INTRODUCTION:
The traditional position of the cerebellum in the hierarchic organization of the central nervous system. The functions of the cerebellum have been obscure, and, accordingly, speculation about them rather free. But the last century has seen the development of a systematic approach to the understanding of brain structure and function. The steps in the approach include: 1) connectivity-- could the known circuitry support the functions proposed? 2) ablation--is the function abolished or impaired by removal of the part in question? 3) stimulation--can the function be evoked by stimulating the part in question? 4) natural activation--does naturally increased activity of the part in question correlate with the behavior in question?
The fundamental organizational plan of the central nervous system in general and the motor system in particular is hierarchical. Hierarchic structuring is evident in its anatomy, phylogeny, ontogeny, and in the effect of discharging and destroying lesions (Jackson, 1870). This principle assumes fresh and crucial significance in understanding cerebellar functions.
In Jackson's scheme, the lowest level in the hierarchy is the spinal cord (Fig. 1); the motor neuron is the final common pathway for all motor commands to muscle. Nowadays, spinal interneurons are thought to control muscle force, length, stiffness, and viscosity, and provide a compliant interface with the environment. Withdrawal and crossed extension reflexes provide protective responses to noxious stimuli, and also the substrate for locomotion. When set to oscillate, the reflex circuits generate the rudiments of walking and running behavior.
The middle level consists of a number of motor pattern generators, each projecting down to the spinal cord mechanisms to provide movements with a specific goal. Vestibulo-, reticulo- spinal and tonic neck reflexes provide the automatic anti-gravity component of upright stance and locomotion. The superior colliculus provides automatic orienting movements of eyes, head, neck and possibly limbs to visual, auditory, and somatosensory stimuli. Cortico- and rubrospinal neurons provide volitional movement of individual muscles, particularly of face, fingers, and toes.
The highest level of motor control in this scheme consists of those parts of cerebral cortex associated with thinking and knowing. The back half of the brain--the posterior and parietal lobes--are devoted to higher order processing of vision, audition, and somesthesis. The front half--the frontal lobe--is devoted to movement and motivation. In the human brain, the left hemisphere specializes in language analysis, the right, spatial relationships within self and extrapersonal space. The upper portions of the parietal lobes partake in localizing objects as potential targets for movement, the lower portions and posterior temporal lobes in recognizing the objects. The upper portions of the temporal lobe partake in language and spatial relations on the left and right sides, respectively; the lower portions, declarative memory. Cognition is the product of associations between the secondary receiving and sending areas.
The traditional role of the cerebellum. In animals, cerebellar ablation by Rolando (1790), Flourens (1824), and especially Luciani (1890) clearly showed that what the cerebellum was involved in: the control of posture and movement. The studies of human cerebellar ablation by Babinski (1899, 1909), Andre-Thomas (1911), and especially Holmes (1917, 1922, 1939) confirmed these observations, and appeared also to determine what the cerebellum was not involved in: sensation (except weight discrimination, which requires movement), perception, attention, learning, memory, mood, apperception, language--that is to say, all forms of cognition (Holmes 1917, 1922, 1939). As to the specific nature of the cerebellar contribution to movement, "fine control" was suggested because large focal cerebellar lesions often caused only the slightest (recognized) deficits in behavior; "coordination", because (to some observers) compound movements appeared to be more affected than simple (Flourens, 1824; Babinski, 1899; but see Holmes, 1939). Since then, the traditional teaching in neurology has been that the cerebellum facilitates the fine control and coordination of movement.
Physiological thinking about what the cerebellum does has also been greatly influenced by what comes into it. Sherrington called it "the head ganglion of the proprioceptive system". The cerebellum receives every sensory modality that has been looked for--muscle spindle, Golgi tendon organ, skin and joint receptors, vestibular, acoustic, visual and lateral line. And the "sensory" input is only a small part of the overall input to the cerebellum. In man, motor, sensory, cognitive and associational cerebral cortex contributes the greatest input via the large ponto-cerebellar projection. Yet even the "sensory" input is in all cases from second order (or greater) relay neurons, and is therefore pre- processed. Further, the ventral and rostral spino-cerebellar paths report on inter neuronal locomotion generators that are premotor and distinctly not sensory (Arshavsky et al, 1972). Nonetheless, some modern physiologists regard the cerebellum more as a sensory analyzer than a movement controller (Nelson and Bower, 1990).
The cerebellum was thus supposed to sit off the main line ("metasystemic", MacKay and Murphy), to receive information virtually from all parts of the nervous system, and to funnel down only to the motor generators of the middle level. The purpose was to provide for their fine control and coordination. The cerebellum was seen as being both motor and sensory, a "middleman" (Vermittler). The middleman used sensory information to facilitate or optimize the motor operation (Bloedel, 1992).
CEREBELLAR MOTOR LEARNING THEORIES.
The need: what would the cerebellum learn about movement? We do not and cannot think about all of the muscle actions in compound and sequential movements. There are too many of them, their transitions too fast, their temporal and magnitude relationships too precise. They occur through some process that is automatic, subconscious. One is aware of individual muscles and joints only when one is beginning to learn new compound movements. Then the movement is altogether different: it is slow, irregular, and "uncoordinated"; all the muscles that are to be involved are not working properly together in amplitude and time.
The theories: automation through trial-and-error learning of context-response linkage. Several theories and a number of lines of evidence have pointed to a crucial role of the cerebellum in the adaptation and learning of movement. Brindley first suggested (1964) that the acquisition of skilled movements, such as playing the piano, begins as a conscious act mostly under the control of the cerebral cortex, without help from the cerebellum (Fig. 1). But the cerebellum itself can also initiate the performance, and it immediately begins to acquire control of the task. It recognizes the contexts in which each "piece" of consciously initiated movement occurs, and after repeated tries, links that context within itself to the movement generators so that the occurrence of the context automatically triggers the movement. Thus, with time and practice, the cerebellum largely controls the process, with little or no help from the cerebrum. The cerebrum and the conscious mind is free to do and think about other things. The control of the task has been shifted from a conscious cerebral cortical process to a subconscious one mostly under the control of the cerebellum.
Both David Marr (1969) and James Albus (1971) independently modeled the process using the cerebellar circuit design and function as sketched by Ramon y Cajal and updated by Eccles, Llinas and Sasaki (Eccles et al., 1967; Llinas, 1981). Gilbert (1974) added that synapses were not simply turned "on" (Marr) or "off" (Albus) but were adjusted to give the continuum of Purkinje cell firing frequencies which are actually observed in awake behaving animals (cf Thach et al, 1992). The circuit models were based on the great differences between the two main cerebellar input systems. The highly convergent mossy fiber-parallel fiber-Purkinje cell system brought information from most parts of the nervous system, and the information was represented as modulations in high- frequency firing of 0-500/sec. The relatively one-to-one climbing fiber Purkinje cell system arose exclusively from neurons of the inferior olive; the synaptic contact was very powerful, but the firing rates were so low (around 1/sec) as to raise questions about the information content.
Mugnaini (1983) showed that the parallel fiber is 6 to 10 times longer than had been supposed, and contacts Purkinje cells along a beam spanning 1/3 to 1/2 the width of the cerebellar cortex. Recent animal studies have shown at least one more or less complete body map in each of N. fastigius, interpositus, and dentatus (Asanuma, et al, 1983--Fig. 5). Further, each nucleus appears to control a different aspect, or mode of movement for the entire body it maps (Thach, et al., 1992): dentate, synergist muscles in visually guided movements (e.g., pinch and reach); interpositus, agonist-antagonist synergy and stretch reflexes at a single joint (Smith and Bourbonnais; Frysinger et al; Wetts et al; Schieber and Thach); fastigius, synergists in upright stance and locomotion (Arshavsky et al; Thach, et al., 1992). Thach et al (1992) and Goodkin et al (1992) reported that cerebellar lesions impair compound movements more than simple, and suggested that a cardinal role of the cerebellum is to combine (through the learning mechanisms above) the elements of movement using the parallel fiber contacts on the long beam of Purkinje cells.
Proposed mechanism: parallel fiber beams represent the context and combine the response... The proposal was that the conditions and "context" in which a movement is to be learned and performed is represented in the modulated discharge of the mossy fibers (Fig.2), which monitor not only sensory information, but also much of the ongoing activity of most of the nervous system. This afferent information is transmitted to the parallel fiber, which branches to contact thousands of Purkinje cells, which in turn project to the somatotopic motor representations within each of the deep nuclei. The parallel fiber is the critical middle layer between sensory and other input and motor output, representing both the context in which the movement is made, and also being a chief instrument for organizing the motor response.
...the climbing fiber detects and corrects errors in performance, and changes the strength of parallel fiber-Purkinje cell synapses, thereby creating novel context-response linkages and response combinations. When a new movement needs to be learned or an old one adapted, the climbing fiber (Fig.2), which normally fires irregularly at a rate of around 1 Hz, and in no particular relation to movement, suddenly (driven by error between intended movement and actual movement--Albus) begins to fire (once) immediately after the error occurs, reliably time after time. The effect of this low-frequency but synaptically very powerful climbing fiber firing is to reduce (Albus) the strength of the synapse on the Purkinje cell of those parallel fibers that are active at the time (and helping to cause the inappropriate movement). What is left after practice, and repeated firings, are those parallel fibers whose action causes the correct movement, and gradually improving performance. Once the behavior has changed to what is correct, the error that drove the change is eliminated, the climbing fiber returns to its random background firing, and the remaining potent parallel fibers are left to drive the system in the particular movement context.
Context-triggering, fan-in, and fan-out. Nearly a million parallel fibers contact the human Purkinje cell. Over mossy fibers, vestibular, somatosensory, visual and auditory sensory information arrives in medial and intermediate zones, and cerebral cortical (presumably "cognitive") information in the lateral zone cerebellar cortex. There, the information is conveyed to parallel fibers. But the parallel fibers are so long (Mugnaini, 1983), that any one Purkinje cell could conceivably receive via parallel fibers the mossy fiber information from all three zones. The input--and the context that any one Purkinje cell "sees"--could include virtually all the sensory modalities, all the feedback from movement of each of the moving body parts AND the feedforward of the plans for the control of each body part, as well as the environmental conditions as the subject perceives them and contemplates what to do about them.
Movement sequences. Learned context triggering also suggests the possibility of generation of sequences of behavior automatically across time. It has been said that "reflex chains" as Sherrington envisaged were improbable because of the time involved (see Keele, 1981). Thus, in the sequence of notes A,B,C,D,E,F,G, E cannot be triggered by feedback from the movement D, because it would take too long to run over real neural pathways. But if E could be triggered by the to play B or C, and each of the notes could be triggered by the pre-conditions for some of the earlier elements in the sequence, then the process might be made to play out with realistic speed. The phrase--or indeed an entire piece--might thus be generated automatically by preceding contexts. There is evidence that behavioral sequences do show such linkages, and are generated in "chunks" of linked elements, whose linkages are less dissociable than are the boundaries of the chunk (see Keele, 1981). In this scheme, the triggering information would come from several different levels in the hierarchical scheme, and this is consistent with the observations from psychological experiments that different chunks are controlled from different levels.
Combining the actions of many muscles. Moving the skeleton is an engineer's nightmare. Over 100 bones are angulated at connecting joints which allow 2 or 3 degrees of freedom by over 600 muscles. In the simplest possible case, if one Purkinje cell controlled one muscle, and one muscle controlled only one joint, and each joint were controlled only by 2 muscles (agonist and antagonist), then moving three contiguous joints could be programmed by summing the weighted activities of the 6 Purkinje cells (and the nuclear cells to which they project). There is evidence that a single Purkinje cell may influence a single muscle (Simpson, 1994). But the actual computations must be much more complex, because most muscles control 2 joints, and a single joint may controlled by many more than 2 muscles. Yet, in principle, the actions of the muscles could be programmed by linking and weighting the activities of Purkinje cells (and their deep nuclear cells). This combining of muscles is but one level of somatomotor coordination (Thach et al, 1992, 1993), and one that may be addressed at the level of the MPG.
