The cerebellum and the Mirror Neuron System: A matter of inhibition? From neurophysiological evidence to neuromodulatory implications. A narrative review

Mirror neurons show activity during both the execution (AE) and observation of actions (AO). The Mirror Neuron System (MNS) could be involved during motor imagery (MI) as well. Extensive research suggests that the cerebellum is interconnected with the MNS and may be critically involved in its activities. We gathered evidence on the cerebellum ’ s role in MNS functions, both theoretically and experimentally. Evidence shows that the cerebellum plays a major role during AO and MI and that its lesions impair MNS functions likely because, by modulating the activity of cortical inhibitory interneurons with mirror properties, the cerebellum may contribute to visuomotor matching, which is fundamental for shaping mirror properties. Indeed, the cerebellum may strengthen sensory-motor patterns that minimise the discrepancy between predicted and actual outcome, both during AE and AO. Furthermore, through its connections with the hippocampus, the cerebellum might be involved in internal simulations of motor programs during MI. Finally, as cerebellar neuromodulation might improve its impact on MNS activity, we explored its potential neurophysiological and neurorehabilitation implications.


A brief overview of the Mirror Neuron System
In 1992, a group of neuroscientists at the University of Parma identified a new class of visuomotor neurons in the monkey's ventral premotor cortex (specifically, area F5) that discharged both when the animal performed a goal-directed action (i.e.grasping pieces of food) and, of note, when the animal saw another individual performing the same action (di Pellegrino et al., 1992;Gallese et al., 1996;Rizzolatti et al., 1996).The same research group named this particular class of cells "mirror neurons" (MNs) to underscore the motor system's ability to "mirror" observed actions by engaging the same brain areas implicated in action planning and execution.Since this discovery, numerous studies in monkeys have identified neurons with mirror properties (as they "mirror" or reflect the action, i.e. represent both the execution and the observation of the action itself) in other brain areas, including the anterior cingulate cortex (ACC), prefrontal cortex (PFC), inferior parietal lobule (IPL) and primary motor area (M1) [for a review, see (Bonini et al., 2022)].
Importantly, the existence of neurons with mirror properties, within crucial nodes of the motor network, has led to speculations regarding their functional significance across the circuit responsible for action planning, programming, and execution, at various levels (Casile, 2013).In addition, three categories of MNs have been identified, namely: strictly congruent (discharging during both the observation and execution of the same action), broadly congruent (active during both the execution of one action and during the observation of another similar but not identical) and logically related (active during the execution and observation of different, but related, actions) (Gallese et al., 1996;Heyes and Catmur, 2022;Kilner and Lemon, 2013).The first two categories were those that mostly attracted research interest; however, the third category above all made it possible to hypothesise that MNs do not merely match observed and performed actions from a motor point of view (e.g.kinematic features) but, rather, that they also encode higher-order aspects of actions, in particular their purpose (Rizzolatti et al., 2014).Moreover, each area provides specific functional contributions.For example, while MNs in the ventral premotor cortex primarily encode the goal of the observed actions (even in cases of incomplete sensory information, as long as the available data are adequate to deduce the intention behind the observed behaviour), those in the parietal cortex are more sensitive to sensory and kinaesthetic features (Carr et al., 2003;Miall, 2003;Umiltà et al., 2001).
Subsequently, the presence of MNs has also been demonstrated in other species, including birds, rodents, bats and, crucially, humans [see (Bonini et al., 2022)].The first indirect evidence in humans emerged from a transcranial magnetic stimulation (TMS) study, which highlighted that the observation of actions (AO) increases the excitability of the motor system in an effector-specific fashion, showing a close alignment with the observed pattern (Fadiga et al., 1995).This allowed the authors to suggest the existence of a human system matching AO and execution, a seminal aspect to promote and inspire the extensive research efforts of recent decades in this field.In 2010, their presence was confirmed through a single-cell recording study, which documented neuronal activity during action execution and AO in the human medial frontal cortex and medial temporal cortex (Mukamel et al., 2010).Further research has shown that, in humans, in addition to the homologous areas in the monkey, mirror-like activities are also evident in the inferior frontal gyrus (IFG) and superior temporal sulcus (STS), which are involved in the regulation of language production and the integration of multi-modal information, respectively (Molenberghs et al., 2010).Moreover, a key role in action understanding has also been revealed for the ventral (mainly mediating distal motor actions, irrespective of intrinsic motor features and effectors) and dorsal premotor area, which, together with the superior parietal lobule, encode peripersonal space and guide reaching movements in this context (Cattaneo et al., 2010;Clavagnier et al., 2007).These areas have collectively been referred to as the Mirror Neuron System (MNS) or Action Observation Network (AON), i.e. a neural network that is activated both when individuals perform an action and when they observe someone else performing the same or similar action.It has been hypothesised that MNs are involved in a wide range of functions, including imitation and social learning, empathy and emotion recognition, communication, and theory of mind (i.e. the ability to attribute thoughts and intentions to others also to predict their behaviour) (Keysers and Fadiga, 2008;Rizzolatti, 2005).
