As already discussed, the cerebellum is unlike the neocortex, for it is dedicated to motor control (Eccles 1967). To establish this control, circuits at the level of the Purkinje neurons are modified as animals learn new motor routines, and this learning takes place no matter how much or how little declarative information supports the learning. For example, something like learning to ride a bicycle is often presumed to entail little declarative assistance, but for you parents out there, you will remember the hours of declarative and procedural agony that went into teaching your child how the ride a bicycle, often being reminded by your child daily and for weeks: “Mommy...Daddy!!! Are we going to train today?” Some have suggested that since lesions of the hippocampus do not interrupt the learning of delayed classical conditioning (whereby the conditioned stimulus and unconditioned stimulus are overlapped in time) that the neocortex is concerned solely with declarative learning and not procedural learning (Swain et al. 2011). This view is being challenged empirically (Schiller, Phelps et al. 2010; Søgård, Monoach et al. 2024; Vorster and Born 2015), and rightly so, since the senses (i.e., consciousness, current and remembered) and movements of the body work in tandem when one learns (Andersen and Mountcastle 1983; Darlington and Lisberger 2020; James 1890; Kimura 1993; Sacks 1976, 2012; Schiller and Tehovnik 2015; Vanderwolf 2007; but see: Bushnell, Goldberg et al. 1981); separating the two makes about as much sense as separating one’s genes from one’s cytoplasms (Noble 2023).

An efference copy is a representation in the central nervous system that allows animals to predict a response per context once learning has been finalized. This allows for the shortening of response latencies, such that any latency limitation of a sensory system will not interfere with the rapid execution of a response, which is a hallmark of the vestibuloocular reflex (Miles et al. 1980ab; Miles and Lisberger 1981). Changes to the efference-copy code are managed by cerebellar Purkinje neurons, and the first study to show this (convincingly) was done in the electric fish. Bell et al. (1997) showed that Purkinje neurons could be made to decrease their gain (i.e., their firing rate) or increase their gain (i.e., their firing rate) depending on how the simple spikes of the Purkinje neurons overlapped or did not overlap with the complex spikes. When the simple-spike EPSPs overlapped with the intracellular evoked complex spikes by 30 ms in a Purkinje cell, the simple spike EPSP decreased in amplitude (Fig. 2A,b & Fig 2B, top, Bell et al. 1997), which translates into fewer simple spikes. On the other hand, when the simple-spike EPSPs did not overlap with the intracellular evoked complex spikes by 210 ms in a Purkinje cell, the simple spike EPSP increased in amplitude (Fig. 2A,c & Fig 2B, bottom, Bell et al. 1997), which translates into more simple spikes. This basic relationship has now been demonstrated in mammals including rodents and primates (e.g., Boyden et al. 2004; Catz, Thier et al. 2005; Gilbert and Thach 1977; Kitazawa et al. 1998; Medina and Lisberger 2008; Miles and Lisberger 1981; Soeterdjo and Fuchs 2006; Soetedjo et al. 2008; Swain et al. 2011; Thach et al. 1992; Wang et al. 2023), and it is now accepted that the cerebellum is the center of efference-copy control (Cullen 2011; De Zeeuw 2021; Ebner and Pasalar 2008; Fukutomi and Carlson 2020; Huan et al. 2013; Ito 2008; Loyola et al. 2019; Miall et al. 1993; Shadmehr 2020; Tehovnik et al. 2021; Thach et al. 1992; Wolpert et al. 1998), and perhaps for all vertebrates.

Based on the model of Marr (1969), the relationship between the cerebellum and the neocortex is as follows for efference-copy control: One of the first studies to deduce how structures external to the cerebellum are involved in setting the adaptation signal in the cerebellum during learning has been performed by Soetedjo, Fuchs and colleagues (2019). Monkeys were trained to generate a saccadic eye movement to a punctate visual target, such that during the execution of a saccade the target was stepped toward or away from the starting point of the saccade. When subjected to such conditions repeatedly for one direction of target shift, monkeys learn to anticipate the direction of the shift by altering the amplitude of the saccadic eye movement. Prior to learning, a monkey produced a corrective saccade to acquire the target. It is the change in complex spikes by way of the inferior olive that alters the Purkinje cell firing rate in the cerebellum, in that decreases in saccade amplitude are accompanied by a reduction of simple spikes, and increases in saccade amplitude are accompanied by an addition of simple spikes. The adaptation signal via the complex spikes can be manipulated by electrically stimulating or by chemically inactivating populations of cells in the superior colliculus that encode saccade metrics (Soetedjo et al. 2019). Once adaptation is finalized, however, such manipulations are ineffective (much like what happens for the learning of the eyeblink conditioning response, Swain et al. 2011). The superior colliculus contains a topographic map for the generation of visually-guided saccadic eye movements (Schiller and Tehovnik 2015); therefore, it sends a ‘where’ signal to the inferior olive so that a visual object is acquired efficiently and any change in the sensorimotor circumstances produces a change at the cerebellar Purkinje cell synapses, so that the eyes arrive on target via the climbing fibre adaptation channel. This is the ‘where’ component of the response (see Fig. 12, circuits colored in purple).