Preventing errors due to interactive torques: feedforward predictive control. Yet there is another aspect of somatomotor coordination, beyond summation of muscle activities. For every action, there is a reaction. Nowhere are the consequences of this law more evident than in the efforts to coordinate the actions of a skeleton. Reaching out to touch an object throws the body backward at hip, knee and ankle. Extending the elbow causes the shoulder to flex; flexing the shoulder causes the elbow to extend. Yet these reactions and inter-joint torques are compensated, and effectively, by contractions of muscles of which we are usually completely unaware. The preventive actions are driven by afferent information from the moving limb; they are missing in patients with severe peripheral neuropathy (Sainburg et al, 1993) and cerebellar damage (Bastian et al 1995). Yet these operations are not in the category of strictly linear feedback control, because the neural conduction is too slow for such a process to work. The preventive actions are driven by the movement but early enough in the movement for the corrective muscle torques to exactly match and nullify the passive interactive torques. Context-response linkage through learning is the way the cerebellum provides a mechanism for accomplishing this compensation. Each muscle's action is conditioned by other muscles' actions, and vice versa. In pinching, closure of the thumb on a grape is conditioned by closure of the index on the grape, and the sensory information monitoring the contact of each digit. The forces and positions of each digit are in part functions of each other: they are not controlled independently. While the exact "fanin" information on a Purkinje cell putatively controlling the thumb flexor is not known, the smear of sensory representation in the studies showing "fractured somatotopy" at the level of the mossy fiber terminals onto the granule cell is consistent with this idea (Bower and Nelson, 1992). The amount and variety of input information, and the number of granule cells and their synapses on Purkinje cells begin to make sense when considers them as a candidates for a "lookup table" in this kind of learned automatic control.
EXPERIMENTAL SUPPORT FROM ANIMAL STUDIES.
The different techniques mentioned in the introduction for studying systems neuroscience have been fruitfully employed to test the motor learning theories.
Ablation-behavior correlation. Cerebellar cortical ablation prevents or abolishes motor learning. Ablation of the cerebellar cortex has completely removed adaptation once it has been established, and has prevented any further adaptation (Ito et al., 1974; Robinson, 1976; Yeo et al., 1984; cf Thach et al, 1992) (Cf. Ito, 1984; Thompson, 1990)
Neuron discharge-behavior correlation. Purkinje cell recording during motor learning gives results that are consistent with the Marr-Albus-Gilbert theories. During familiar movements in the awake alert animal, the mossy fiber-parallel fiber system drives the Purkinje cell to modulate its simple spike firing frequency over a range of 0-500/sec; the climbing fibers fire at around 1/sec, and in no relation to movement. During movement adaptation, the climbing fibers fire in relation to the movement, but then decrease as the adaptation is complete (Gilbert and Thach, 1977; Ebner and Ojakangas, 1993); simple spikes (caused by parallel fiber action) conjoined with natural climbing fiber activity change over performance and time as predicted by the Albus hypothesis (Gilbert and Thach, 1977; but see Ebner and Ojakangas, 1993, and reviews of Thach, 1980; Ito, 1984, 1989; Thach et al., 1992). In optokinetic reflex eye movements, climbing fiber discharge signals eye movement error (Simpson and Alley, 1974).
Electrical stimulation conjointly of climbing fibers and mossy or parallel fibers reproduces the synaptic modification proposed to underlie motor learning. Paired stimulation of climbing fibers and mossy fibers gives results consistent with the Marr-Albus-Gilbert theories( Long Term Depression). Conjoint electrical stimulation of climbing fibers and parallel fibers repetitively over time decreases the strength of those parallel fiber-Purkinje cell synapses while sparing others (Ito et al., 1982; Ekerot and Kano, 1985).
Classical conditioning is dependent on the cerebellum. One of the most influential contributions in recent years has been the work of Thompson and colleagues on a cerebellar role in classic conditioning of the nictitating membrane response in the rabbit (McCormick et al 1981; Thompson 1986, 1990; Thompson and Krupa 1994; Topka et al 1993). A puff of air to the cornea (the unconditioned stimulus) coupled with a tone (the conditioned stimulus) will lead after time and repetition to an eyeblink to the tone alone. This classic conditioning is prevented (Steinmetz et al, 1992) or at least greatly altered (Welsh and Harvey, 1989) by cerebellar ablation. While there has been debate whether the critical lesion is in the interposed nucleus or in the cerebellar cortex (Yeo et al, 1984) this work was the first to clearly claim a cerebellar location for classical conditioning. It is not yet clear whether this represents the de novo creation of a new pathway from stimulus to response, or the adaptation of a latent one, such as the acoustic startle response (Ito, 1984; cf Mortimer, 1973; Leaton and Supple, 1986). More recently, it has been shown that the learning can take place and remain localized within the cerebellum even though it is temporarily kept from being expressed (Krupa, et al 1993). These observations critically uphold the Marr-Albus hypothesis.
PREDICTIONS ABOUT WHAT TO EXPECT IN CEREBELLAR PET AND LESION STUDIES OF MOTOR LEARNING.
PET and functional MRI human brain activation. Radiologic brain imaging is the most exciting and promising approach to understanding human brain function of the past several decades. Not only can it safely confirm in humans the results of invasive studies of sensory and motor functions in animals, it can get at the emotional, perceptual and cognitive functions in a way that animal studies cannot. Measurement of changes in blood flow or metabolism (the indirect indices of neural activity) can be correlated directly with known and controlled human mental activities. Human PET and animal invasive single unit recording studies are not competitive; nor is either sufficient itself. Each supports and complements each other's critical strengths and weaknesses.
It is easy to point out the weaknesses of PET scanning. The subtleties behind the conventual displays of "hot spots of neural activity" surrounded by volumes of "quiescent brain" are more than meet the eye. The primary data are not measurements of neural activity, but rather of factors which are thought to correlate with neural activity. The primary data is positron emission or magnetic resonance that varies from background in proportion to local alterations in blood flow or sequestration/uptake of oxygen or glucose. The changes in blood flow or metabolism are usually small, 2% or so of baseline activity, and often require averaging across subjects and other statistical manipulations to be detected at all. Commonly the subject is scanned during a "rest" state and again during a behavioral "test" state, and the scans of the two states are subtracted to detect the differences, if any. Increased blood flow and glucose uptake are thought to be much greater in synapses than in somata or dendrites (Collins and Wooten). Thus a "hot spot" on a PET scan reflects increased activity in the terminals of inputs to that area, more so than the activity of dendrites and somata within the area, and therefore the output from the area. For example, increases in climbing fiber firing should cause a "hot spot" on the PET scan in the cerebellar cortex. But this doesn't mean there is an increased output from the cerebellar cortex: from alert monkey and cat studies, the commonest immediate and long term effect of increased climbing fiber input is a reduction in Purkinje cell output from the cortex (Gilbert and Thach, 1977; Ito et al., 1982; Ito and Karachot, 1989). Furthermore, the increased blood flow or metabolic activity that is correlative with an increased nerve terminal firing is independent of whether those synapses are excitatory or inhibitory. If two inputs to a nucleus (e.g. dentate), the one excitatory (e.g. mossy fibers) and the other inhibitory (e.g. cortical Purkinje cells), are both active, the PET scan should show a "hot spot" in the nucleus. But the inputs might cancel, so as to lead to no change in the output from the nucleus. If only the Purkinje cell inhibitory input to the nuclei were active, the nuclei would still show as a "hot spot". But in this case, the normal tonic firing of the nuclear cell output would actually be reduced. The only way to infer increased or decreased activity in the nucleus per se is to look at its efferents and their synaptic termination on its downstream targets. For the dentate nucleus, these target sites would include the VLc, VLps, VPLo, X, MD, and CL subregions of the thalamus, the parvo-cellular red nucleus, the nucleus tegmenti reticularis pontis, the principle olive, and the superior colliculus.
But each of these areas receives inputs from areas other than the dentate. The influence of activity in dentate (or any other deep cerebellar nucleus) and thus cerebellar output depends on seeing increased blood flow or glucose or oxygen uptake in these structures that is not attributable to activation by some non- cerebellar input.
Because of these many caveats, and because of the many differences in imaging experiments across labs, the main initial test of validity is one of consensus. What a number of PET scan studies have now shown is that there are changes in blood flow and metabolic activity in the human cerebellum during several examples of "motor learning" (Fig.2). They have also shown that the cerebellum is active only in concert with a number of other parts of the cerebrum and brain stem, all together forming a network that appears to be active during motor learning. Perhaps most important, this network is more active during the learning than during the previously learned and now stereotyped "automatic" motor performance of the task. This final observation has been used to argue for a role that the cerebellum is specialized for motor learning per se.
Human brain ablation complements PET activation studies. As with strategies for imaging brain function, it is not the purpose here to review in any detail the pros and cons of ablation. It is the oldest technique for learning about brain function; it is still crucial in determining whether and how a brain part may implement a behavioral function. Ablation uniquely supports and checks the activation results of the PET or the animal unit study. Activation studies may show that an area is active during a behavior, but not that it is contributing to that behavior. Rather, it may instead contribute to some other coincidental behavior. Ablation of the part and a resulting change in behavior helps to support the causal connection. But like any technique, ablation has its limitations. Impairment of function following focal ablation unequivocally demonstrates involvement of a part in a function: it does not show that the function resides within the damaged part, and there alone, nor does it necessarily say what the part contributes to the function. Nonetheless, it may provide a clue. This is particularly so in humans, who through communication may permit a more detailed and introspective analysis of essentially what the normal behavior that ceases after the lesion (the negative deficit) and what abnormal behaviors may occur in its place (the positive deficit).
Specific predictions for PET studies, based on theory and the results of animal studies (Fig.3):
1. At rest (without movement), motor, premotor, prefrontal cortex and SMA neurons are relatively inactive, firing at rates below 10 per second or so. These should register as "inactivity" on PET scanning.
The vestibular and spinocerebellar inputs to the medial and intermediate cerebellar cortex are tonically active at rates of 30- 50/sec., and with the tonic activity of the intrinsic inhibitory neurons cause these portions of the cerebellar cortex to be intermediately active. The nuclear cells should also show moderate activity, since their Purkinje cell inhib itory inputs fire at their baseline frequencies of 50-70/sec, and their mossy fiber inputs are tonically active in a similar range of frequencies.
2. For learned, familiar movements, motor, premotor cortex and SMA neurons should be fully active, firing at up to 200-300/sec.
In the cerebellar cortex, mossy fiber inputs to medial, intermediate and lateral cortex neurons will be active in relation to movement at 0-200-300/sec. This will probably show up in the PET scan as an increase in activity over the tonic baseline. The same will be true for the deep nuclei, receiving from both mossy fibers and Purkinje cells.
If the movement is indeed overlearned and performed "automatically", Brindley would have predicted that the prefrontal cortex would be inactive.
3. During learning, according to Brindley, prefrontal cortex for the first time will be fully active, as it thinks about, pieces together and initiates the movement. In this phase, the movement is "consciously" made, with greater awareness of the details of the movement.
Cerebellar cortex--medial, intermediate, and/or lateral (depending on the nature of the task that is being learned) will show greatly increased activity, because of the increased discharge of the climbing fiber. Animal studies have confirmed the Marr-Albus prediction that under this condition the inferior olive and the climbing fiber will increase firing frequency from their low baseline, in order to "learn" the context in which prefrontal cortex is ordering up the performance, and seize control itself in an "automatic" mode. Since blood flow is proportionate to synaptic activity, and since the climbing fiber-to-Purkinje cell is one of the most massive synaptic structures in the nervous system, the cerebellar cortex should be highly active in those areas controlling task performance. This activity will continue only as long as the subject is learning--according to the theory and the evidence from animal experiments, it defines this phase.
Motor and premotor cortex and SMA activity will be the same whether commanded by prefrontal cortex in a "conscious" mode or by cerebellum in an "automatic" mode. These areas should be fully as active as during learned familiar performance.
4. During learned "automatic" performance, the activities become what they were for familiar performances. Intermediate and lateral cerebellar cortex will be moderately active (mossy fiber-parallel fiber-Purkinje cell synaptic activity controlling performance; climbing fiber activity at low-level baseline). Different sets of synapses will be active in controlling the newly acquired behavior, the overall level of activity will be the same as for the previous "familiar" performance.
Motor, premotor and SMA cortex will be fully active, as the execution of the task requires their full participation.
Prefrontal cortex becomes inactive, as task control has passed over to the cerebellum and an "automatic" mode.
Specific predictions for human cerebellar lesion studies: inability or slowness in:
1. Linking a novel context to the motor pattern it should trigger. Movements will continue to be slow, irregular, and requiring of mental effort, as they are normally when first attempted to be learned.
2. Acquiring a novel combination of muscle actions in the triggered motor pattern. Muscle components will be grossly irregular in their presence (or absence) and time-amplitude profile in the attempted combination.
