The Cerebellum: Brain for an Implicit Self (8 page)

In the aforementioned climate, David Marr (1945–1980), James Albus, and a few other theorists proposed theoretical models of the cerebellar neuronal machine (e.g.,
Marr, 1969
;
Albus, 1971
). This was an eagerly awaited breakthrough for computational neuroscience. I remember its great impact on me after reading
Marr’s 1969
article. I felt that his theory converted our wiring diagrams of the cerebellum into a meaningful blueprint.

The crucial assumption adopted in Marr’s theory was the use of synaptic plasticity as a memory element in neuronal circuits. At that time this was but a theoretical possibility and totally lacking in supportive experimental evidence. As mentioned in
Chapter 1
, “
Neuronal Circuitry: The Key to Unlocking the Brain
,” Hebb (
1949
) had already proposed the concept of Hebbian synapses, whose transmission efficacy increased when the presynaptic and postsynaptic neurons fired in synchrony. Brindley (
1964
) pointed out the possibility that the convergence of parallel fibers and climbing fibers onto Purkinje cells implied the presence of Hebbian synapses, since climbing fiber signals are so powerful that these inevitably excited Purkinje cells. Thus, if both parallel fibers and climbing fibers were activated synchronously, parallel fiber-Purkinje cell synapses were activated both presynaptically and postsynaptically, that is, the type of condition that induced a Hebbian form of plasticity. In Marr’s (
1969
) model, as based on Brindley’s suggestion, learning actions were considered to occur as follows. Each climbing fiber conveyed a cerebral instruction for an elemental movement, and the receiving Purkinje cell was also exposed via the mossy fiber input to information about the context in which the climbing fiber fired. During rehearsal of an action, each Purkinje cell could learn to recognize such contexts, and later, after the action had been learned, the occurrence of the context alone was enough to fire the Purkinje cell, which then caused the next elemental movement.

Albus’ model (
1971
) was a close analogy to the simple perceptron, assuming that climbing fibers played the role of the outside teacher as a supervisor (recall
Figure 6
). When a successful performance of the cerebellum was recognized, relevant climbing fibers sent signals that potentiated concurrently activated parallel-fiber synapses on Purkinje cells (i.e., potentiation of the synapses that brought about success). On the other hand, when the performance was unsuccessful, relevant climbing fibers sent signals to depress concurrently activated, parallel-fiber synapses on Purkinje cells (i.e., depression of the synapses involved in failure). However, it is impossible to use the same climbing fiber for both potentiation and depression in real synapses. This meant that one of them had to be chosen. Albus (
1971
) selected depression for several technical reasons, whereas Marr used potentiation after success. Theoretically speaking, learning was possible using either model. Thus, these models raised alternative possibilities to be selected on an experimental basis.

It is to be noted that the simple perceptron is primarily designed for discrimination of spatial patterns and has no capability of discriminating temporal patterns. A decade after Marr’s and Albus’ models, Fujita (
1982a
) proposed an adaptive filter model of the cerebellum able to discriminate temporal patterns by assuming that the neuronal circuit involving mossy fibers, granule cells, parallel fibers, and Golgi cells constitutes a phase converter, which generates a set of multiphase
versions of a mossy fiber input.
Figure 16
shows schematically the early idea of the operation of Fujita’s adaptive filter model of the cerebellum when the input signal is sinusoidal. Fujita (
1982b
) incorporated successfully this phase converter concept into a model of VOR adaptation and reproduced successfully the adaptation of the VOR (
Chapter 10
, “
Ocular Reflexes
”). The importance of the granule cell-Golgi-cell-granule cell pathway as a clock in the cerebellum has now been well recognized (
Chapter 9
, “
Network Models
”).

Figure 16. Adaptive filter model of the cerebellum.

