The Cerebellum: Brain for an Implicit Self (18 page)

In vestibular nuclei, a component (N2) of the field potentials evoked by electric stimulation of the vestibular nerve has been considered to represent the monosynaptic activation of vestibular nuclear neurons. After sustained optokinetic stimulation was applied, the N2 potential was enhanced, thereby suggesting the occurrence of LTP (
Shutoh et al., 2006
) (
Chapter 10
, “
Ocular Reflexes
”).

Rebound depolarization.
A robust rebound depolarization observed in cerebellar nuclear neurons following the offset of current-induced membrane hyperpolarization (
Chapter 6
, “
Pre- and Post-Cerebellar Cortex Neurons
”) has been related to an unusual form of synaptic plasticity. Pugh and Raman (
2006
) recorded EPSCs elicited in cerebellar nuclear neurons and showed that they were potentiated only when high-frequency synaptic activation was coupled with sufficient postsynaptic hyperpolarization evoked by Purkinje cells. This potentiation occurred as long as rebound current was in operation and even when postsynaptic spiking was prevented by a voltage-clamp. These data support Medina and Mauk’s (
1999
) postulate adopted in a large-scale computer simulation of cerebellar circuits. They proposed that inhibition from Purkinje cells controls potentiation of mossy fiber synapses in cerebellar nuclear neurons (see
Chapter 9
, “
Network Models
,”
Section 7
for a possible learning rule in cerebellar nuclear neurons).

Activity-dependent synaptic plasticity is generally considered to be a transitory cellular process, which is eventually consolidated into a permanent memory trace, albeit by way of an as-yet-unknown mechanism. Information in at least initial memory is conventionally supposed to be stored as a patterned set of activity-dependently modified synapses in a neuron or a neuronal network. On the basis of this Hebbian model of memory, efforts have been devoted to exploring the long-lasting effects of activity-dependent synaptic plasticity. Several relevant findings are presented below.

How long does synaptic plasticity persist? Using current technology, it is usually difficult to continuously follow a neuronal event, such as conjunctive LTD, for more than 2 hours. In Purkinje cells co-cultured with granule cells, miniature EPSCs (mEPSCs) can be intermittently observed by repenetration of the test cell. Murashima and Hirano (
1999
) found under this specific condition that a reduced form of conjunctive LTD lasted longer than 36 hours but shorter than 48 hours. Under the assumption that conjunctive LTD is required for adaptive behavior in reflexes, adaptation of optokinetic eye movement suggests that conjunctive LTD in the flocculus lasts one day, whereas LTP in vestibuloocular relay neurons lasts one week (
Shutoh et al., 2006
). A study of eye-blink conditioning suggests that the effect of conjunctive LTD lasts even up to one month (
Chapter 10
).

Spine configuration.
When one is studying the potential relation between synaptic plasticity and memory processes, a critical question is whether this plasticity involves an obligatory structural change in the tested synapse. Indeed, in dendrites of pyramidal neurons, LTP is associated with spine addition and enlargement, and LTD with spine loss and shrinkage (
Lippman and Dunaevsky, 2005
). In Purkinje cell dendrites, however, neither local synaptic induction nor global chemical induction of LTD changed spine number and size (
Sdrulla and Linden, 2007
). In contrast, a significant decrease in the number of excitatory synapses in the outer layer of the cerebellar cortex occurred by the third day of eyeblink conditioning in rabbits. This change was presumably brought about by the actions of conjunctive LTD (
Connor et al., 2009
). During long-term adaptation of the OKR, the number of parallel fiber-Purkinje cell synapses also decreased significantly in the mouse flocculus (by ~1/3 after 5 days). This certainly suggested that conjunctive LTD was followed by a loss of spine synapses (
Nakadate et al., 2004
).