3. Showing any improvement resulting from practice. The "learning curve" will be retarded or flat.
4. Patients with prefrontal lesions may also show inability to learn a novel task. However, their behavior should distinguish them from that of patients with cerebellar lesions. Prefrontals should fail or be delayed in initiating even the first trial, and any subsequent trial. Nonetheless, they may "automatically" initiate well-learned movements, upon presentation of the appropriate context.
PET STUDIES OF MOTOR LEARNING IN HUMANS
Manipulation and "tactile learning". The first PET studies of cerebellar O2 consumption doing a learning paradigm were performed by Roland and colleagues (Roland 1987; Roland et al., 1988). The tasks consisted of "tactile learning" and "tactile recognition". In tactile learning, subjects used the right hand to manipulate small metal objects of similar size and weight, so as to learn to be able to recognize them by touch. In tactile recognition, the learned objects were presented interspersed amongst similar "novel" objects. The goal was to distinguish the learned from the novel by manipulation and touch. Both task scans were compared to a "rest" scan where no movements or discriminations were made. Video analysis showed that the subjects manipulated objects on the discrimination task about two times faster than during the learning task.
Comparing the rest with both the learning and the recognition scans showed similar patterns of rCMRO2 increase in the latter two conditions. Bilaterally, there were increases in "six" different areas in prefrontal cortex, supplementary motor and premotor areas, anterior insula, lingual gyri, hippocampus, basal ganglia, and parasagittal portions of the anterior lobe of the cerebellum and lateral portions of the posterior lobe. On the left, there were increases in primary motor and sensory areas, the anterior superior parietal lobule, and the secondary somatosensory motor area. These areas are known from previous studies to participate in movement and sensation. In this study, there were "no differences in the anatomical structures participating in storage and retrieval".
But in comparing the learning and the recognition scans, "the rCMRO2 was significantly higher in the neocerebellar cortex during tactile learning", while "the CMRO2 increases in the left premotor cortex, supplementary motor area and left somatosensory hand area were larger during tactile recognition". The authors attributed the higher rCMRO2 during learning in the cerebellum to the increased climbing fiber activity that had been seen in animal studies on motor learning (c.f. Roland, 1987, discussion; Gilbert and Thach, 1977). They attributed higher rCMRO2 during recognition in the various motor sensory areas to the higher rates of movement during that condition. Although movements increased in frequency through the "tactile learning" training procedure, they could identify no motor learning per se. They therefore attributed the learning to tactile discrimination. These results are diagrammed in Fig.2.
Learning sequences of single finger movements. Nevertheless, the first PET study specifically addressed to motor learning in man soon followed (Seitz et al., 1990). In the task to be learned, the right hand was used in touching the tip of the thumb (1) to the tips of the other four fingers in the sequences 3 2 2 4, 4 4 5 5, 5 5 4 4, 4 3 2 2, the sequence having been instructed verbally. One PET scan was made at rest, a second when subjects had had little practice ("initial learning"), a third after 50 min of further practice ("advanced learning"), and the last after another 50 min of training ("skilled performance"). Over the three movement scans movement speed nearly doubled. Errors dropped to "very few" in the third stage and to "none in less than 6s" in the fourth stage.
Areas of the brain to show changes in rCBF were similar to those in the previous study. During initial learning (compared to rest), increases were seen bilaterally in the inferior frontal gyri (= premotor of prior paper?), in the right parasagittal anterior lobe of the cerebellum and the right lateral posterior lobe of the cerebellum, left greater than right premotor, supplementary motor and primary motor areas, left primary and supplementary somatosensory areas, anterior part of the superior parietal lobule, cortex in the intraparietal sulcus, and the left ventral posterolateral thalamus. Changes in subcortical structures included the basal ganglia, ventrolateral thalamus, red nucleus, region of the substantia nigra and of the pontocerebellar nuclei (oddly, mid-sections of putamen-globus pallidus, the red nucleus region, and the pontine nuclear region showed decreases in rCBF rather than increases).
As training proceeded through stages, there were progressive changes in the above pattern (Fig.2). The bilateral activity in the inferior frontal gyri dropped out entirely. The authors noted: "At this time the subjects said they no longer needed to count internally the number of times the fingers should touch the thumb". This is a crucial insight, that could only have occurred in human studies: the significance will become clear. There was also diminution of the increases in somatosensory association cortices, more so on the right side (ipsilateral to the finger movements).
There was an increase in the rCBF in left (contralateral) primary motor hand area. The authors attributed this to increased speed and frequency of finger movements.
The increased rCBF of the right (ipsilateral) anterior lobe of the cerebellum remained at the same level from learning to learned performance. (There was no change in the activation in the left primary somatosensory hand area or in the ventrolateral and postero-lateral thalamus.) The authors interpreted the abrupt initial increase of the cerebellar activity at the initial stages of motor learning again to increased climbing fiber activity (seen in animal studies). The decreased rCBF seen initially in mid portions of the regions of putamen-globus pallidus, red nucleus, and pontine nuclei-diminished with practice: that is, rCBF increased in these areas from the drop during "initial learning" to the state during skilled learning (=learned performance).
This study again suggested that a battery of brain parts was involved both in motor learning and motor performance. Again it was not clear which part was responsible for what aspect of the learning, and what essentially was learned. Of note was the activity of some areas at the start of the learning process. For inferior frontal cortex, the increase was absolute: it dropped out entirely during practiced performance. For the cerebellum, it was relative: the increase in activity at the beginning was maintained in relation to a rising movement frequency and speed and a rising rCBF in motor cortex. The cerebellar increase was inferred to be relatively greater for the learning per se than for the performance per se. Inferior frontal cortex and the cerebellum then must have had preferential roles in the learning per se. The authors speculate that the inferior frontal inclusion and subsequent drop out may have been due to the trained use of remembered linguistic instructions (verbal rehearsal) before the performance became automatic. The maintained activity in the cerebellum was consistent with Brindley's idea of a cerebellar role in motor learning (see below). The initial state of volitional conscious control, where the cerebellum acquires control of performance through learning, gives way to a final state of automatic control, directed by the cerebellum.
In the subcortical areas, the drops in activity from rest to initial learning could be consistent with their known high firing frequency at rest. A basal ganglion role in the automatic control would be suggested by the putamen-pallidal resurgence, and a cerebellar role by the resurgence of red nuclear activity as the task became learned. We shall return to this point.
Learning to play the Tetris Video game. The next work addressed learning of a complex visuo-motor task (Haier et al., 1992) (Fig.2). This used fluoro-deoxyglucose (FDG) uptake during learning and performance of the Tetris video game, compared and contrasted with the passive watching of numbers as they appeared on the video screen. The Tetris game required that subjects use the right hand to push keyboard buttons that would orient, move and place groups of video screen blocks amongst larger groups of blocks so as to complete an overall block figure (a straight line of blocks across the bottom of the video screen). Subjects doing the Tetris task were first given three minutes of practice at the game to learn the rules and the goal. Then they had their ("Naive") test trial in which they played for as high a score as possible for 35 min. while FDG was injected. They then had the "Naive" scan to determine the distribution and amount of sequestered FDG. After 4- 8 weeks of 30-45 min per day, 5 days a week of playing Tetris, they had their "Practiced" test trial during which they played for 35 min while FDG was injected. Following this performance they had their "Practiced" PET scan again to localize and quantitate sequestered FDG. With practice, the performance scores improved dramatically across all subjects. Subjects continued to improve with time out to almost 50 days of practice.
Naive and Practiced Tetris Scans were then both compared to scans taken after 30 min of passively watching numbers appear on a video screen, and with each other:
Comparing Tetris (Naive and Practiced) with passively viewing numbers, the former showed increases in the right inferior frontal and right superior temporal gyrus, and the cerebellum. The latter showed greatest FDG uptake in left occipital, left supramarginal gyrus and right area 17, with lesser increases in left postcentral and bilateral precentral cortex.
Comparing scans of the naive (i.e., learning) with the practiced Tetris performance, the former showed more activity in the left superior frontal cortex, left anterior cingulate and right posterior cingulate gyrus, left anterior and middle cerebellar cortex, and right posterior cerebellar cortex. the latter scan showed more activity in the right area 18, right hippocampus, and left cingulate.
This study concluded that with practice there was a greatly diminished uptake of glucose across all the structures that had shown an increase during "naive" performance (Fig.2). These structures included not only the cerebellum and the prefrontal cortical areas, but the primary motor cortex as well. This was despite the fact that the amount and frequency of movements had increased.
Learning sequences of single finger movements, controlled for frequency of movement. In a PET study of rCBF during "motor practice and neurophysiologic adaptation in the cerebellum", Friston et al., (1992) compared three pairings of rest with "right handed, brisk sequential finger to thumb opposition with each digit (2 to 5) in turn...To prevent gross performance changes over trials, the movements were entrained by a metronome at three per 2s (presented only in the task condition) (Fig.2). No measurements of task performance were made. The subjects were familiarized with the task 30 min before scanning, but were allowed to practice. Each motor activation started 30 s before administration of radioactivity, and lasted 2 min. In all scans the eyes were closed".
There was no "global" (overall) change in brain blood flow across the six conditions. In the first motor practice scan, there were relative increases in regional blood flow in the following areas: left sensorimotor cortex and bilateral cerebellar cortex, right greater than left. Active to a lesser extent were premotor cortex (left greater than right), supplementary motor area (left greater than right), the left putamen, the left lateral thalamus, and the cerebellar nuclei. (There was also bilateral activation of the primary auditory areas and nearby insular cortex. This was attributed to the sound of the metronome.
With practice over the sessions, there was an attenuation in the activity in the "right lateral cerebellum" and in the "medial cerebellar structures". They suggested that the lateral cerebellar adaptation was centered in the cerebellar cortex and that the medial location was deep to the cortex at the level of the deep cerebellar nuclei. Less extensive interactions (adaptations) were also seen in the right brainstem at the level of the inferior colliculus and in the left SMA. "There was no evidence for an interaction in the putamen or in the thalamus at this threshold".
The authors concluded that during motor learning the cerebellum commences with higher activity, which then drops off with practice (Fig. 2). They like Roland (1987) and Seitz et al. (1990) attributed this to the transient increase then decrease of climbing fiber activity reported in animal studies during motor learning (Gilbert and Thach, 1977). Of note, the frequency of finger movements was controlled, and the sensorimotor cortex activity remained constant.
This study focused on activity in the cerebellum, and it is not clear whether they would have seen changes in prefrontal cortex.
Learning to track a rotating disk. A somewhat different result was seen in the following study of Grafton et al., (1992) (Fig.2). They used PET to study rCBF during "procedural learning" in which subjects were asked to use the right arm and hand to move a stylus to track a 2 cm target on a 20 cm small disc rotating at 60 rpm. During the first and sixth (control) scans, the subjects tracked the target with eyes only (EOM not monitored). During the second through fifth scan, the subjects tried to keep the stylus on the moving target. In between test trials, there were practice sessions. Performance was monitored and dramatic improvement was documented with practice from 17% to 66% mean time across the four PET scans obtained during pursuit rotor performance.
"Motor execution was associated with activation of a distributed network involving cortical, striato-nigral and cerebellar sites" (Fig.2). As performance improved with practice, increases in rCBF occurred in the left primary motor cortex, supplementary motor area, and pulvinar thalamus. By contrast, there was no change in an initial increase in rCBF in cerebellum. The authors contrasted this with the previous cited study, noting that the task was more difficult, and not enough practice had been allowed for performance to become "automatic". They concluded with an expression of their intent to study performance after prolonged practice to the point of automaticity to see if then the cerebellar activity diminishes as previously.
Learning sequences of single finger movements, controlled for frequency of movement, and learning vs performance. The most recent paper (Jenkins et al., 1993) used PET to measure rCBF during "motor sequence learning" (Fig.2). The task was similar to that of Friston et al., 1992 and attempted to control and demonstrate motor learning and to distinguish it from performance. Thus subjects again had a period of rest, a period of performance of previously learned sequences of thumb-to-finger tapping at clocked intervals, and a period of trial and error "new" learning of a novel sequence.
During both new-learned (and learning) and pre-learned (and practised) performance: the contralateral sensorimotor cortex was activated and to the same extent for both conditions.
Prefrontal cortex (right greater than left) was active only during new learning.
"The cerebellum was activated by both conditions, but the activation was more extensive and greater in degree during new learning. Similarly, lateral premotor cortex was more active during new learning.
By contrast, the supplementary motor area was more active during performance of the pre-learned sequence than of the learning sequence. Putamen like primary motor cortex was equally active during the two conditions.
These authors again concluded that "the cerebellum is involved in the process by which motor tasks become automatic".