Although different, their common aspects suggest that the MNS enables the subject to mentally simulate observed actions, find their correspondence ("direct resonance" or "direct matching") in their own motor repertoire, and use this knowledge to infer the implications of the behaviour of others, also because of its wide connections with areas involved in planning, value attribution, and emotions (Du et al., 2020;Ferrari et al., 2017).Consistently, many studies suggests that the MNs matching mechanism maps the observed action onto the observer's own motor repertoire, leading to a direct comprehension of the action's goal (Errante and Fogassi, 2020).In other words, MNs process the observed actions by associating them with our representations and are consistently sensitive to the degree of correspondence between the observed action and the observer's motor skills (Fadiga et al., 1995;Gallese et al., 2004).Indeed, observing a movement performed by an artificial hand elicited notably less activity in the MNs compared to human hand movements and, similarly, biomechanically impossible actions do not activate these neurons (Perani et al., 2001;Stevens et al., 2000;Tai et al., 2004).These insights enabled significant improvements in robotics and its applications to neurorehabilitation, a promising and rapidly developing field (Maggio et al., 2021;Zahra et al., 2022).Moreover, the observation of complex movements (e.g.dancing) may recruit regions of the MNS to the extent that the observed action is represented in the subject's personal motor repertoire, i.e. to a greater extent in those who are experienced than those who are not.Specifically, the neural response to AO is shaped by both previous visual exposure and experiential motor engagement, meaning that AO entails matching the observed behaviour to one's motor repertoire beyond mere perceptual matching (Calvo-Merino et al., 2006).This also suggests that the AON encodes particularly complex action patterns, not just simple component movements (Calvo-Merino et al., 2005).However, it has been suggested that observed actions are segmented into basic motor acts from one's motor repertoire, allowing for motor encoding and immediate forwarding to other structures within the MNS for replication (i.e. for the purpose of imitation, if needed) (Buccino et al., 2004).If, conversely, AO also requires the learning of a novel motor pattern, it entails the recombination of motor actions that have been previously "resonated" (Byrne, 2002;Rizzolatti and Craighero, 2004).Interestingly, neurons with mirror properties have also been shown in the hippocampal regions, and their functional significance is still only partly understood, although an involvement of AO in mnestic processes has been suggested (Calvo-Merino et al., 2005;Mukamel et al., 2010).
Importantly, MNS dysfunction has been highlighted in a number of neurological and psychiatric disorders, which have enabled the exploration of the functional significance of MNs and the evaluation of potential prognostic and therapeutic implications arising from their dysfunction (Farina et al., 2020;Hamilton, 2013;Mehta et al., 2014).Nevertheless, it is worth mentioning that even the most recent research tends to emphasize the role of neocortical regions, often overlooking the potential contribution of subcortical structures, like the cerebellum, to the MNS.Importantly, recent studies deepened cerebellar contribution in cortico-subcortical brain functions in physiology and pathology, such as in the context of motor control [see, for example, the review by (Caligiore et al., 2017b)].However, the evidence on the influence of the cerebellum on brain areas with a specific focus on its role in the context of the MNS, as well as the neurophysiological and neurorehabilitative implications of cerebellar modulation on MNS activities, is still poorly explored.Therefore, in this review, we aimed to gather evidence, theoretically and experimentally, on the cerebellum's involvement in MNS functions and to discuss the hypothesis that cerebellar modulation on brain areas also depends on its influence on the MNs network.Given the fundamental role of MNs in many essential human activities, we hope to provide a comprehensive overview that enhances the understanding of the neurophysiological mechanisms linking the cerebellum to the MNS in both healthy and pathological contexts and to suggest potential neurorehabilitation and neuromodulation implications for aiding recovery in patients with various neurological conditions.

The human cerebellum: feed-forward internal models, predictive coding and action selection
The cerebellum houses over half of the neurons of the central nervous system (CNS) (Roostaei et al., 2014).Although early evidence, primarily derived from observations of neurological patients with cerebellar damage, initially hinted at its role primarily in motor functions, particularly in coordinating and sequencing movements smoothly, recent research has unveiled its involvement in a broad spectrum of perceptual and cognitive processes (Manto and Mitoma, 2023).Indeed, anatomically, the cerebellum is segregated into two distinct regions: the anterior cerebellum, which interfaces with the sensory-motor regions of the cerebral cortex and primarily modulates motor functions, and the posterior cerebellum, which projects to an extensive network of frontal-parietal areas crucial for higher-level cognitive functions (Schmahmann, 2019).Moreover, there are two distinct body representations in the cerebellum, one located in the anterior lobe (lobules I-V) and the other in the inferior region of the posterior lobe (primarily lobules VIIb and VIII) (Grodd et al., 2001;Schlerf et al., 2010).
