca²⁺/Na⁺ exchanger (210).

In CF-PC synapses, all three subunits of the AMPA receptors (GluR2, GluR3, GluR1) and mGluR1 are distributed in a similar manner to GA-PC synapses (43). The ₂-receptors are expressed in both GA-PC and CF-PC synapses during developmental stages, but after postnatal day 21 and in adulthood, ₂-receptors are scarce in CF-PC synapses, while they are predominant in GA-PC synapses (579, 431). The report that L-homocysteinic acid generates substantial responses in cultured PCs (572) may draw attention because of the earlier proposal that L-homocysteate might be a transmitter (sect. IIIA3), but these responses are mediated by AMPA receptors.

Receptors in CF-PC synapses are blocked by CNQX (29) similarly to AMPA receptors in GA-PC synapses (290). However, these two types of synapses differ in their sensitivity to another antagonist, a synthetic analog of Joro spider toxin, 1-naphthyl-acetyl-spermine (NAS). NAS blocked the GA-PC synapses in cerebellar slices but not the CF-PC synapses (11). It is noted that, in cultured hippocampal neurons, NAS differentially blocked AMPA receptors associated with a strong inward rectification and permeability to Ca²⁺, but not those associated with a slight outward rectification and low Ca²⁺ permeability (294). In recombinant glutamate receptors expressed in Xenopus oocytes, NAS differentially blocked AMPA receptors composed of GluR1, GluR3, GluR4, or GluR1/3, but not those composed of GluR1/2, GluR2/3, or GluR6 (56). These results would suggest that the AMPA-type receptors in the CF-PC synapses may have a different subunit composition from those in GA-PC synapses.

The soluble form of guanylyl cyclase is a heme-containing protein involved in the enzymatic conversion of GTP to cGMP. Soluble guanylyl cyclase is localized in PCs, particularly in the primary dendrites (27), and the mRNA for soluble guanylyl cyclase is expressed in the granule cells and PCs (543). Soluble guanylyl cyclase acts as a receptor for NO and, in the NO-activated form, synthesizes cGMP. In cultured PCs, immunofluorescence associated with a monoclonal antibody recognizing both - and -subunits of soluble guanylyl cyclase was shown to increase in response to NO generated by a NO donor (525). Intracellular application of the potent and selective inhibitor of soluble guanylyl cyclase, 1H-[1,2,4]oxadiazole[4,3-a] uinoxalin-1-one (ODQ), has been shown to effectively block LTD induction (62).

Carbon monoxide (CO) is an activator of guanylyl cyclase and is formed by the catalytic action of inducible type 1 and constitutive type 2 heme oxygenase. The heme oxygenase-2 mRNA is densely distributed in the granular cells and PCs, similarly to the mRNA for soluble guanylyl cyclase, which may suggest that CO replaces NO in activating soluble guanylyl cyclase (533). Nevertheless, this possibility is doubtful because of the low enzymatic activity of the soluble guanylyl cyclase-CO complex (212). In primary cultures of olfactory neurons, zinc-protoporphyrin-9 (ZnPP), a heme oxygenase inhibitor, was shown to deplete endogenous cGMP (533). However, the specificity of ZnPP to heme oxygenase is questionable because ZnPP may act directly on soluble guanylyl cyclase or may deplete intracellular L-arginine that is required for NO synthesis (347, 419). Another potent activator of guanylyl cyclase is arachidonic acid (524) (sect. IVB5).

The CRF type 1 receptor (CRFR1) and its mRNAs are present on the somata and dendrites of PCs (437). CRFR1 is positively coupled with adenylyl cyclase through G_s proteins (42), but there is no evidence indicating that adenylyl cyclase, cAMP, or PKA has any role in LTD induction (sect. IVC5). There is evidence that PKC is activated via CRF receptors in cerebellar tissues (382).

Involvement of CRFR1s in LTD induction is indicated by the finding that CRF antagonists, -h CRF and astressin, blocked LTD induction (382). For LTD induction, however, CRF release driven by CF impulses does not appear to be required each time because the CRF antagonists effectively block the LTD induced by GA/Ca²⁺ spike conjunction without stimulating CFs. CRF released spontaneously from CF may play a permissive role in LTD induction.

PCs express receptors for IGF-I (60, 162, 551). Involvement of IGF-I in LTD induction has been suggested in a microdialysis experiment where glutamate applied to rat cerebellar cortex induced the release of GABA from the cerebellar nucleus (82). The glutamate-induced GABA release was persistently inhibited when IGF-I, but not the basic fibroblast growth factor (bFGF), was also applied to the cerebellar cortex (81). Electrical stimulation of the inferior olive in conjunction with glutamate application to the cortex also inhibited subsequent glutamate-induced GABA release from the nuclei. A PKC inhibitor and a NOS inhibitor blocked the effect of glutamate/IGF-I conjunction in depressing the glutamate-induced GABA release. A phorbol ester, a PKC activator, or L-arginine, a NO donor, when applied in conjunction with glutamate, mimicked the effect of IGF-I in depressing the GABA release.