Unlike the superior colliculus, which codes for where a visual object is located, the neocortex contains cells that code for the ‘what’ aspects of an object such as its luminance, color, shape, depth, motion, and so on (Schiller and Tehovnik 2015). The temporal lobes are known to house neurons that fire to a specific collection of features of an object, and damage to this part of the brain impairs object recognition in monkeys and humans (as does damage to the middle temporal cortex that impairs identifying objects defined by motion, Schiller and Tehovnik 2015). Based on the observations of Soetedjo, Fuchs and colleagues (2019), if one were required to generate an eye movement to a new object such as the letter B after having learned to make such a response to letter A, two things need to happen: (1) the firing of cells in the temporal cortex should shift from those coding for object A to those coding for object B (which are induced by changes in reward contingency); (2) during the transition period from object A to B the inferior olive alters the synaptic configuration at the cerebellar Purkinje cells mediating the response such that object B is made to evoke the response that was once evoked by object A, as signaled to the Purkinje neurons by altering the composition of mossy fibre activation (Marr 1969) (see Fig. 12). Once the new configuration is complete (with optimal performance in terms of percent-correctness at the shortest latency) the inferior olive is disengaged from the cerebellum. Object B has now replaced object A for generating the response, fictively from the cerebellum (for the cerebellum is an efference copy decoder) that receives the same information as the saccade generator for response execution. The foregoing process applies universally to any type of body movement, since the cerebellum is privy to all oculo- and skeletomotor responses subserved by the muscles.

Alterations to Purkinje firing via the climbing fibres (which generate the complex spikes) is such that one climbing fibre is dedicated to every Purkinje neuron for control over plasticity, and each climbing fibre innervates up to ten Purkinje neurons (Eccles 1967; Gibson et al. 2004; Ito 2008; Rawson and Tilokskulchai 1981; Thach 1967). The source of the climbing fibres, the inferior olive[1], receives input from the neocortex, brain stem, and spinal cord (Berkely and Worden 1978; De Zeeuw et al.1998; Eccles et al. 1967; Ekerot, Schouenborg et al. 1987; Hasanbegović 2024;Loyola et al. 2019; Saint-Cyr 1983; Sladek and Bowman 1975; Swenson and Castro 1983; Swenson et al. 1989; Toonen et al. 1998; Watson, Apps et al. 2013). In addition to reward and visual information (as mentioned above), the climbing fibres carry auditory, tactile, and vestibular information, as well (Ekerot, Schouenborg et al. 1987; Gallistel et al. 2022; Lueke and Robinson 1994; Miles and Lisberger 1981; Swain, Thompson et al. 2011; Watson, Apps et al. 2013). During the elicitation of volitional acts such as reaching, orienting, or walking, the climbing fibres are disabled (Apps 1999; Armstrong et al. 1988; Bauswein et al. 1983; Carli et al. 1967; Gellman et al. 1985; Smith and Chapin 1996), which means the efference-copy code cannot be altered at this time[2]. It is believed that climbing fibre disablement is under the control of the neocortex (Apps 1999), which also controls volition (as suggested previously). The climbing fibres tend to be most active during periods of immobility or during passive body movement (Armstrong et al. 1988; Bauswein et al. 1983; Gellman et al. 1985). Note that while learning a new task, there will be more occasion for interruption (or immobility), during which time the Purkinje neurons can be further modified to improve motor performance. Lastly, the climbing fibres are suppressed during REM sleep, and active during slow-wave sleep[3], as evidenced by electrical stimulation of the forelimbs (Carli et al. 1967). Once a task is automated, complex spike activity returns to baseline (Sears and Steinmetz 1991; Swain et al. 2011).