In summary of these PET studies (Fig.2), one shows higher cerebellar activity in "tactile" learning compared to practised performance, even though movement is greater in the latter (Roland, 1987, Roland et al., 1988). In the second, cerebellar activity remained the same from motor learning to motor performance despite increased movement in the latter, from which the authors inferred a relative increase in relation to the learning phase (Seitz et al., 1990). In the three studies in which movement was controlled (Friston et al., 1992; Grafton et al., 1992; Jenkins et al., 1993), cerebellar activity was greater in novel than in prelearned tasks (Friston et al., 1992; Jenkins et al., 1993) or remained elevated (Grafton et al., 1993).
Raichle et al., 1994 pointed out the difficulty in comparing these results, where the tasks and methods differed so widely. The crux is in keeping the movement performance constant and adequately dissociated from the motor learning. Since the cerebellum may be involved in both, dissociation is needed to demonstrate the extent to which it participates in each. In the above studies, this aspect is unclear. Of the three studies that controlled for the amount of motor activity, "one noted no change in primary motor cortex (Friston et al., 1992), one noted a decrease (Mazziotta et al., 1991) and the other noted an increase (Grafton et al., 1992). In two studies that did not control the amount of motor activity and, consequently, observed an increase in the number of acts performed during the same scanning session, one reported an increase in the activity in primary motor cortex (Seitz et al., 1990) and one reported a decrease (Haier et al., 1992). Two of the groups that controlled motor activity commented about changes in SMA but reported opposite results; Grafton et al. (1992) reported an increase whereas Friston et al. (1992) tentatively reported a decrease. The varied results reported in these five imaging studies of motor learning in normal humans do not permit us to draw any conclusions about consistent changes in neuronal activity or organization in primary motor or supplementary motor cortex to be expected from practice of various motor task or to anticipate our results" (Raichle et al., 1994).
Nonetheless, the studies do appear to show parallel changes in prefrontal cortex and cerebellum in going from a learn-ing to a learn-ed performance. This is exactly what Raichle et al. (1994) subsequently observed in their study of "non-motor" learning (Fig. 2). Cerebellar activity is increased during the learning, and is diminished or absent during the over learned or automatic performance.
Yet Raichle's warning is appropriate: with this much variability across studies, what can one conclude? An obvious caveat is one given by this same group on the need to watch out for coincidental behavior and muscle activity that is not part of the main task performance. In a study of panic disorder, in which increased temporal blood flow was attributed to increased activity in limbic regions of the brain, it was subsequently found to localize instead to temporalis muscles, which were fortuitously active in the near vicinity (Drevets et al., 1992). The same could hold for regions of the brain that are active and covary with a particular task performance, but whose causal connection is to some coincidental behavior and muscle activity that is not at all necessary to task performance. In animal studies, one is always on the lookout for coincidental "suspicious behaviors" and consequent spurious correlations between unit activity and the primary task performance. And indeed, the cerebellum may be particular active in relation to synergistic associated movements (Schieber and Thach, 1985a,b; Thach et al., 1993b). Perhaps too much trust is placed in human subjects following the investigator's instructions and wishes to the letter; certainly, spurious correlations seem to be a possibility (Drevets et al., 1992).
But here is where ablation studies can come to the rescue, and indeed, many have recently supported a cerebellar role in motor learning.
CEREBELLAR DAMAGE IMPAIRS MOTOR LEARNING
Adaptation of pointing while wearing magnifying lenses. Following the Brindley-Marr-Albus-Gilbert theoretical suggestions for a cerebellar role in motor learning, and the early animal experiments showing impairment by lesion and unit discharge correlation with motor learning, the first study showing human relevance was that of Gauthier et al. in 1979 (Fig. 3).
In normal humans and in patients with "posterior fossa involvement", they examined visuo-motor performance in an eye-hand pointing task as it adapted to the wearing of magnifying lenses. When the target field was thus visually magnified, the normal response was to misreach by a distance proportionate to the magnitude of the magnification factor. With practice, and given knowledge of the error (open loop), normal subjects adapted sufficiently to compensate for about half the prism-induced error. Proof that a true adaptation (not just a change in strategy) had occurred came upon removal of the lenses: normal subjects pointed and overshot the target by a similar magnitude in the opposite direction. The patient had had prior superior vermal tumor removal, with accompanying loss of superior vermal and adjacent cortex, had had transient hydrocephalus, palato-pharyngo-paraspinal and diaphragmatic myoclonus. He had had a persistent cerebellar deficit that included saccadic hypermetria, eccentric gaze holding nystagmus, rebound nystagmus, scanning speech, symmetrical limb dysmetria, dysdiadochokinesia, Romberg sign, truncal titubation, and broadbased staggering gait. The patient was unable to recalibrate his gain during the visual-motor training, and he alone had no post-lens exposure overshoot. The authors concluded: "we believe that the persistent motor deficit of the cerebellar type displayed by patient 1 parallels the absence of visual-motor adaptation. Such adaptation clearly involves recalibration of visual-motor coordination. The same patient formed part of a study of saccadic overshoot dysmetria interpreted and modeled as a deficit in recalibration of oculomotor gain. Other cerebellar signs may have related explanations".
They suggested that the preserved performance of the saccade trajectory and of Hering's law of conjugacy implied the preservation of the fundamental oculo-motor control mechanisms in the brainstem. By contrast, the prior episode of palatal myoclonus suggested to them an impairment of the inferior olive, which the Marr-Albus theory and the experimental results of Llinas et al. (1975) and Gilbert and Thach (1977) had implicated in motor learning. They agreed with all these authors that these inputs to the cerebellum and the cerebellum itself were thus involved in motor learning, and that their individual roles might not be further revealed by ablation alone.
Adaptation of pointing while wearing laterally-displacing wedge prisms. A similar experiment with similar results was performed by Weiner et al. (1983), who studied visuo-motor adaptation in a pointing task across a variety of neurological patients during the wearing of wedge prisms (Fig.3). With these prisms, the optic path is bent (e.g. to the right), and the subject has to look to the left along the bent optic path to see the target which is directly in front of him. In pointing, the arm, hand, and finger point in virtually the same direction as the gaze, and thus overshoot the target to the left. But if the subject can see either the site pointed to and thus the error within a second or so after the point, or if he can see the hand itself during the point, then there is a progressive adaptation with each subsequent point, in which the point comes gradually closer to the target. When the prisms are removed, and the eyes now look straight at the target, the adapted hand points to the right of the target in the opposite direction and of a magnitude similar to the original error. This occurrence of overshoot and its persistence, requiring re- adaptation with practice, is what convinces one that a true adaptation has taken place. The authors pointed out that the phenomenon was described by Von Helmholtz in the last century, and has been studied by a number of psychologists since as to its essential mechanism. They summarized the then-existing beliefs as "This adjustment is attributed to the combined effects of two processes: a true visual adaptation to the prisms and a cognitive correction in pointing when there is a perceived error, or the difference between pointing where the target is "seen" and where the target is thought to be" (Italics ours). The authors also reviewed the different thoughts as to the mechanism of the adaptation, which separated mainly into that of an altered sense of position or of posture (peripheral) and an altered perceptual or mental set (central). They also reviewed the various animal ablation experiments seeming to exonerate corpus callosum, hippocampus, anterior and posterior commissure, massa intermedia, optic chiasm and [sic] cerebellum (Bossom and Hamilton, 1963). Bossom had reported reduced levels of adaptation after bilateral frontal lobectomy or bilateral caudate lesions (1965). Baizer and Glickstein briefly reported that cerebellar lesion in monkeys impaired prism adaptation (1974).
Weiner et al.'s study involved normal humans and patients with cerebellar disease, Parkinson's Disease, right and left cerebral hemisphere disease with affected corticospinal pathways and Alzheimer's Disease and Korsakoff's syndrome, the latter two both with "declarative" memory defects. Only the cerebellar patients showed "significant reductions in the after effect" (Fig. 3), which the authors like Gauthier et al. regarded as the best measure of true adaptation. The authors therefore likewise labelled the effect a visuo-motor adaptation. "Poor adaptation in cerebellar patients may be due to impaired motor learning, defined as a change in motor program after environmental change. Adaptation to lateral displacement of vision requires new motor responses in response to induced alteration in visual input."
Learning to trace complex figures and mirror tracing. This view was ostensibly supported by the experiment of Sanes et al. (1990) on normal controls and cerebellar patients with either focal or atrophic diffuse disease of cortex or with olivo cerebello ponto atrophy (OPCA) (Fig.3). The tasks were two: i) the tracing of a 5-sided "complex figure" repeated for 50 trials, and ii) the mirror-tracing of two different 4-sided figures, for 50 trials and 10 trials, respectively. The 5-sided figure was traced while being viewed directly; the two 4-sided figures only while being viewed as reflected in a mirror (and thus right-left reversed).
The details of the method and instruction are important. "The patterns were sufficiently large that subjects had to move more than the fingers and wrist to accomplish the task. Typically, the movement involved shoulder and elbow joint rotations, as well as adjustments of the wrist, and occasionally of the fingers. Speed of execution was emphasized as the most important movement parameter (italics ours). Subjects were instructed to slide the stylus across the data tablet as rapidly as possible, but with the proviso that each movement segment begin and end in the small squares that enclosed each vertex of the pattern. Subjects were cautioned not to pause at the vertices, but rather to execute the movement in one continuous sequential [sic] movement (italics ours). Subjects were informed that the lines connecting the vertices were intended as a general guide for the sequential movement."
A multivariate analysis was undertaken across patients and across conditions, in the hopes that distinct abnormalities would stand out. Measured variables included: 1) Movement time, measured for the entire four or five segment movement; and 2) Tracking errors, calculated as the average, peak, and endpoint deviations of the tracing from the target lines, and acceleration changes that crossed zero. Reversals of acceleration were necessary only at endpoints; more than that reflected lack of "smoothness" of trajectory. The authors state that the patients made many performance errors, and that normalizing and averaging corrections were required to extract errors in learning from errors in performance.
Common features across groups and conditions included a tendency to speed up with practice, more so for the normals than the cerebellar groups. All groups continued to make errors of the various types across tasks from start to finish.
Differences across groups and conditions were the tendency on direct vision tracing for the normals and to a lesser extent the OPCAs (but not the cerebellar corticals) to increase the average and endpoint error (though not the peak and tangential acceleration zero-crossing errors). This was regarded as a natural phenomenon in accordance with Fitt's law of the speed/accuracy trade-off: as speed increases, accuracy should fall off. The failure of the cerebellar cortical patients to follow this law was regarded as some failure to adapt to the normal strategy.
Another difference occurred on the mirror tasks and consisted of the relative inability of the OPCAs to reduce their tangential acceleration zero crossings, as compared both to normals and to the cerebellar cortex patients. This was interpreted as a relative inability of these patients to learn a new skill. This purported deficit was likened operationally to associational conditioning, a failure to achieve a link between a new or different behavior to a given stimulus. The authors suggested an analogy to the work of Thompson and colleagues on conditioned eyeblink in the rabbit (McCormick and Thompson, 1984; Thompson, 1986, 1990).
The authors interpreted that both the normals and the cerebellar cortex patients both adapted to the mirror task, because they both increased their speed and reduced their tangential acceleration zero-crossing errors. The authors also interpreted that there was a carry-over from the 50 trials of mirror learning on the first figure to the ten trials of performance on the second figure. They gave as the most direct evidence for this the fact that no normals or patients (except one) made direction errors on the first movement on the figure.
Greatly to their credit is the authors' serious effort to define what indeed is learned in motor learning and what essentially is impaired by cerebellar disease. Their careful presentation of their methods, results and interpretations allows one to interpret further. The first question one might raise is whether failure to obey Fitt's law is a specific neurologic deficit. The instructions given to the subjects to move as fast and as nearly continuously as possible (despite a segmented motor pattern task) clearly encouraged errors in approximating the target line. It is not at all clear why the OPCAs (who were slower) made fewer (when normalized) of these errors, but it does not seem quite appropriate to call the failure to make these errors a deficit.
The second question one might raise is to what extent any of the subjects really "learned" mirror drawing. Certainly, errors in performance persisted to the end. Further, in the claimed "carry- over" of mirror learning from the first mirror-traced figure to the second, one could question this interpretation. In mirror-tracing, movements parallel to the face of the mirror are not reversed, and no adaptation is needed to make them. Only movements toward or away from the mirror are reversed, and require adaptation. In the first mirror-traced figure, all lines had a component moving toward or away from the mirror. In the second figure, the first line to be drawn was parallel to the face of the mirror. This then was the only movement that a subject informed of the nature of the task but who was relatively poorly practiced could have made without errors. It is therefore not surprising that only one error (in a patient) was seen. The conclusion of "learning carry-over" seems questionable.
