+
+

The cerebellum is among the most clearly organized circuits in the vertebrate brain and also among the most thoroughly studied. By the early 1970s a wealth of information about the anatomy and physiology of the cerebellum had been amassed, inspiring elegant theoretical models of its function. The cerebellum was proposed to automatize motor patterns by storing associations between contextual information and motor commands. In the early 1980s key predictions of the models were verified, including the establishment of a candidate cellular plasticity mechanism for the theorized associative learning capacity of the cerebellum. Many believed that the cerebellum would be the first major circuit in the vertebrate brain to be “cracked.” Yet decades later, links between the orderly circuity of the cerebellum and its function remain elusive. A major difficulty has been relating activity patterns of cerebellar neurons to particular aspects of behavior. In other words, what is the cerebellar code for movement? A recent study by Herzfeld and colleagues (2015) suggests a surprisingly simple answer to this question in the context of rapid eye movements known as saccades.

+

Although it has long been known that damage to the cerebellum results in deficits in motor coordination, the precise contribution of the cerebellum to motor control has come into focus only recently. Current thinking is that the cerebellum plays a key role in generating what are known as “forward models.” Forward models use the current state of the motor system and a copy of outgoing motor commands to generate a prediction of the change in state that will result from current motor commands. In short, a forward model predicts the sensory consequences of motor commands and thus the future state of movement circuitry. Such predictions are thought to be especially critical for rapid movements because they provide a means to correct movements without the need to wait for sensory feedback.

+

Some of the clearest evidence that the cerebellum functions as a forward model comes from studies of saccadic eye movements. Saccades are among the fastest movements that humans and other primates make; they are too fast for sensory feedback to play a role in correcting them. Behavioral studies show that under some circumstances humans reliably correct their saccades while they are happening to improve the accuracy of eye movement. Clinical studies show that patients with cerebellar damage fail to exhibit these fast corrections, consistent with the idea that the cerebellum functions as a forward model. These human behavioral and clinical studies provide a compelling motivation to understand how saccades are controlled by neurons in the cerebellum.

+

Herzfeld and colleagues used electrophysiological recordings to characterize the activity of cerebellar Purkinje cells in monkeys trained to make saccades to visual targets. Purkinje cells do not control movements directly. Rather, they inhibit neurons in the deep cerebellar nucleus that project to brain regions that control movements, including saccades. Lesion and electrical stimulation studies have identified specific regions of the cerebellum, in the oculomotor vermis, that are required for normal saccades. Puzzlingly, however, the patterns of action potentials produced by individual Purkinje cells in these regions do not relate in a straightforward way to the real-time motion of the eye during a saccade. For example it has been found that although individual Purkinje cells exhibit bursts or pauses in firing during saccades, the durations of bursts and pauses are much longer than the saccade. Such findings are difficult to reconcile with the ideas that the cerebellum participates in predicting (as in the forward model hypothesis) and controlling saccades.

+

The first key insight of this study is that although the activity of individual Purkinje cells is not predictive of eye motion, the combined activity of multiple Purkinje cells does accurately predict eye motion. The reason is that some Purkinje cells increase firing in relation to saccades while others decrease. When added together these two opposite response patterns partially cancel each other out, shortening the duration of the combined response to match that of the saccade. Although Herzfeld and colleagues combined the spiking activity of many Purkinje cells using a computer, such a coding scheme is not at all artificial. Neurons in the deep cerebellar nucleus integrate input from around 50 Purkinje cells (the same number used by the researchers to generate accurate predictions of eye motion). For the first part of the study Herzfeld added up the responses of Purkinje cells selected at random. However, it is not known whether the population of Purkinje cells that synapse onto a deep cerebellar nucleus cell is random or selected according to some rule.

+

The authors chose to test a rule related to a distinctive feature of Purkinje cells—the climbing fiber response. All Purkinje cells receive two classes of excitatory input: mossy fiber–granule cell input and climbing fiber input. The mossy fiber–granule cell system conveys diverse sensory and motor information originating from many brain regions. Each Purkinje cell receives weak input from hundreds of thousands of granule cells, which distribute inputs broadly to neighboring Purkinje cells via their long-ranging axons, known as parallel fibers. In contrast, climbing fiber inputs originate from a single source, the inferior olivary nucleus in the brainstem, with each Purkinje cell receiving powerful input from just one inferior olive neuron. This so-called climbing fiber input wraps around the Purkinje cell dendrite, evoking a distinctive response in the Purkinje cell, known as a complex spike, every time the olivary neuron fires. A considerable body of evidence suggests that the climbing fiber acts as an error or teaching signal by modifying the strength of synapses between granule cells and Purkinje cells. Herzfeld and colleagues categorized Purkinje complex cell spikes according to their response to saccadic errors. By displacing the saccadic target in different directions after the start of the saccade, they found that different Purkinje cells exhibited complex spikes that fired most reliably in response to errors in a specific direction.

+

Summing Purkinje cell responses grouped according to these complex spike responses led to a second remarkable finding. Combining responses of cells with similar sensitivity to error direction produced accurate, real-time predictions of saccade speed, amplitude, and direction. Purkinje cell populations responded most strongly for saccades in the direction opposite to that of error tuning of the complex spikes. The origins of this direction tuning related to a subtle shift in the timing of the pausing Purkinje cells for different saccade directions. Strikingly, the directional tuning was completely absent if Purkinje cell populations were chosen randomly. Indeed there is a compelling rationale for organizing Purkinje cells according to their climbing fiber responses. Extensive anatomical and physiological mapping studies suggest that the cerebellum is subdivided into functionally related groups of Purkinje cells that receive input from the same or similar climbing fibers and that converge onto the same or nearby neurons in the deep cerebellar nucleus. The results of this study suggest that such precise anatomical arrangements may be the key to deciphering how Purkinje cell activity relates to movement.

+

A goal for future studies will be to test experimentally the idea that deep cerebellar nucleus neurons receive input from populations of Purkinje cells with similar directional error tuning. Although such experiments are challenging, optical methods for monitoring and manipulating activity in neural populations are advancing rapidly. By combining optical recordings and manipulations of Purkinje cells with intracellular recording from deep cerebellar nucleus neurons, the ideas put forth by Herzfeld and colleagues could be directly tested. Fully understanding how cerebellar activity relates to saccades will also require understanding the integration of the two additional major inputs to deep cerebellar nucleus neurons. These inputs are (1) collaterals of the mossy fiber and (2) climbing fiber inputs that project to the cerebellar cortex. It will also be of interest to see whether the logic applied in this study—i.e., the pooling of Purkinje cells with similar complex spike responses—similarly sheds light on how the cerebellum encodes other types of movements—e.g. arm movements. Finally, a fascinating task for developmental neuroscientists will be to understand the mechanisms through which the selective wiring of functionally related Purkinje cells onto deep cerebellar nucleus cells is actually achieved at the molecular level.

+

Although the function of the cerebellum, including its putative role as a forward model, is still far from known, Herzfeld and colleagues have taken a major step toward cracking the cerebellar code for saccadic eye movements. The key to their approach was to move beyond the consideration of the responses of single neurons and to think instead in terms of neural populations, operations for combining them, and rules for selecting them. Their results are all the more encouraging because the operations and rules they arrived at are both quite simple and firmly rooted in the known anatomical structure of the cerebellum. The crystalline circuitry and precise organization of the cerebellum may yet mark the road toward understanding its function.

Reference +
+
Herzfeld  DJ, Kojima  Y, Soetedjo  R, Shadmehr  R. 2015. Encoding of action by Purkinje cells of the cerebellum. Nature 526:439–442.