Silent synapses.
A rather surprising recent finding is that stimulation of a single granule cell produced a detectable EPSC in only a small fraction of those Purkinje cells whose dendritic trees were traversed by the parallel fibers of the stimulated cell. Because ~54% of traversing parallel fibers form synaptic contacts with one Purkinje cell, it was suggested that up to 80% of parallel fiber-Purkinje cell synapses are silent or have a very low transmission efficacy (
Isope and Barbour, 2002
). In another experiment on a sagittal slice of the rat cerebellum (
Wang et al., 2000b
), parallel fiber stimulation enhanced Ca
²⁺
transients in a dendritic area 11 micrometers in radius (380 micrometers wide) containing 1,140–2,660 parallel fiber-Purkinje cell synapses (as determined from the anatomical finding that 1 um
²
of a dendritic area contains 3–7 of these synapses). In contrast, parallel fiber stimulation evoked 60–370 picoampere EPSCs in a Purkinje cell. Because unitary parallel-fiber synaptic current is 12 picoamperes, 5–30 parallel fiber-Purkinje cell synapses must have been activated. Hence, the electrical stimulation of parallel fibers in a small dendritic area of a Purkinje cell activates at the most 2.6% (30/1,140) of the parallel fiber-Purkinje cell synapses located within the test area. These findings are consistent with another observation obtained by an entirely different approach (
Ekerot and Jörntell, 2001a
,
b
). In the forelimb area of the C3 zone, each Purkinje cell responded via a mossy fiber-parallel fiber pathway to stimulation of only one of the 30–40 types of receptive fields that cover the forelimb skin in its entirety. The receptive fields of Purkinje cells vary even between neighboring Purkinje cells located along the same parallel fiber bundle. Hence, each Purkinje cell is activated via only ~2.5%–3% (1/40–1/30) of the total parallel fiber input received from forelimb skin. Thus, the three types of measurement described previously were consistent in their indication that each Purkinje cell computes on the basis of a small fraction of parallel fiber-Purkinje cell synapses. Of the 175,000 parallel fiber synapses each Purkinje cell receives (
Chapter 4
), 5,000 could thus be functional.

Also surprising is the conspicuous effect on Purkinje cells of the repetitive high-frequency stimulation of parallel fibers (trains of 15 pulses @ 100 Hz were repeated at 1/3 s for 10 minutes) when unpaired with climbing fiber activation (
Jörntell and Ekerot, 2003
). A long-lasting, very large increase in the receptive field size of Purkinje cells then develops over 2–6 hours. In contrast, the stimulation of parallel fibers paired with climbing fiber activity reverses such changes in the size of the receptive field. The same repetitive stimulation of parallel fibers induces opposite effects on inhibitory interneurons. On the basis of these findings, Ekerot and Jörntell (
2003
) suggested that silent synapses are produced functionally by LTD and that learning converts silent synapses to active ones by the induction of LTP, the reverse occurring for LTD. The predominance of functionally silent parallel fiber synapses in Purkinje cells may imply that up to 97% of these synapses are long-term depressed during repeated learning trials. During this time “neurocomputing circuits” are emerging from the original randomly connected parallel fiber synapses.

Whereas short-lasting memory is linked to functional changes in existing synapses, long-lasting memory is thought to be associated with a structural change such as the number of synaptic connections. The protein synthesis required for development and maintenance of synaptic structures should play a key role in the production and maintenance of a long-lasting memory (see
Kandel, 2009
). Proteins are synthesized not only in the somata but also locally in the dendrites of neurons in association with synaptogenesis and synaptic plasticity (
Schuman et al., 2006
). Accordingly, dendrites contain many different mRNAs, and the translational machinery includes ribosomes, polyribosomes, and elongation factor 2, as well as the endoplasmic reticulum and cisternae of the Golgi apparatus.

In studies related to protein synthesis, translational inhibitors (anisomycin, puromycin, and cycloheximide) and transcriptional inhibitors (actinomycin D and 5,6-dichloro-β-D-ribofuranosyl-benzimidazole, DRB) are commonly used. The translational inhibitors depress conjunctive LTD in cultured Purkinje cells only in its late phase. This suggests a role for protein synthesis in sustaining conjunctive LTD, that is, providing proteins for structural changes that take place as activity-dependent synaptic plasticity evolves (
Linden, 1996
). However, a puzzling observation was that the entire phase of conjunctive LTD in cerebellar slices was abolished by a 5-minute perfusion of translational inhibitors during conjunctive stimulation (Karachot et al., 2000). When a translational inhibitor was applied after a delay of 15 minutes or more, it no longer blocked conjunctive LTD. It appears that in slice conditions, a “quickly-turning-over” protein is required for the induction of LTD, probably in addition to that required for maintenance of LTD during its late phase. The composition of such a protein is not yet known, however.