In electric fish, the efference-copy representation allows for the following discriminations (Fukutomi and Carlson 2020): (1) electrical communication as generated by an electric fish and such communication as generated by another electric fish, (2) electrical detection of the presence of others (conspecifics as well as inanimate objects) by noting changes in the self-generated electric field, and (3) electrical detection of others by detecting the electric field of others. To establish these relationships, there needs to be continuous communication between the cerebellum and the telencephalon, as is true of communications between the cerebellum and neocortex in mammals (Hasanbegavić 2024). In humans, this is especially the case when engaged in dialogue, delivering a speech in a particular language, or carrying out rapid translations between two languages given that each language is stored by a separate network of neurons in the cerebellum and neocortex (Corkin 2002; Kimura 1993; Levinson and Torreira 2015; Mariën et al. 2017; Ojemann 1991, 1983). Schizophrenics are known to have great difficulty discriminating between self-initiated body movements and those produced externally (Blakemore et al. 2000), which indicates a problem with the efference-copy code.

In oculomotor neurophysiology, many decades have been spent trying to identify the source of the efference-copy code in primates (Goldberg 2015; Rao et al. 2016). To study this, pre-saccadic activity has been examined throughout the neocortex (mainly in the lateral intraparietal area and the frontal eye fields, Duhamel et al. 1992; Goldberg and Bruce 1990), with the suggestion that this activity predicts the future eye position, and that this prediction is independent of knowing where the eyes are in orbit (Goldberg and Bruce 1990). Two issues have arisen regarding this story: (1) Does the updated receptive field that predict the future transition in parallel with the direction of a saccadic eye movement (to indicate a sensory remapping), or does it converge onto the endpoint of an ongoing saccade (to indicate a motor remapping)? (2) Does the source of this activity come from the superior colliculus? Regarding what is being remapped (perception or movement), when a complete set of target positions and saccade directions is studied throughout the visual field, it is found that the remapping signal converges onto the endpoint of a saccade (Tolias et al. 2001; Zirnsak et al. 2014). This fact coincides with the idea that the remapping signal originates from the cerebellum as a motor vector, rather than from the neocortex as a sensory vector.

As to the source of the remapping signal, experiments were performed in which a pathway between the superior colliculus and frontal eye field via the thalamus (i.e., the mediodorsal thalamus) was disabled, while monkeys performed a double-step task (i.e., two visual targets were presented in succession and a monkey was required to generate a saccadic eye movement to each from memory) (Sommer and Wurtz 2004). It was found that inactivation of the thalamus with muscimol (mildly) disrupted the performance of monkeys on this task. This was used to suggest that the remapping signal originated from the superior colliculus. A serious confound in this experiment is that injecting inhibitory agents into one side of the thalamus produces position habits, which could account for the mild deficit. Here we predict that bilateral inactivation of the thalamus would fail to disrupt the double step task, and that a more meaningful inactivation would include cerebellar circuits.

Electrical stimulation has been used to assess what part of the brain stores the eye position signal for the generation of saccadic eye movements. A monkey can be trained to generate a saccadic eye movement to a remember target position, but before execution of the saccade, electrical stimulation of brain can be used to determine if the monkey’s brain registers the displaced eye position by generating a corrective saccade to the remember position. It has been found that displacement by activation of sites at or above the superior colliculus (including the frontal and medial eye fields) can be corrected (Schiller and Sandell 1983; Sparks and Mays 1983; Tehovnik and Sommer 1996)[4]. But activation of sites within the cerebellum, the brain stem saccade generator, or the oculomotor nuclei failed to be corrected (Noda et al. 1991; Schiller and Sandell 1983; Sparks et al. 1987)[5]. This suggests that the eye position signal is maintained by the cerebellum and shared by the brain stem generator and oculomotor nuclei, as animals learn new oculomotor tasks. Indeed, it has been shown that damage of the trigeminal fibres (which carry eye position information) failed to interrupt the corrective saccades produced after stimulation of the superior colliculus in trained monkeys (Guthrie et al. 1983). We predict that untrained monkeys would not generate corrective saccades, since the efference-copy signal at the level of the cerebellum would not have been updated.