This model explains how the cerebellar network recognizes temporally encoded signals. It is assumed that a phase-converter consisting of the mossy fiber (MF)-granule cell (Gr open circle)-Golgi cell (Go filled circle) circuit generates a set of multiphase versions of mossy fiber (MF) input (represented by sinusoidal discharge). When Purkinje cells (PC) use conjunctive LTD in their learning, a certain phase-shifted version of the input, which is out of phase to the climbing fiber error signals, is selected by Purkinje cells. On the other hand, granule cell to Purkinje cell transmission in phase with the climbing fiber (CF) input (indicated by a left-directed arrow) will be depressed. (Explanation based on Fujita’s [
1982a
] model; see also
Dean et al., 2010
for another explanation of the model.)

In the late 1960s and early 1970s, many laboratories apparently tried to reveal such synaptic plasticity, but in vain. It is widely known that Eccles invited Marr to sit in front of a cathode ray oscilloscope with him while they tested the effects of conjunctive stimulation of climbing fibers and parallel fibers using stimulus parameters chosen by Marr. No sign of synaptic plasticity was then observed, however. At that time, experiments were conducted
in vivo
such that stable intracellular
recording was not possible for a period sufficiently long to detect synaptic plasticity. Accordingly, transmission across parallel fiber-Purkinje cell synapses was examined only by extracellular recording of field potentials. However, as compared to field potentials recorded in the hippocampus to reveal long-term potentiation (
Bliss and Lomo, 1973
), those in the cerebellar cortex were ten times smaller. This meant that before the availability of high-performance electronic averagers, any potential long-term modification of synaptic transmission could not be detected. The observation was further impeded because several factors in
in vivo
experiments were later shown to interfere with the occurrence of the synaptic plasticity: postsynaptic inhibition caused by basket/stellate cells (
Ekerot and Kano, 1985
), local bleeding resulting in the release of hemoglobin that absorbs nitric oxide (
Nagao and Ito, 1991
), and general anesthesia (
Vigot et al., 2002
).

Despite the preceding evidence to the contrary, I agreed with the two theorists, Marr and Albus, because, as shown below, the flocculus hypothesis of the vestibuloocular reflex (VOR) that I was proposing at that time matched very well with their models. In 1979, I visited Professor David Hubel at Harvard Medical School to present a seminar. When it ended, Marr approached me, this being our first and only interaction. He mentioned his interest in my flocculus hypothesis for the VOR and asked me to send him any related publications. He also said that he would soon leave for the U.K. for leukemia treatment but would possibly visit Japan the following year to receive a prize from an artificial intelligence group. I told him that I had been waiting to meet him for ten years and that I was continuing my research on synaptic plasticity. Upon returning home, I received a letter from Marr, in which he mentioned gracefully that he, too, had been waiting ten years to meet me. Sad to say, Marr did not come to Japan, and I regretted that I could not tell him person to person about the new positive evidence of synaptic plasticity, which I reported at the XXVII Congress of the International Physiological Union (IUPS), which was held in Budapest in June 1980 (
Ito et al., 1981
). Béla Julesz (1928–2003) consoled me to some extent, however, when he informed me that he had written to Marr, who was by then quite ill in bed in Cambridge, Massachusetts, to tell him about my Budapest report. Sadly, Marr died in late 1980.