Cyclic AMP response element-binding protein (CREB).
This nuclear protein regulates the transcription of genes with a CRE site in their promoter (
Silva et al., 1998
). When synaptic activation causes a high intracellular Ca
²⁺
concentration, CREB is phosphorylated at the site of its transcriptional regulatory residue, serine 133, by involving CamKIV. When transfected with a dominant inhibitory form of CREB, which prevents DNA binding of endogenous CREB, or with dominant-negative constructs of CamKIV, Purkinje cells failed to develop the late phase of LTD, just as they did under the influences of transcriptional inhibitors (
Ahn et al., 1999
). CREB and CaMKIV are therefore suggested to contribute to the late phase of LTD. Genes encoding the a and b polypeptides of CaMKIV are highly expressed in cerebellar granule cells, but their expression in Purkinje cells is still unclear (
Sakagami and Kondo, 1993
). Nevertheless, CREB is a survival factor
for many neurons, and its deletion results in massive neuronal apoptosis. This might complicate the effects of CREB deletion on synaptic plasticity.

Serum response factor (SRF) is another transcription factor that is regulated by synaptic activity, but its deletion in the adult brain does not cause neuronal death or gross malformation. Rather, such deletion produces a near-complete blockade of the induction of immediate early genes, including those that encode the synaptic protein, Arc. Smith-Hicks et al. (
2010
) secured the following findings. Deletion of the Arc gene blocked the late phase of LTD in cultured mouse cerebellar Purkinje cells, and the inhibition of SRF or its cofactor MAL displayed similar effects. Furthermore, when Arc-/- cells were transfected with a wild-type Arc, the late-phase LTD was rescued. However, mutation of one SRF-binding site in the Arc promoter, SRE 6.9, blocked this rescue. Co-transfection of wild-type Arc and SRF (engineered to bind mutated SRE 6.9) restored late-phase LTD in Arc-/-, SRE 6.9 mutant cells. Thus, SRF binding to SRE 6.9 in the Arc promoter is required for the late phase of cerebellar LTD.

Various subtypes of synaptic plasticity found to this point must surely play their respective roles in the operational mechanisms of cerebellar neuronal circuits. The relationship between activity-dependent synaptic plasticity and memory traces remains unclear, and this issue stands out as a pressing future problem. Indeed, there seem to be endless possibilities for the discovery of novel components in cerebellar neuronal circuits!

Having just considered the diverse cellular and molecular events occurring in the cerebellum, we can now examine how information is processed in its neuronal circuits. Since the monumental Marr and Albus contributions, various models have been proposed. In this chapter, we see how our understanding of such models has advanced recently.

Mossy fibers, the major inputs to the cerebellar cortex, make synaptic connections with the most numerous small neurons in this cortex, the granule cells. These connections follow a characteristic asymmetric pattern of divergence and convergence; one mossy fiber branches to contact with 400–600 or more granule cells, whereas each granule cell is contacted by 4–5 mossy fibers (cat,
Palkovits et al., 1971
; rat,
Jakab and Hamori, 1988
).

Marr (
1969
) assumed that afferent information communicated by mossy fibers to the cerebellar cortex was returned to a combination of small subsets of active mossy fibers that he called “codons.” Each codon represented a feature of the input. The size of a codon that could activate a given granule cell varied from cell to cell, depending on its threshold, which could be regulated by Golgi cell inhibition or LTP (
Chapters 5
and
8
). A large-scale simulation of codon behavior was performed subsequently and it generally supported Marr’s model (Tyrrell and Wilshaw, 1992).

Various ideas have been proposed for the functional significance of several mossy fibers converging onto single granule cells. Each granule cell may sample different types of mossy fiber input and associate them in patterns that are discriminated later by Purkinje cells. The small convergence number in the mossy fiber-granule cell pathway may imply a “sparse coding” mechanism, which features strong activation of a relatively small number of granule cells for each item of information. This would help transmit a complete contextual account of mossy fiber activity to a Purkinje cell and thereby minimize interference between the tasks being learned, a process that would augment information storage capacity (
Brunel et al., 2004
;
Philipona and Coenen, 2004
).