More direct evidence for the involvement of IGF-I and its receptors in LTD induction has recently been reported (542). Application of IGF-I or insulin that also binds to IGF-I receptors effectively induced LTD in cultured PCs as represented by a reduction of AMPA currents, but not NMDA currents, and this effect of IGF-I was blocked by an antibody or an inhibitor of IGF-I receptors. How the activation of IGF-I receptors results in LTD induction is not clear for the present. Because IGF-I stimulates production of 1,2-diacylglycerol (DAG) (295), it could in turn activate PKC as required for LTD induction (sect. IV, B1 and C1). Because partial peptides of dynamin and amphiphysin that interfere with endocytosis blocked the IGF-I-induced LTD (541), this form of LTD may be caused by internalization of AMPA receptors by endocytosis, as suggested to occur in the final stage of signal transduction for LTD (sect. IVE3). Insulin-induced LTD also reduced the number of GluR2-containing AMPA receptors in tissue-cultured hippocampal neurons, and this form of LTD is caused by clathrin-dependent endocytosis (356).

Because PCs do not develop NMDA receptors associated with Ca²⁺ channels (see above), Ca²⁺ entry to PCs is mostly mediated by voltage-sensitive Ca²⁺ channels, occurring during either Ca²⁺ spikes or current-induced membrane depolarization (315, 373, 377, 513). A significant transient entry of Ca²⁺ in PC dendrites occurs during Ca²⁺ spikes and associated plateau potentials, but some Ca²⁺ signals also occur in somata (263, 315, 373). Ca²⁺ entry is induced not only during CF-evoked Ca²⁺ spikes, but also during AMPA-EPSPs induced by GA stimulation of below the threshold of the generation of Ca²⁺ spikes. This Ca²⁺ entry is confined to a compartment consisting of a small number of branchlets of terminal spiny dendrites (142) or even in individual dendritic spines (112). The dendritic Ca²⁺ signals depended on the size of GA-evoked EPSPs so that 20-30 GAs must be activated within the local dendritic area to produce a detectable increase in the postsynaptic Ca²⁺ concentration. Ca²⁺ entry is indispensable for LTD induction (sect. IVA1).

Six types of voltage-dependent Ca²⁺ channels have so far been identified (termed T, L, N, P, Q, and R). These channels are heteromultimeric complexes composed minimally of one main subunit, ₁, serving as both pore and voltage sensor, and auxilliary - and ₂-subunits. PCs are immunoreactive to _1A in the somata as well as the entire length of dendrites (553). The high-threshold P channel was originally described in PCs (339). The high-threshold Q channels were first described in cerebellar granule cells as distinct from P channels in its inactivation kinetics and sensitivity to -agatoxin IVA, a spider toxin (442). P and Q channels appear to be generated by alternative splicing of the _1A-subunit gene (61). The funnel web spider toxin (FTX), a specific P channel blocker, abolished the Ca²⁺ spikes in PCs (529). The -agatoxin IVA, which blocks Q type of Ca²⁺ channels, abolished the depolarization-induced Ca²⁺ spikes, the spike-associated steep increase in the dendritic [Ca²⁺]_i, and the steady intrasomatic increase in Ca²⁺ concentration (546). Thus P/Q channels are responsible for the generation of dendritic Ca²⁺ spikes. The N and L types of high-voltage-activated Ca²⁺ channels are also expressed in PC dendrites (552), but evidence supporting the possibility that N or L types of Ca²⁺ channels play significant roles in PC functions is limited (546).

The presence of low-voltage-activated (also called low-threshold or T type) Ca²⁺ channels has been controversial. The results of electrophysiological recordings and optical Ca²⁺ measurements suggest the presence of the T type of Ca²⁺ channels in the PC dendrites (392, 546). Of the three T-channel-related ₁-subunit genes (_1G, _1H, and _1l), _1G-mRNAs were found to be strongly expressed in PCs (509). The _1E-subunit of another low-voltage-activated/fast-inactivating type of Ca²⁺ channel was also reported to be present in the fine dendrites of PCs (571). The blockade of low-voltage-activated Ca²⁺ channels by Ni²⁺ delayed the onset of generation of depolarization-induced Ca²⁺ spikes, suggesting that the T-type Ca²⁺ channels or those containing the _1E-subunit generate transient inward currents that facilitate the generation of Ca²⁺ spikes (546).

Na⁺ enters PCs via glutamate-gated cation channels, voltage-sensitive Na⁺ channels, and Ca²⁺/Na⁺ exchangers. An increase in [Na⁺]_i was observed in both dendrites and somata after GA stimulation (77). However, the increase in the somatic [Na⁺]_i was observed only when regenerative spikes were detected in the somata and therefore would represent Na⁺ entry through voltage-sensitive Na⁺ channels. The increase in the dendritic [Na⁺]_i evoked by GA stimulation was blocked by CNQX and should be associated with AMPA receptors.