If the "carry-over" from the first mirror tracing task to the second is suspect, so then is the presence of an after effect or persistence of learning. An after-effect was clearly demonstrated in the prior papers on lens and prism adaptation, respectively. It is not clear what exactly was learned here, or if anything was truly learned. The alternative is that there was the conscious adoption of a strategy to reverse visuo-motor coupling proportionately as one moved from a plane parallel to the face of the mirror to one toward or away from it.
A further reason for questioning whether visuo-motor adaptation occurred is the relatively short time course claimed for it. An equivalent experience would appear to be that reported by Gonshor and Melvill-Jones ('76). In these experiments, dove prisms were placed over the eyes and worn continuously. These reversed the visual world in the left-right dimension without altering the up- down dimension, which would appear to be similar to but just opposite that of mirror drawing. What was measured was the phase and gain of the VOR; other motor behavior was not studied systematically. Changes in the VOR occurred gradually over several weeks, and did not fully reverse until after five weeks of continuous wearing. At this time other motor behaviors appear also to have adjusted: one subject resumed driving his motorcycle! Whether he would have attempted this after 50 short trials is doubtful. In sum, the task of learning mirror tracing seems a good one, but one likely to take a much longer time (Fig.3).
Learning to couple a novel postural response to a novel perturbation of stance. Another line of study attempting both to demonstrate a cerebellar role in motor learning and also the identity of the controlled variables is that of Horak (1990) and Horak and Diener (1993) on postural responses to the perturbation of stance (Fig.3). In this study, subjects included normals, bilateral vestibular nerve lesions, anterior lobe cerebellar cortex degenerations, and OPCAs. Subjects were instructed to stand on a platform which was driven backward by a motor. Two perturbation parameters were varied independently, velocity and distance of backward displacement: four velocities over a given distance, and four distances at a given velocity. Any and all pairings of a velocity and a distance were repeated over a block of 10 critical trials, so that the subject had the opportunity to experience and adapt to those particular conditions. Only the last three trials were analyzed in a block of trials, after ample opportunity for adaptation. Postural (corrective) responses were measured as torque generated under the toes and ball of the feet, (to prevent falling forward), and the EMG of various muscles in the leg and trunk.
Normals adapted to each of four different velocities (fixed distance) and four different distances (fixed velocity) with proportionate differences in plantar flexor torque and EMG amplitude. Both the vestibular and the cerebellar patients also adapted to velocity of displacement, and with gains (velocity/torque) similar to normals. However, both patient groups showed excessive bias, generating hyperactive responses. That is to say, plots of velocity (abscissa) and torque (ordinate) were linear and of the same slope for all three groups, but had different intercepts on the torque ordinate, vestibular and cerebellar patients being progressively higher. Thus, translation of stimulus velocity to graded torque response clearly continued to take place in both patient groups and with identical and normal gains. What seemed missing was an inhibitory tonic bias on brainstem and spinal circuits. This could be understood as the loss of Purkinje cells in the case of cerebellar cortex disease (the most extremely abnormal). In the case of relatively milder excess bias in the vestibular patients, the mechanism is less obvious, but was attributed to somatosensory compensation for the vestibular loss (Bles et al., 1984).
As for variations in distance of displacement, the normals and vestibulars adapted, the vestibulars again with normal gain, but biased high (generating a fixed excessive amount of torque for each torque increment proportionate to greater displacement). But the cerebellar cortical atrophies were distinctly different, showing no gain (a torque/amplitude plot with zero slope) and a high bias (intercept displaced high on the torque ordinate). This led to hyperactive responses with overshoot of the proper endpoint position, without any stimulus-response scaling. Since the magnitude of the displacement (at fixed velocity) could not have been inferred from stimulus parameters at onset, the magnitude of the response had to be learned through trial and error. The adaptation is one of association, and not of scale. The authors refer to it as the acquisition of SET, which is specifically impaired in the cerebellar cases.
Oddly, the OPCA patients, though presumably similarly "ataxic" in walking and heel-knee-shin tests, showed normal scaling of distance vs response.
Cerebellar patients showed excessive activity of antagonists, as well as of the agonists that led to the excessive responses and the hypermetria. This was interpreted as a compensatory cocontraction as if to restrain and reduce the response that would otherwise be even greater. Specifically, the authors denied that there was any disorder in the selection, sequencing, or latency of the muscles themselves. The interpretation was that this was exclusively a disorder of set (and bias), and not of coordination per se. Thus, the authors conclude " the anterior lobe of the cerebellum appears to play a critical role in modifying the magnitude of autonomic postural responses to anticipated displacement conditions based on prior experience. The cerebellum may modify the gain [sic] of the somatosensory loops for posture by adjusting the threshold, and not the slope, of stimulus/response relations. The cerebellum does not appear to be critical for using somatosensory information on displacement velocity to scale response magnitude, despite a hypermetric response offset. The major effects of anterior lobe damage in humans appears to focus on gain [they mean bias] and set control and not on disrupting the temporal synergic organization of multijoint postural coordination." As for bias and learning or "set" acquisition, this careful study shows that the cerebellum must be involved in at least these aspects of movement.
Adaptation of throwing while wearing laterally-displacing wedge prisms. Others have also shown that cocontraction is common in cerebellar disease (Hallett et al., 1975), which demonstrates an impaired selection of muscles appropriate to a task (cf. Thach et al., 1992, 1993b and below). Still other work has raised the old question of whether the role of the cerebellum is simply to provide control of bias (threshold) or gain (stimulus/response proportionality) of input output relations in downstream movement generators, or whether it might in addition combine the elements within and across generators (Thach et al. 1992). There can only be so much behavior that is hardwired within the nervous system. Any novel behavior, developed in response to novel environmental conditions, must obviously be "programmed" into the preexisting hardware. One kind of novel behavior is the novel synergic combination of muscle actions. The word "skill" is used for this novel acquired pattern of coordination. A paradigm that illustrates the learning of a synergy is the adaptation of eye-hand coordination in throwing a ball or a dart at a target while wearing wedge prism spectacles. In throwing at a target, the eyes fixate the target, and serve as the reference aim for the arm in throwing. The coordination between the held position of the eye and the synergy of the arm throw is a skill: it has to be developed and kept up with practice. If wedge prism spectacles are placed over the eyes with the base at the right, then the optic path will be bent to the right, and the eye will have to look to the left to see the target. The arm, calibrated to the line of sight, will throw to the left of target. With practice, the calibration changes, and the arm throws with each try closer to and finally on-target. Proof that gaze direction and eye position is in fact the reference aim for the arm throw trajectory comes when the prisms are suddenly removed and the arm throws. The eyes are now on-target, but the eye-arm calibration for the previously leftbent gaze persists; the arm throws to the right of target an amount almost equal to the original left error (Thach et al., 1992). But with practice, the eye position and the arm throw trajectory are recalibrated back to the original setting: the throws move closer back to and finally on-target. A good analogy is the relation between sighting and shooting a gun: the linkage between the sight and the bore trajectory is calibrated by adjustment, and kept true through practice.
But is this adaptation in the visual sensory domain, the perceptive, the cognitive, the motor, or somewhere in between? Our evidence that it is at least largely motor comes from its specificity for body part. When one arm is trained and the other arm is tested, the initial arm adaptation does not carry over to the opposite arm throws, but persists through to throws again by the initial arm, only to readapt with repeated throws (cf also Prablanc et al., 1975). The lack of carry-over of the adaptation to the untrained arm, and the fact that throwing with the untrained arm does not degrade the adaptation in the trained arm, both speak for privacy of the storage to use of the trained body part.
But can one train the body parts to a specific task, without the adaptation spilling over to other tasks performed by the body parts? When one arm is trained on overhand throws and then tested on underhand throws, for most individuals the overhand training does not carry over to the subsequent underhand throws Thach et al., 1992b). Yet the overhand training persists through to subsequent overhand throws, readapting with repeated throws. Prism adaptation is thus specific to the throwing arm and to the type of throw. This may be analogous to using a similar movement that is differently calibrated for two different contexts--such as hitting a baseball with a bat and a tennis ball with a racquet. One may modify the one activity, for example, using a heavier bat, and not have the training affect the other activity--one's tennis game. When the two movements become more nearly identical one is more likely to get carry-over. Some find it difficult to play tennis and squash alternately; more have trouble alternately playing squash and racquetball.
Can one learn to store more than the one gaze-throw calibration simultaneously? We asked subjects to throw 200 throws while wearing the prisms and 250 without daily, 4 days per week for 7 weeks (Martin et al., 1993). We measured the progress on the 5th day of each week with 25 throws before, 100 throws during, and 75 throws after wearing the prisms. This made a total of 900 throws with prisms and 1100 throws without prisms each week for 7 weeks. Over time and practice, the first throw with the prisms landed closer to the target, and the first throw without the prisms (aftereffect) also landed closer to the target. By 7 weeks, throws are on-target for the first trial wearing and the first trial after removing the "known" prisms. This suggests that 2 adaptations (no- prisms and known-prisms) may be stored simultaneously and separately. This may be analogous to the fact that one can maintain eye-hand coordination while wearing one's spectacles and when not wearing them. It also accounts for the period of adjustment required to "get used to" a change in ones lens prescription.
What was the controlled variable in these dual calibrations? The gaze-throw angle for no-prisms was 0 degrees, that for the known-prisms of 30 diopters, about 15 degrees. When the two overtrained subjects donned the known-prisms, their gaze-throw angle immediately shifted to 15 degrees; the context of the prisms must have introduced a bias within this system. The bias was compartmentalized and specific: components of the gaze-throw shift consisted of changes of position of eye-in-head, head-on-trunk, and trunk-on-shoulder (Martin, et al, 1995). In one subject, the relative proportions were 5, 5, and 5 degrees, respectively, for each component. In the second subject, the relative proportions were 6, 2, and 8 degrees, respectively, for each component. Thus, in each subject, varying amounts of tonic activity in specific muscles contributed to the compounded variable of static gaze. Each subject had learned two such patterns, which could be immediately changed from one to the other, depending on the behavioral context (prisms or no-prisms)
Baizer and Glickstein (1974) first showed in macaques trained to point to a visual targets while wearing wedge prisms that the adjustment mechanism was abolished by cerebellar lesion; Weiner et al. confirmed the result in patients with cerebellar disease, and showed further that adaptation was not impaired in disease of corticospinal or basal ganglia systems. We (Thach et al., 1991) have confirmed these results in the throwing task in patients with pure cortical cerebellar disease, and have also seen that patients with MRI-documented inferior olive hypertrophy and infarcts of olive output at the inferior cerebellar peduncle could not adapt, despite otherwise near-normal performance Fig.3). By contrast, patients with infarcts involving the dentate nucleus despite severe ataxia could adapt. This suggests that the adaptation mechanism could be dissociated at least in degree from those of coordination and performance.
What then is learned in "cerebellar motor learning"?. What basically does the cerebellum control in the motor domain? We have interpreted these studies as showing a marked capacity for storage of different types of adaptation for the same body part for different contexts and different movements. It is important that the different adaptations be kept private, without carryover from one to another. We have argued elsewhere on theoretical grounds that the best place to store such context specific adaptations is somewhere far "upstream" (Fig. 4) from the controlling motor pattern generators and the motor neurons. The reverse would be true for types of adaptation that should best apply to all uses of a body part. A weakening of a muscle would best be compensated by increased synaptic strengths of inputs to the alpha motor neuron. Damage of the labyrinth would best be compensated by increased synaptic strengths of vestibular nerve input to vestibular nuclei. In the various training paradigms that have been employed in animal studies, both or either kind (site) of adaptation may have been produced. Thus the controversy about the one exclusive of the other may result more from experimental method than from biologic restriction.
In sum, the experimental observations summarized here are compatible with the Brindley-Marr-Albus-Gilbert theory. A number of PET studies show activity in prefrontal cortex ("voluntary, conscious activity") at the very beginning of the task learning, only to drop out as performance becomes practiced. Cerebellar activity is maximal at throughout the learning phase, when learning theories predict and animal studies show increased discharge in the massive climbing fiber synapse onto the Purkinje cell. The activity drops sharply but DOES NOT stop as task performance becomes more automatic, theoretically under control of a "new" pattern of activity in parallel fiber synapses onto Purkinje cells. The "new" pattern of parallel fiber-Purkinje cell activity should be greatly different from the "old" one in its metabolic needs, and blood flow should be the same or similar. These observations are compatible with two of the most critical predictions of Brindley, Marr,and Albus, and agree with the findings in animal studies.