The new evidence presented in Budapest (
Ito et al., 1981
) was a result of my change in strategy from using field potentials to test for parallel fiber-Purkinje cell transmission to measuring the rate of Purkinje cell discharge in response to half-maximum parallel fiber stimulation (“firing index”). While recording from a Purkinje cell in the flocculus, Masaki Sakurai, Pavich Tongroach, and I witnessed that conjunctive stimulation of vestibular mossy fibers and climbing fibers decreased unfailingly the firing index (
Ito et al., 1982
). Even though we were stimulating vestibular mossy fibers, field potentials in the vestibular nuclei and flocculus granule layer were confirmed not to reveal any related changes. We also recorded from putative basket cells, in which conjunction induced no depression like that observed in Purkinje cells. Because Purkinje cells and basket cells share the mossy fiber-parallel fiber pathway, we reasoned that the depression specific to Purkinje cells must have taken place in the Purkinje cells, themselves. Moreover, we demonstrated that the sensitivity of Purkinje cells to iontophoretically-applied glutamate (the transmitter released from parallel fibers), but not to aspartate or N-methyl aspartate (not a transmitter for parallel fibers), was depressed for a considerable duration after combining climbing fiber stimulation and glutamate application. Shortly thereafter, we received a grant to purchase a high-performance electronic averaging instrument. Its use enabled Masanobu Kano and me to record the field potentials representing monosynaptic activation of Purkinje cells by parallel fiber impulses and to demonstrate that conjunction induced long-lasting depression of these potentials, this being definite evidence of the manifestation of LTD (
Ito and Kano, 1982
). Next in my laboratory, Karl-Frederic Ekerot and Kano used direct stimulation of parallel fibers combined with Purkinje cell firing indices to reveal the occurrence of LTD (
Ekerot and Kano, 1985
). Later, the successful recording of LTD in cerebellar slices (
Sakurai, 1987
) prompted many more studies of LTD, which were undertaken worldwide. By 1990, LTD was established as a unique type of synaptic plasticity (
Ito, 1989
). Nowadays, conjunctive LTD can be observed routinely in tissue cultured Purkinje cell preparations developed by Linden’s group and in the cerebellar slice preparations used in other laboratories, including my own (
Figure 17
).

Figure 17. Induction of LTD in a slice of the cerebellum.

(A) Intracellular recording from a Purkinje cell in a slice of the mouse cerebellum. 1, EPSPs evoked by double shock stimulation of parallel fibers (2PF). 2, Five Ca
²⁺
-spikes induced by application of a membrane depolarizing current pulse (md). 3, Similar to 2, but initial two Ca
²⁺
-spikes were driven by 2PF (at upward arrows) superimposed on an md. 4, An averaged single shock-evoked parallel fiber EPSP recorded before (x) and after (y) conjunction of PF stimulation and an md at 1 pulse/second for 5 minutes to bring on LTD (x minus y). 5, A record similar to that in 4 except for the stimulation being restricted to before and after 2PF. 6, A record similar to 5 except for the stimulation being restricted to before and after an md. (B) Time course of LTD in two mouse strains (C57BL WT and C3H WT). Abscissa, time in minutes (min) relative to onset of stimulation. Ordinate, relative rising rate of Purkinje cell EPSP responses to PF stimulation. For both mouse strains, plots are shown for control 2PF and md stimulation versus conjunction of these stimuli. x and y, time for recording of the traces x and y in A4. In brackets, number of tested cells for each stimulating condition. (From Le and Ito, unpublished material.)

In the 1990s, signal transduction processes underlying LTD became a subject of extensive investigation in many laboratories (see
Daniel et al., 1998
). I recall that when I moved to RIKEN (Institute of Physical and Chemical Research) in 1990, little was known about this subject. Now, however, a complex flow chart is available. It shows chemical signals involving more than 30 different molecules (for review, see
Ito, 2001
,
2002
). While I was concentrating on the mechanism of signal transduction for LTD, there were notable research developments in several directions on the nature of cerebellar synaptic plasticity. Postsynaptic LTP as the counterpart of conjunctive LTD had long been missing, but Lev-Ram et al. (
2002
,
2003
) finally found it. The involvement of cerebellar/vestibular nuclear neurons in learning, in addition to LTD in the cerebellar cortex, was suggested early on (
Miles and Lisberger, 1981
;
Lisberger and Sejnowski, 1992
;
Raymond et al., 1996
). It has now been shown quite clearly (
Kassardjian et al., 2005
;
Shutoh et al., 2006
;
McElvain et al., 2010
). Moreover, a wide variety of synapses in the cerebellar cortex have been shown to be activity-dependent and subject to plastic modification (see
Hansel et al., 2001
). These advances are reviewed in later chapters.