In cultured PCs, multiple types of voltage-sensitive K⁺ channels having different single-channel conductances have been identified in both PC somata and dendrites (180). An excitability increase in PC dendrites, presumably due to changes in a K⁺ current, has been detected even 1 mo after eye-blink conditioning (468).

There are ~20 different species of G proteins exerting different effector actions. G_s is a family of G proteins that activates adenylyl cyclase, whereas G_i inhibits adenylyl cyclase or activates cGMP-phosphodiesterase. A new class of G proteins, the G_q family, has recently been identified and found to be associated with group I mGluRs (sect. IIIB3) and involved in PLC activation (sect. IIIC4). Of the four isoforms of G_q subunits (G_q, G₁₁, G₁₄, G_15/16), the highest transcriptional rate of G_q was observed in PCs and hippocampal pyramidal cells. That of G₁₁ was also noted in hippocampal pyramidal cells, but the other two were scarce in the central nervous system (511). The isoforms G_q and G₁₁ share a similar amino acid sequence (88%), receptor specificity, and ability to activate PLC-. Immunoreactivity for an antibody against the COOH terminus common to G_q and G₁₁ is abundant in the dendrites of PCs (352). With immunogold labeling, G_q/₁₁ immunoparticles are distributed in the perijunctional zones within 600 nm from the edge of the postsynaptic densities, corresponding to the perijunctional distribution demonstrated for mGluR1 (511) (sect. IIIB3).

Certain bacteria toxins are used to disrupt G protein activities by ADP-ribosylation of the -subunit of G protein (30). Cholera toxin blocks the G_s family, and pertussis toxin blocks the G_i family. The G_q family, however, is resistant to both of these toxins. The action of mGluR1s to release Ca²⁺ from intracellular stores in PCs (sect. IIID2) is insensitive to pertussis toxin (575), as would be expected because this action is mediated by G_q protein. However, quisqualate-induced reduced form of LTD, recorded by the grease-gap method, was blocked by pertussis toxin (236), suggesting the involvement of another type of G protein, probably G_i, in LTD induction.

Hydrolysis-resistant analogs of GTP such as guanosine 5'-O-(2-thiodiphosphate) (GDPS) and guanosine 5'-O-(3-thiotriphosphate) (GTPS) are also used to interfere with G protein activities by persistently activating effector systems. Photolytic release of GTPS in PCs produces a large inward current followed by a small outward current, mimicking the response to an mGluR agonist (1S,3R-ACPD) (sect. IIIB3) (501). On the other hand, prolonged intracellular infusion of either GDPS or GTPS prevented the 1S,3R-ACPD-induced generation of the inward current (210). In mice deficient in G_q, CF-PC and GA-PC transmission was functional, but CFs remained to multiply innervate PCs (414) and LTD was lacking (381).

E.  Phospholipases

1.  PLC

PLC is a family of enzymes that hydrolyze the membrane phosphatidylinositol 4,5-bisphosphate (PIP₂) producing two second messengers: DAG (see sect. IIID1) and inositol trisphosphate (IP₃, see sect. IIID2). Among its five isoforms of , , , , and , three (, , and ) are phosphoinositide specific, and PLC- alone is associated with G proteins (199). The activation of metabotropic receptors leads to the activation of PLC- via the activation of G_q/11. The four isoforms of PLC- have recently been cloned, of which PLC-3 mRNA is specifically expressed in PCs, and PLC-4 mRNA is most highly expressed in PCs, whereas PLC-1 and -2 are localized in other areas of mouse and rat brain (456, 545). PLC-4 is distributed equally in both the rostral and caudal cerebellum, whereas PLC-3 is abundant in the caudal compared with the rostral cerebellum. In other words, PLC-4 functions dominantly in the rostral cerebellum, while PLC-3 and PLC-4 function to a similar extent in the caudal cerebellum. In PLC-4-deficient mice, the activation of mGluR1s, which normally induces an increase of [Ca²⁺]_i in the cerebellum, resulted in no such effect in the rostral cerebellum and only a small increase in the caudal cerebellum (501). PLC-3 and PLC-4 are activated differently from each other by subunits of G_q proteins (199, 248); -subunit activates PLC-4 more potently than PLC-3, whereas -subunit is much more effective in activating PLC-3 than PLC-4 (501).

The PLC-mediated pathway, presumably involving G_q/11 and PLC-3/4, plays a crucial role in LTD induction, since its downstream effects, both activation of PKC by DAG and release of Ca²⁺ from intracellular stores by IP₃, are required for LTD induction (sect. III, B1 and B2). A PLC inhibitor actually greatly attenuated the mGluR agonist-induced [Ca²⁺]_i increase (410), which was also blocked in mice deficient in the PLC-4 gene (264). On the other hand, PLC does not seem to contribute significantly to the generation of mGluR-EPSPs because PLC inhibitors, PKC inhibitory peptides, or heparin sodium, a nonspecific inhibitor of IP₃ receptors, have no significant effects on the mGluR-EPSPs (514) or mGluR agonist-induced inward currents (210). This is consistent with the report that photolytic release of IP₃ from its caged form did not induce an inward current (501).