The ablation studies of Horak are compatible with a bias adjustment mode of control, and also of context-response linkage. The observations of visual perturbations during pointing of Gauthier et al. (1979), and Weiner et al. (1983), of Sanes et al. (1990) in mirror drawing, and of our own in throwing argue for a compounding of controls across body parts--call it bias, gain, or what one will--that generate movement combinations of such variety and novelty as to easily qualify as "learned new movements". In the prism-altered eye-hand coordination studies, the essential variable that is changed is the angle between the line of gaze and the line of the point or throw. The gaze-throw angle is in turn made up of angles of eye-in-head, head-on-body, and body-on-arm. The angles in turn represent the different tonic activities in the agonist-antagonist muscle pairs controlling the body part in question. These activities are cerebellar-controlled so as to sum and allow precise "coordinated" motor acts.
COGNITIVE FUNCTIONS OF THE CEREBELLUM
The cerebellum sits higher in the hierarchic organization of the central nervous system than it used to. From time to time, there have been suggestions that the cerebellum might play a role in one or another cognitive function (Cf. reviews by Rawson, 1940; Schmahmann, 1991; Macklis and Macklis, 1992). Ideas have ranged from memory (Poseidonius, 9th century AD--cf. Rawson, 1940) to sexual potency (Gall, 1800's--cf Macklis and Macklis, 1992). None of these suggestions have been incorporated in our knowledge of cerebellar functions, because they have not fit with other scientific facts. But there have been recent advances in our knowledge of connectivity, ablation syndromes, and natural activation correlations in animals and man which require us to reevaluate this question.
The so-called "motor association" (Adams and Victor, 1994) areas are presumably concerned with motor planning--these include the supplementary motor area (medial area 6), the premotor area (lateral area 6), the frontal eye fields (area 8) and the accessory frontal eye fields of Schlag, and the motor speech areas in man (areas 44, 45). Anatomically, they receive from posterior areas associated with perception and awareness (see below), and project to the middle level motor pattern generators. They receive multi- modal sensory inputs, and send to different movement generators. These parts are active in animal recording and human PET studies during the movement, and their ablation impairs movement. For these reasons, they have been called the motor "association" areas of cerebral cortex.
It has recently become known that these areas may be active in anticipating or rehearsing a movement without actually performing it (the studies of anticipatory signals in monkey SMA and PMC by Evarts, Tanji and Kurata (Tanji and Evarts, 1976: Tanji, 1985); of mimicry signals in monkey PMC by Rizzolatti (Di Pelligrino et al., 1992); and of mental motor rehearsal signals in human SMA by Roland (Roland, 1987). The two roles in purely mental imaging of movement and in movement planning would appear to go together.
Recent work has added projections from the cerebellum to include virtually all levels of the motor system--spinal motor and inter neurons (cf. Asanuma et al, 1983d), the superior colliculi (May et al., 1993), and (via thalamus) the cerebral cortical "motor association" areas--premotor cortex, primary and secondary frontal eye fields, and now even areas 44,45, and 46 (Schell and Strick, 1983; Lynch et al., 1992; Yammamoto et al., 1992; Matelli and Luppino, 1993; Middleton et al, 1994). As such, the cerebellum is "upstream" from movement pattern generators at all levels. But since each of these movement generators has other prominent excitatory inputs, and since cerebellar ablation impairs but never abolishes movement, cerebellar functions have traditionally been characterized as regulatory and modulatory, rather than executive.
Does the cerebellum only modulate what others have begun, or can it run the whole show? Nevertheless, these connections put the cerebellum in a position to excite any or all of the major motor generators, at every level from motor neuron to motor association cortex. Why has it thus been delegated to the role of regulator and modulator, rather than executor? The historic arguments include: 1) cerebellar lesions impair but never abolish movement, and often cause little or no observed motor defect; 2) until recently, cerebellar electrical stimulation has not been reliable in causing movement, and 3) cerebellar target motor structures each have an additional major excitatory input, which has been credited with the major driving effect. However, newer knowledge questions all three premises. First, focal cerebellar lesions do indeed abolish particular categories of movement without affecting others. The effect of the lesions depends upon its precise localization, and our knowledge of the particular region and its control functions. The effects may be disabling (cf Thach et al, 1992 for review). Second, electrical stimulation of the output nuclei does reliably cause movement (cf Thach et al, 1993 for review). Stimulation of the cortex may not, because of mixed excitatory and inhibitory effects. Third, cerebellar excitatory effects on a target structure are often stronger than those of the "second" excitatory input. This is known to be so for the red nucleus (cf Tsukahara, 1972). Also, the dentate nucleus fires before and apparently helps initiate output from motor cortex (Meyer-Lohmann et al, 1975). These facts shed a new light on the cerebellar control of movement, and on the so-called cerebellar motor learning theories. They suggest that the cerebellum may operate at the highest level of direction and coordination.
Language and non-motor learning. In their PET study, Petersen and colleagues (1989) found activity in the right cerebellar hemisphere to increase during a language task. Subjects were give lists of words to read and say, hear and repeat. Then they were given lists of nouns for each of which they had to "generate" and say an appropriate verb. Examples would be "nail" prompting a response of "hammer", or of "boat" a response of "row". Scans from the prior spoken tasks were then subtracted from scans during the "generate" task. This produce activation in the right cerebellar hemisphere, the left anterior cingulate gyrus, the left posterior temporal cortex, and in the left frontal lobe.
Raichle and colleagues (1994) have recently reported the effects on the PET scan of practice of the verb-generate task. In novel trials where word lists were presented for the first time, areas that specifically became active included left prefrontal cortex, left anterior cingulate gyrus, left posterior temporal cortex, and right lateral cerebellum. With continued practice, these areas dropped out, and other areas previously inactive became active. These areas included sylvian-insular cortex bilaterally and the left medial peristriate cortex. They commented: "These results indicate that normal subjects change areas of the brain used during tasks performance following less that 15 minutes of practice. One critical factor in determining the circuitry used appears to be the degree to which a task is learned or automatic".
Mental movement. There is growing evidence that premotor cerebral cortex and the cerebellum participate in imagined movement. We have already referred to the word-generate tasks studied by Petersen et al., 1989. These first showed activity in the right cerebellum hemisphere during silent speech--but only on some tasks. Decety et al., 1990 then showed that the cerebellum is activated during imagined movement--a game of tennis. Surprisingly, silent counting also produced activation of the cerebellum. It is not clear whether the regions were the same. This was confirmed by Ryding et al., 1993.
Again, with PET as for animal recording studies, one cannot be certain the brain activity was not causally related to some unobserved activity such as movement of body parts, which we have found to be particularly the case with the cerebellum (Thach et al., 1993). Therefore ablation studies form an important control and confirmation that the observed phenomenon is indeed correlated with and causally connected to the putative behavior.
Cerebellar deficits: impaired language generation, planning in games; impaired learning and error detection in both. Fiez et al. (1992) supplied this confirmation in their study of a patient who had had an infarction of the right cerebellar hemisphere (PICA distribution).
His language was not noticed to be abnormal during his hospital admission, and after discharge he returned to his law practice. Thereafter he noted only an "increase of 'slips of the tongue' (which, based upon his self explanation, appeared to be semantic paraphasia); a decline in his 'instant recall abilities', e.g. the ability to instantly associate the clients' names with details of the cases; and a shortened attention span, which he attributed to the cessation of smoking. When questioned, RC1's [the subject] wife also mentioned his 'slips of the tongue', but did not report any other problems or changes in his personality."
Specific tests included memory quotient, digit span, tapping span, delayed match to sample, Boston Diagnostic Aphasia Exam, Boston Confrontation Naming, Token Task, Wisconsin Card Sorting Test, Picture Arrangement, Word Fluency (CFL test). His scores were at or above normal on all of these.
Yet on the verb generation task, his performance was abnormal in two ways. First, he did not improve his reaction time as a function of practice. Whereas normals reduced theirs by approximately 30%, he showed little or no reduction. Second, he made errors by choosing inappropriate words. Examples included: for the test word "money", the incorrect response "market" (a control gave "spend", the most common response across controls); for the test word "razor", the incorrect response "sharp" (the control gave "shave", the most common response across controls). Whereas controls made between 10 to 20% such errors, the patient made between 40 to 75% on subsequent repetitions. He did not correct and seemed unaware of these errors.
On the Tower of Toronto Task, he required more moves than normals to complete the task, and failed to improve to the same extent with practice. On the first block of trials, normals averaged about 30 moves, while the patient required over 60 (the minimum necessary was 15). After 5 blocks of trials, normals had reduced to 20 moves, and the subject to 45. Normals continued to improve over blocks 6-10, while the patient plateaued.
The authors summarized: "performance on standard tests of memory, intelligence, 'frontal function', and language skills was excellent, [but] he had profound deficits in two areas: (1) practice related learning; (2) detection of errors. Considered in relation to cerebellar contributions to motor tasks, the results suggest some functions performed by the cerebellum may be generalized beyond a purely motor domain".
How "profound" the deficit was is a matter of scale and of focus. The failure to improve in reaction time and in the Tower game would pass as normal unless specifically looked for and quantitated against controls. The word generation errors appeared to stand out to interested observers, including his wife. Even so, many might consider the deficit rather subtle. Certainly, he could perform the tasks, and did show improvement with practice, albeit at a measured and grossly reduced magnitude.
In sum, one PET one ablation study of "non-motor learning" gave results surprisingly similar to those of "motor learning". In PET, similar parts "lit up" during the learning, which "dropped out" during learned performance (prefrontal cortex and lateral cerebellum). Some parts involved in learned performance were not active during learning (sylvian-insular cortex bilaterally and the left medial peristriate cortex). In ablation, there was impairment (but not elimination) of task learning.
Sequences. Inhoff 1989 tested the ability of cerebellar patients to generate sequences of one, two and three movements. In normals, the reaction time increased sharply with the number of the movements in the sequence. This indicates that some preprogramming has occurred somewhere in the nervous system, whose duration is proportionate to the length of the sequence. Cerebellar patients failed to show this result, and in proportion to the degree of their disease. Those with mild disease showed less delay in reaction time with each added segment in the sequence. Those with severe disease had almost the same reaction time independent of the length of the sequence. The authors commented that it was as though they were "decomposing" the movements in time, as Babinski and Holmes had demonstrated for compound movements. That is to say, they generated each of the series as if dealing with it singly, rather than in a unit.
Fiez et al., 1992 in the single patient and subsequently Grafman et al., 1992 in a series of patients with cerebellar disease showed difficulties in performing the Tower games of Toronto and Hanoi, respectively. Both achieved lower scores (took more moves) and made more errors (more wrong moves) than did normals. Fiez et al. give quantitative information showing that the impaired subject made about twice as many moves and errors as compared to normals. Grafman et al. only say that the impairment was statistically significant. It is important to point out that the subjects could still plan ahead after cerebellar disease well enough to play and finish the game. Whatever the cerebellum contributed, it was apparently useful, but not necessary--in these tasks.
Timing. A very robust finding has been that of Ivry et al., 1988 on mental timing. In subjects asked to compare control and test tone bursts of different durations (of the order of half a second), patients with lateral cerebellar infarcts scored poorly, as if random. The deficit was felt not to be due to making timed occult movements: the laterals did not show deficits in motor timing, and those with intermediate zone disease showed deficits in motor timing but not in perceptual estimates. The result was interpreted to imply that the lateral cerebellum was the clock of the nervous system, independent of motor activities.
Autism and attention. There is a considerable literature that connects the cerebellum with schizophrenia, autism and attention. Berman, Berman, and Prescott, 1974, Weinberger et al., 1980, and Berntson and Torello, 1982 have shown that brains of schizophrenic patients often show cerebellar atrophy in the posterior vermis. Floeter and Greenough, 1979, showed that "deprived rearing" of infant monkeys produced both an autistic condition and a lack of development in Purkinje cell dendrites. Yet these studies have not dissociated the mental from the known motor abnormalities that human schizophrenics and deprived-reared monkeys both show. Thus it remains to be seen as to what the correlation means.
Other. Appollonio et al., 1993 have recently performed a number of tests on their series of patients with cerebellar atrophy. Tests included those of general intellectual ability, different aspects of memory (effortful, automatic, and implicit), speed of information processing, verbal fluency (both category and letter fluency tests). Cerebellar patients were significantly impaired only of "tasks requiring the use of executive [italics ours] functions, such and the initiation/perseveration subtest of the Mattis Dementia Rating Scale or the fluency tests, and on memory measures requiring greater processing effort." The authors concluded that "the impairment is secondary to a deficit in executive functions". We shall comment on this shortly.