Finally, it has been shown that if one activates the ocular proprioceptors in a monkey immediately before the monkey elicits a saccadic eye movement toward a visual target, the eyes are systematically displaced off target (see Figure 13 from Chen 2019). Such results have been confirmed in human subjects whose ocular proprioceptors were activated using vibratory stimuli (Roll and Roll 1987; Roll et al. 1991; Valey et al. 1994, 1995, 1997). Whether animals can be trained to redirect their gaze directly toward the target after proprioceptive perturbation is not known. We know that one of the dominant inputs to the cerebellum is proprioception, which arrives at the cerebellum at a latency as short as 3 ms (Fuchs and Kornhuber 1969), which is an order of magnitude shorter than the arrival of such signals at the neocortex (Xu, Goldberg et al. 2012). This affirms further that the cerebellum is centrally involved in maintaining the eye position signal for oculomotor control rather than the neocortex.

Summary:

1. The efference-copy representation is updated at the Purkinje neurons by decreasing and increasing the simple spike rate, respectively, via the presence and absence of complex spikes, with the presence reducing the gain and the absence increasing the gain; this property may be universal throughout the vertebrate line, given that it was discovered first in the electric fish.

2. Alterations of the efference-copy code are triggered by inputs to the inferior olive, which is the source of the complex spikes.

3. Changing the gain at the Purkinje neurons is most often made during immobility, for volitional movement disables this input. Also, once motor learning is finalized, the activity of the complex spikes returns to baseline.

4. In communicating animals such as the electric fish and humans, an efference-copy representation is necessary to discriminate between self-generated communications and those originating from others. Schizophrenics have problems with this discrimination.

5. Even for oculomotor control, changes to the efference-copy code occur in the cerebellum, even though the consequences of these changes are registered throughout the neocortex (Goldberg 2015).

6. The eye-position signal is maintained by the cerebellum (rather than the neocortex) for the generation of saccadic eye movements.

Footnotes:

[1] The inferior olive is innervated by gap junctions. The neurons in the olive are electro-tonically coupled, which allows for synchronized discharges (Llinas et al. 1974).

[2] The climbing fibres may be disabled during self-generated movement, so that the sensations experienced during the movement (due to re-afference) are not attributed to external sources (Apps 1999).

[3] The consolidation of memory takes place during slow-wave sleep in both the neocortex and cerebellum (Apps 1999; Wilson and McNaughton 1994). Furthermore, myelination and synaptogenesis are augmented during periods of sleep based on the expression of specific genes in both invertebrates and vertebrates (Cirelli et al. 2004; Gilestro et al. 2009).

[4] The approximate error between the endpoint of the corrective saccade and the target location was about 1 to 2 degrees, which compares to the resolution of the oculo-proprioceptive system (Gauthier et al. 1990ab, 1994), suggesting that the proprioceptive system provided the memory here.

[5] Electrical stimulation of some sites within the saccade generator could be corrected (Sparks et al. 1987); here it is believed that the collaterals from the neocortex and superior colliculi were being activated to enable the correction by not distorting the eye position signal through stimulation.

Figure 12. The neocortex of primates mediates the ‘what’ characteristics of a visual target and the superior colliculus mediates the ‘where’ or location characteristics of a visual target. Symbols ‘A’ and ‘B’ are stored in the temporal cortex declaratively after learning. To communicate the ‘what’ and ‘where’ properties of the symbols to the cerebellum, the mossy fibres carry the pattern of the symbols to the cerebellar cortex (Marr 1969), and the climbing fibres carry the location signal to the cerebellar cortex (Soetedjo, Fuchs and colleagues 2019). The two signals are merged during learning so that once associated, a saccadic eye movement is generated to a symbol at the appropriate time, i.e., ‘when’. To drive the saccadic behavior, i.e., the ‘why’, a reward signal is transmitted from the ventral tegmental area (VTA) to the cerebellar cortex via the climbing fibres. Other labels are as follows: visual cortex (V1), motor cortex (M1), and superior colliculus (SC). For other details see Tehovnik et al. 2021.

Figure 13. (A) A monkey was trained to generate a saccadic eye movement to a punctate visual target (T) from a fixation spot (F). Before evoking an eye movement, electrical stimulation was delivered to the ocular proprioceptors which displaced the eye by six degrees (to F’). After termination of the stimulation, the eye failed to acquire the visual target and instead went to position T’. (B) On a percentage of trials, control trials were interleaved in which no stimulation was delivered to the proprioceptors. On such trials, the monkey successfully acquired the target (T) from the fixation spot (F). Data from Chen (2019).