Bracke-Tolkmitt et al. (1989) previously had conducted a broad neuropsychological study of cognitive functions in their patients with cerebellar disease. IQs were found to be slightly lower in the cerebellar group, but most tests were in the normal range. They were significantly abnormal on 3 tests. These included the immediate re-drawing of the Benton figures, the learning of verbal paired associates, and in learning arbitrary associations between colors and abstract words. Again, the impairment was that of reduced--not abolished--performance. At their worst, in the word- color learning, and the Benton figure recall, subject scores were about half those of normals.
These cognitive contributions of the cerebellum could seem to be very like the popular conception of its motor contributions-- "fine control" and "coordination". In the mental as in the motor performance, the ability to act is never entirely lost, but only degraded. This in turn could conceivably result from a general "tuning up" function across many brain parts, as originally proposed by Holmes (1939) for movement. Somehow, this is not intuitively very satisfying.
WHAT DOES THE CEREBELLUM SPECIFICALLY CONTRIBUTE TO COGNITION?
What does the cerebellum contribute to mental movement? The Leiners have argued for a cerebellar influence on frontal lobe mentation. Much of their careful reasoning depends on there being anatomical projections from the cerebellum via thalamus to cerebral frontal association cortex. Whereas evidence for these connections for a time rather scant, it has steadily increased. Anatomical connections from cerebellum to far frontal association areas were proposed in monkey by Sasaki et al. (1976) using electroanatomical methods. In man, a phylogenetically new and unique posterolateral part of dentate was proposed by Hassler (1950) and Leiner et al. (1991) to project via thalamus to far frontal cortex. The Leiners inferred that the cerebellum could contribute to whatever was processed in these areas. Though they did not specify the type of mental operation performed in these areas, they did specify regions that are now identified as being higher order motor. And while they did not say exactly what the cerebellum might provide to these areas, because of the uniformity of the cerebellar architecture and the likelihood that the algorithms were similar or identical to those used for motor control, they used the terms "coordination" and "skill".
Let us return to the fact that the cerebellum is active during mental movement (Decety et al. and of Ryding et al.). That in itself could have been a spurious correlation; the cerebellum could have been active in relation to some unobserved synergic muscle activity that was associated with--but was not necessary for--the performance of the task. But the fact is now clear that ablation of the cerebellum does impair mental task performance. So somehow the cerebellum is involved, and the question is "how"?
Context triggering and combination. For motor learning, we have pursued the idea that the cerebellum may link a behavioral context to a motor response. The response may be a combination of many downstream neural elements firing together. Both the context- response linkage and the response composition would be achieved through trial and error learning. After practice, the occurrence of the context would trigger the occurrence of the response. This would explain how combinations of muscles may become active all at once, especially in skilled movements outside the capabilities of the motor pattern generators, so as to produce coordinated behavior appropriate to a specific context.
We extend the idea to include the premotor parts of the brain, to which the cerebellum is now know to project, and which are known to play a role in mental movement and cognition. These areas are active in the planning of movements that are then executed, and they plan movements that are not to be executed. They "think" movements. From recent evidence, the cerebellar output extends even to what has been characterized as the ultimate frontal planning area, the "prefrontal" cortex, area 46. The cerebellum may be involved in combining these cellular elements, such that, through practice, an experiential context can automatically evoke an action plan. The plan would be in the realm of thought. The plan either could--or need not--lead to execution. Again, the specific cerebellar contribution would be one of linkage of the context to a specific response, the combination of the response from simpler elements, and these accomplished through repeated practice. The prefrontal and premotor areas could still plan without the help of the cerebellum, but not so rapidly, automatically, or so precisely linked to context. At some level of task complexity, cerebellar damage would reveal itself in behavioral errors. This would have nothing to do with other cognitive activities--visual, auditory, attention, etc., and areas of the brain to which the cerebellum does not project.
Mental rehearsal of motor performance. It has been well documented that mental rehearsal improves motor performance (See Jeannerod, 1993). This is common knowledge amongst musicians, athletes, chess players, actors and lecturers. Repetitively playing through the performance in one's mind can remarkably improve the next actual performance. Is there any way in which the cerebellum might play a role in motor learning in which only mental movement is practiced?
A hitherto curious and unexplained oddity in the structure of the primate motor system is the evolutionary change in the connectivity of the red nucleus. As mentioned previously, the magnocellular (phylogenetically older) red nucleus gives rise to a prominent rubrospinal tract in carnivores, and receives from motor cortex and the cerebellar interposed nucleus. It is supposed to play a role similar to that of the corticospinal pathway in providing voluntary control of small distal muscle groups. By contrast, the parvocellular ((phylogenetically newer) red nucleus receives from premotor cortex (area 6) and the cerebellar dentate nucleus. Its output is thought to go not to the spinal cord but exclusively to the principal portion of the inferior olive, which in turn projects back to the lateral hemisphere of the cerebellum. This has seemed odd that a system capable of firing at high sustained frequencies should funnel into and dead-end in the conspicuously low frequency inferior olive. The suggestion had been made that this might fit into the general cerebellar role of motor learning (Kennedy, 1990), but a tighter rationale had not been provided. Mental motor rehearsal could well be the specific framework for its role in learning motor performances that are only practiced internally and not overtly expressed. The contexts would be brain states corresponding to prior elements of the performance, the combined responses would be of premotor neuron assemblies in area 6 (and 8,9,44,45,46?), the linkage would be established through repetitive practice. It is not clear what mental performance errors might consist of, or how they might be detected.
The cerebellar contribution to mental (and motor) performance would thus become critical when: 1) the task involves imagined movement that is rapidly and automatically triggered by context, 2) contains a number of linked neural (or body part) components, and 3) when these properties are being adapted or newly acquired. Paradoxically, the context-response coupling and the response combinations would be "unconscious" aspects of thought. They would have been at a level of awareness only during the learning phases. Ironically, this would be something that we have learned and no longer "know" anything about because it has been given over to the cerebellum for the implementation of automatic motor control, actual or imagined.
"Generated" language. It has been known that cerebellum lesions produce dysarthria--Holmes localized them to the midline vermis, Lechtenberg and Gilman (1978) to the midline and just to the left of it. The deficit is one of articulation--speech is variable in pronunciation (as if all the component muscles of speech can not be got together at once in the proper combination), and slow (as if all the components can not be got together linked in a sequence, and are instead emitted one at a time). It is not known to be context or task specific.
There is the fairly common condition of "cerebellar mutism" seen uniquely in children after cerebellar surgery (e.g. Rekate et al., 1985). In earlier descriptions, it was unclear whether this condition was one of abulia, autistic withdrawal, aphasia, aphonia, or anarthria. As more cases have been reported, the features have become clearer. Except for one case reported in a young adult, the syndrome has been confined to children. Characteristically its onset is delayed (in over half the reported cases) as much as 4 days after the injury. Often it does not occur with the initial surgery, but after some complication such as infection, hemorrhage, or reoperation. There may (or may not) be frank signs of brainstem involvement. Opinion is divided as to whether this condition results specifically from cerebellar damage, or due to some remote delayed effect--e.g., anarthria of brainstem origin. The problem in definitely implicating the cerebellum arises from the facts that: 1) most cerebellar surgery--in all respects equivalent--is unaccompanied by the mutism, 2) the delay after the cerebellar damage, and 3) the restriction to childhood. By contrast, the practicing neurologist and neurosurgeon, who see many acute and subacute cerebellar injuries in adults, never see them produce mutism, autism or schizophrenia.
Quite a different situation is that with the verb generate tasks in the studies of Petersen et al., Raichle et al., and Fiez et al. There, localization is distinctly to the right cerebellar hemisphere, the phenomenon occurs with some speech and language tasks and not others, and the effect wears off with familiarity.
Is variegated vocal sound production a basis of language? Lieberman (1969) has suggested that language began phylogenetically as an increased ability to make different vocal sounds. Proposed to originate with phylogenetic developments in the primate larynx, it paralleled the ability to fractionate movements and the dexterous use of the right hand, which is dependent on the development of the primate neo motor cortex. The argument runs that a capacity for making a great number of sounds (movements) gives one a vocabulary and dictionary of yet meaningless items to which one can then begin to assign symbolic identities. With the passage of time, the brain evolved to use the motor substrates for communication, symbolic representation, and some forms of thought.
"Automatic" speech. We begin to speak before we understand language. Beginning with babbling in infancy, we proceed through "rote learning" of nursery rhymes, nonsense poems and jingles without necessarily understanding them. We learn and can recite "The Jabberwocky" as movement and not at all as language.
A somewhat similar case is mnemonic sayings. We rote-memorize something that has so little linguistic or logical connection among the elements that it is learned as a movement. We can listen to what we say in order to get at what we otherwise can't remember. For example, "Thirty days hath September..." allows us to remember how many days there are in each month. But it is not something we know. It is buried in a rote-learning movement sequence. Physicians have and use a similar formula to recall in order the 12 cranial nerves; those few American students who can recall all the Presidents in the correct order will have committed the first syllable of each name to a catechism to verbal (motor) memory. One can suggest that recitation of the alphabet, multiplication tables, are similar in nature.
Speech and gesture. Another common observation is the coupling of gesture, especially of the face and hands, with speech. At what level in behavior and in the brain is this coupling? Do we gesticulate when we make mental speeches? Do we think of a word more quickly by thinking of it as a whole with its associated gesture?
Is such rote-learned, automatic language dependent on any one brain-part to the exclusion of others? People with lesions of the left frontal lobe operculum may suffer loss of "propositional" speech, yet retain relative amounts of other utterances--nursery rhymes, songs, the alphabet, multiplication tables. This is not to say that all of these sayings are "stored" in the cerebellum. No study yet has shown loss of nursery rhyme, childhood jingle, or verbal mnemonic from cerebellar ablation. Yet Bellugi's studies (Bellugi et al 1990) have suggested a role for the cerebellum in automatic speech. Children with Down's syndrome and a diseased cerebellum are limited in their development of speech, with relative sparing of other intellectual activities. By contrast, children with William's syndrome have an atrophic cerebrum with a relatively preserved cerebellum, and are conspicuous for their precocious "cocktail conversation". Socially responsive, fluent, correct in language content and appropriate to context, it is nevertheless peculiarly devoid of propositional content.
In this light, one is tempted to interpret the PET localization of "language" in the right cerebellar hemisphere (Petersen et al 1989) as a indicative of a fundamental process in automatic speech. The process of attaching a verb (a thought movement) or rhyme (a rehearsed movement) to a noun-context. And in Fiez' study, damage of the area results loss of automatic utterances of "the right word", and inability to make new associations.
Sequences. If it is true that the cerebellum participates in context-response coupling through learning, and of response formation, then the same mechanism could explain sequential behavior. Because of the richness of the input to the cerebellum, it is conceivable that any internal state at any almost any level of the central nervous system hierarchy could serve as a context. If the cerebellar targets are themselves high within the behavioral "planning" parts of the brain, plans for movement component "A" could trigger movement component "B" long before "A" was ever actually executed.
Such a mechanism could also account for the improvement in performance through mental rehearsal, and "non-motor" learning. The primary sequence becomes automatic, and we can proceed to build higher order structures on top of it. Through mental and actual practice, the moves of chess men may become so well learned that one may think at higher levels of strategy in planning the next moves, even through to thinking through an entire game. Being able thus to anticipate errors would help to prevent them.
Timing. Melvill-Jones and Watt, 1971, showed that certain motor acts have precise and stereotyped timing. In studies of hopping and stepping, they found that a group of students hopped at a frequency of 2 Hz, with little variation. They observed that this was the same frequency as the then popular dance music. They measured the latency of the segmental and also of the "functional" or long loop stretch reflex in the gastrocnemius at 120 ms. They surmised that the reflex, triggered by the setting down of the foot, was optimally timed to assist in the lift-off of the next foot rise. The reflex and body mechanics were therefore suggested essentially as being a "clock" for the body in this activity.
Decety and Michel, 1989, reported that imagined movements take proportionately the same time as when actually performed. This is somewhat surprising, since part of the mechanism proposed by Melvill-Jones and Watt must have depended on the mass-mechanics of the system, including inertia, elasticity and viscosity of tissues, and muscle contraction times, which would influence the resonance properties at least as much as the neural conduction times. This in turn suggests that the movements themselves are modeled within the nervous system, and that there is, as Jeannerod has recently argued, a complete neural motor "representation" (Jeannerod, 1994). Whatever the centers and connections are that complete this representation, its capability for precise timing could well serve as a clock for "non-motor" activities.
What Melvill-Jones and Watt showed for hopping is equally true for many other activities. We say "one thousand one, one thousand two..." to measure exposures in the darkroom, which have been observed to come very close to the times given by a clock. We sing to pace and synchronize bodily activities of dancing, marching, rowing, hauling, and many work activities. We can remember the meter of the song, because it is timed to the movement activity. Oddly, despite the physical differences between mental movements and actual movements, the timing is surprisingly similar, as Decety and Michel have shown. The question then becomes "Is there any activity that is mentally timed that does not have a movement referent?" One could imagine that in an attempt to time a process, one set up some internal standard that consisted of a mental motion of some part of the body. Since movements take time, and since the time taken by mental movements is equivalent to that taken by actual movements, the mental movement becomes a "clock" against which any external event may be compared. This could be a phrase, a tapping of the foot, a rocking of the body, or the blink of an eye--consciously, or not quite consciously employed. Implicit is that the movement have some period appropriate to the duration of what's to be measured. The "beat" need not involve the inferior olive (Keating and Thach, 1995)!
Does the cerebellum control everything mental?. Some have suggested that the cerebellum may have a very general influence on cognition, as if to participate anything or everything mental. See the reviews of Botez et al., 1985, 1988, 1989; Leiner et al., 1986, 1987, 1989, 1991; Ito, 1990; Schmahmann, 1991; Macklis & Macklis, 1992.
Murakami et al., 1989, Akshoomoff & Courchesne, 1992, Akshoomoff et al., 1992, have alone raised questions about a cerebellar contribution to attention. These observations have been based on few cases who had disease of other parts of the nervous system as well.
Meyer believed that schizophrenia was essentially a learning disability, a maladapted response to social contexts. Bleuler believed that in this disease, the personality, rather than having the elements properly combined, was dissociated into its constitutive elements. There are the unquestioned correlations listed above between cerebellar atrophy, schizophrenia, autism, "cerebellar mutism", and experimentally socially deprived rearing. Nevertheless, at the present time, none of these correlations satisfy the criteria set out at the beginning for demonstrating cause and effect. 1) The focal anatomy of these conditions is not well enough understood to know where the cerebellum projects to it or not. 2) Cerebellar ablation in the normal individual has not reproduced these conditions. 3) No cerebellar PET or single unit activity pattern has been particularly correlated with these aspects of behavior. This is not to say that these objections will not be overcome in the future.
How might one go about testing such a theory? One might begin with PET and ablation studies where mental task performance requires mental movements. It would be of interest to break these down into categories of context-triggered responses, of simple vs compound responses, and of the learning of each. Many of the tasks described herein that appear to engage the cerebellum fall into one or another of these categories. Examples include games where the player has to "play ahead to see the errors", and develop correct playing strategies. A good checkers or chess player can play an entire game in his head; a really good player, several games at once. The goal would be to identify and dissociate the context- response coupling from the response formation from the learning. The control for these could include mental tasks involving imagination in a spatial domain but not movement--such as solving abstract math problems, hearing music, seeing maps and faces, and learning them. But even these might be deceptive, as one could conceivably employ mental movement in ways more subtle than overt movement.
These tests could be done with human PET and fMRI scans and in human ablation studies. However, tests of mental timing would require greater sensitivity to dynamics. If the cerebellum and premotor cortex were involved in periodic rhythmic activity, simulating movement of body parts, such as might be used as the mechanism of a biological clock in the physiological motor domain, electrical analysis of the circuit (EEG, ERP, macro- or microelectrode) would be required to look for signs of such periodic activity.
SUMMARY
What PET and ablation studies tell us about the role of the cerebellum in motor learning in humans is that the cerebellum is but one part of a larger system that includes primary, supplementary, and premotor cortex, basal ganglia, red nuclei and especially the prefrontal cortex. PET studies show that the cerebellum and prefrontal cortex participate in "motor learning" and "motor performance" in different ways. Ablation studies show that the cerebellum may play a small part or a large part in "motor learning", depending on what is being learned. Ablation may slightly impair abolish motor learning, depending on the task. Both types of study raise but have not themselves resolved the question of what essentially the cerebellum contributes to movement that might be learned.
The greater contribution of human functional imaging and ablation studies has been to show that the cognitive functions may hinge entirely on the involvement of the cerebellum in imagined movement. Our thesis is that imagined movement is similar or identical to the early initiatory phases of actual movement. The cerebellum plays the same specific role in "coordinating" these imagined movements that it does in actual movements. We support the Brindley-Marr-Albus-Gilbert theories and suggest that these cerebellar pathways are used to build through trial and error learning behavioral context-response linkages, and to build up appropriate responses from simpler constitutive elements.
As cerebellar ablation does not abolish actual movements, but only their fine control and coordination, neither does it abolish imagined movements, but only their automaticity, speed of response, stereotypy and ability to improve them with practice. It is presumably due to the some inexactness in the use of the central representation of movement that, after cerebellar damage, errors are made in the movement-associated mental activities which are required for some "cognitive" task performances.
Acknowledgements: Marcus Raichle and Steven Petersen revealed subtleties of PET methodology; Marc Jeannerod, Michael Arbib, Giacomo Rizzolatti, and Hideo Sakata led the way through mental motor imaging; Julie Fiez, Peter Strick, Peter Gilbert, Jeffery Keating, Howard Goodkin, Amy Bastian and Tod Martin helped with these attempts at explanation. Nine referees corrected many errors. David Marr, James Albus, Henrietta Leiner, Alan Leiner, and Robert S. Dow did most of the original thinking.
REFERENCES:
Akshoomoff, N.A. & Courchesne, E. (1992) A new role for the cerebellum in cognitive operations. Behavioral Neuroscience 106:731-738.
Akshoomoff, N.A., Courchesne, E., Press, G.A., & Iragui, V. (1992) Contribution of the cerebellum to neuropsychological functioning: Evidence from a case of cerebellar degenerative disorder. Neuropsychologia 30:315-328.
Albus, J.S. (1971) A theory of cerebellar function, Math. Biosci. 10:25-61.
Antziferova, L.I., Arshavsky, Yu.I., Orlovsky, G.N. and Pavlova, G.A. 1980 Activity of neurons of cerebellar nuclei during fictitious scratch reflex in the cat. I. Fastigial nucleus. Brain Res. 200: 239-248.
Appollonio I.M., Grafman J., Schwartz, M.S., Massaquoi S. & Hallett, M. (1993) Memory in patients with cerebellar degeneration. Neurology 43:1536-1544.
Arshavskii YI, Berkinblit MB, Fuxson OI, Gelfand IM, Orlovsky GN (1972) Recordings of neurones of the dorsal spinocerebellar tract during evoked locomotion. Brain Res. 43: 272-275.
Arshavskii YI, Berkinblit MB, Fuxson OI, Gel'fand IM, Orlovsky GN 1972b Origin of modulation in neurones of the ventral spinocerebellar tract during locomotion. Brain Res. 43: 276- 279.
Arshavsky, Yu.I., Orlovsky, G.N., Pavlova, G.A. and Perret, C. 1980b Activity of neurons of cerebellar nuclei during fictitious scratch reflex in the cat. II. The interpositus and lateral nuclei. Brain Res. 200: 249-258.
Asanuma, C., Thach, W.T., & Jones, E.G. (1983) Anatomical evidence for segregated focal groupings of efferent cells and their terminal ramifications in the cerebellothalamic pathway of the monkey. Brain Res. Rev. 5:267-299.
Asanuma, C., Thach, W.T., & Jones, E.G. (1983) Distribution of cerebellar terminations and their relatiobn to other afferent terminations in the ventral lateral thalamic region of the monkey. Brain Res. Rev. 5:237-265.
Asanuma, C., Thach, W.T., & Jones, E.G. (1983) Brainstem and spinal projections of the deep cerebellar nuclei in the monkey, with observations on the brainstem projections of the dorsal column nuclei. Brain Res. Rev. 5:299-322
Babinski J. (1899) De l'asynergie cerebelleuse. Rev. Neurol. 7:806-816.
Babinski J. (1906) Asynergie et inertie cerebelleuses. Rev. Neurol. 14:685-686.
Baizer, J.S. & Glickstein, M. (1974) Role of cerebellum in prism adaptation. J. Physiol. 23:34-35.
Bastian AJ, Martin TA, Keating JG, Thach WT (1995) Abnormal Control of Multijoint Dynamics in Cerebellar Ataxia. J.Neurophysiolol. (in press).
*Bastian A.J. & Thach W.T. (1993) Impaired pinch and preserved reach after lesion of cerebellar thalamus. Soc. Neurosci. Abstr. (in press).
Bellugi, U., Bihrle, A., Jernigan, T., Trauner, D., and Doherty, S. (1990) Neuropsychological, neurological, and neuroanatomical profile of Williams Syndrome. Am.J.Med.Gen. Suppl. 6:115-25.
Berman, A.J., Berman, D., & Prescott, J.W. (1974) The effect of cerebellar lesions on emotional behavior in the Rhesus monkey. In: The Cerebellum, Epilepsy, and Behavior. Ed by Cooper, I.S., Riklan, M., and Snider, R.S. Plenum Press, New York.
Bernston, G.G. & Torello, M.W. (1982) The paleocerebellum and the integration of behavioral function. Physiol. Psychol. 10:2-12.
Bles W., Vianney de Jong J.M.B., & de Wit G. (1984) Somatosensory compensation for loss of labyrinthine function. Acta. Otolaryngol. (Stockholm) 97:312-321.
Bloedel, J.R. (1992) Functional heterogeneity with structural homogeneity: How does the cerebellum operate? Behav. Brain Sci. 3:1-39.
Bossom J. (1965) The effect of brain lesions on prism adaptation in monkeys. Psychon. Sci. 45-46.
Bossom J. & Hamilton C.R. (1963) Interocular transfer of prism- altered coordinations in split-brain monkeys. J. Comp. Physiol. Psychol. 56:769-774.
Botez, M.I., Gravel, J., Attig, E., Vezina, J.-L. (1985) Reversible chronic cerebellar ataxia after phenytoin intoxication: Possible role of cerebellum in cognitive thought. Neurology 35:1152-1157.
Botez, M.I., Leveille, J., & Botez, T. (1988) Role of the cerebellum in cognitive thought: SPECT and Neurological Findings. In: The Australian Society for the Study of Brain Impairment. eds. M. Matheson & H. Newman, pp. 179-195.
Botez, M.I., Botez, T., Elie, R., & Attig, E. (1989) Role of the cerebellum in complex human behavior. Ital. J. Neurol. Sci. 10:291-300.
Braitenberg, V. and Atwood, R.P. 1958 Morphological observations on the cerebellar cortex. J.Comp.Neurol. 109: 1-33.
Brand, S., Dahl, A.-L., and Mugnaini, E. 1976 The length of parallel fibers in the cat cerebellar cortex. An experimental light and electron microscope study. Exp. Brain Res. 26: 39-58.
Bracke-Tolkmitt, R., Linden, A., Canavan, G.M., Rockstroh, B., Scholz, E., Wessel, K., & Diener, H.C. (1989) The cerebellum contributes to mental skills. Behav. Neurosci. 103:442-446.
Brindley, G.S. (1969) The use made by the cerebellum of the information that it receives from the sense organs. Int. Brain. Res. Org. Bull. 3:80.
Decety, J. & Michel, F. (1989) Comparative analysis of actual and mental movement times in two graphic tasks. Brain and Cognition 11:87-97.
Decety, J., Sjoholm, H., Ryding, E., Stenberg, G., & Ingvar, D.H. (1990) The cerebellum participates in mental activity: Tomographic measurements of regional cerebral blood flow. Brain Res. 535:313-317.
Di Pellegrino G, Fadiga L, Fogassi L, Gallese V, Rizzolatti G (1992) Understanding motor events: A neurophysiological study. Experimental Brain Res. 91: 176-80.
Drevets W.C., Videen T.O., MacLeod A.-M.K., Haller J.W., Raichle M.E. (1992) PET images of blood flow changes during anxiety: correction (letter). Science 256:1696.
Eccles, J.C., Ito, M., & Szentagothai, J. (1967) The cerebellum as a neuronal machine, New York, Springer-Verlag, Inc.
Ekerot, C-F., Kano, M. (1