A. Antonioni et al.Interestingly, the cerebellum shows a uniform modular organisation throughout its extension, characterised by a layer of granule cells, which receive input from both cortical areas (i.e. the mossy fibers) and the inferior olivary nucleus complex (i.e. the climbing fibers) and whose axons, i.e. the parallel fibers, synapse both with Purkinje cells, which are the only output of the cerebellar cortex, and with inhibitory interneurons (i.e. the stellate and basket cells), which determine a feed-forward inhibition on Purkinje cells (D'Angelo, 2018).The latter are inhibitory cells and, when activated, they reduce the excitatory activity of the deep cerebellar nuclei (DCN), whose excitatory axons reach the other CNS stations.Specifically, the thalamus emerges as a significant target of DCNs (Habas et al., 2019).Furthermore, additional cell types such as Golgi, Lugaro, and unipolar brush cells finely modulate cerebellar activity, contributing to its remarkably refined computational potential (Schilling et al., 2008).Fig. 1 schematises the complex circuitry of each cerebellar module, while Table 1 provides an overview of the functional roles of specific cerebellar regions.
Given this ubiquitous organisation across all its regions, it is hypothesised that the distinctive computational functions carried out by the cerebellum stem from its specific afferent and efferent connections.These connections facilitate bidirectional communication with neocortical areas, thereby modulating and facilitating various brain processes (Argyropoulos et al., 2020;Habas, 2023).Each Purkinje cell receives information from all three medio-lateral zones via long parallel fibers.This input could integrate various sensory modalities, movement feedback, motor cortex commands, and plans from the PFC/premotor cortex, along with perceptual input from the parieto-occipital association cortex encompassing both intra-personal and extra-personal contexts (D'Angelo, 2018;Mugnaini, 1983).Nowadays, among its many functions, it is deemed pivotal in motor learning, specifically in error-based learning, allowing the subject to maintain its accuracy despite changes and perturbations in the environment and in the body itself (Spampinato and Celnik, 2021).Remarkably, the firing pattern of climbing fibers seems to be modulated by motor errors, leading to a selective weakening of synapses between parallel fibers and Purkinje cells engaged in erroneous performance, while reinforcing connections associated with correct executions (Zang and De Schutter, 2019).In particular, the cerebellum is thought to be pivotal in creating internal feed-forward models to minimise the difference between the predicted and the obtained sensory outcomes of a motor act based on the copy of the efference (i.e., the copy of the motor programme to be executed, which is sent to the cerebellum through the cortico-pontine fibres), the current sensory state of the system (based on vestibular and spinal cerebellar afferents), and the sensory feedback sent by the periphery upon movement completion (Ishikawa et al., 2016).It is worth highlighting that, while sensory feedback from the periphery reaches the cerebellum after the action, other inputs enable it to anticipate the likely sensory outcome and, in instances where there are disparities between expected and actual outcomes, the cerebellum implements corrective actions to minimize this difference (Tanaka et al., 2019).In this sense, the cerebellum is crucial in predicting the sensory consequences of motor acts and is critically involved in selecting the most appropriate actions to maintain coherence within the internal model of environmental interactions.More dated findings also attributed a pivotal function to the cerebellum in the inverse model, which, in contrast to the feed-forward model, computes the motor command to achieve the intended changes Fig. 1.Schematic representation of the complex cerebellar circuitry and its main components.The vertical lines on the right show the subdivision into layers: the molecular layer (blue line), the Purkinje cell layer (yellow line), the granule cell layer (red line), and the layer of white matter fibers and the deep cerebellar nuclei (green line).On the left, examples of cerebellar glomeruli are shown, formed by the synapse between mossy fibers (in gray, originating from pontine nuclei that transmit information from cortical areas and also send a copy of this information to the deep cerebellar nuclei), the Golgi cell axon (violet cell), and the granule cell dendrites (green cell).In the same context, additional modulation is provided by the unipolar brush cells (light blue cells).The axons of the granule cells reach the molecular layer, where they give rise to parallel fibers, providing excitatory input not only to the extensive and branched dendritic arbors of the Purkinje cells (dark pink cells) but also to the inhibitory interneurons that feed-forward inhibit the Purkinje cells.Specifically, the stellate cells (red interneurons) mainly inhibit the apical portions of the Purkinje cell dendrites.In contrast, the basket cells (yellow interneurons) exert control primarily over the basal portion of the dendrites and the cell body of the Purkinje cells, which are located in the layer named after them, along with the soma of the Golgi cells and the Lugaro cells (dark orange cells).The climbing fibers, originating from the inferior olivary nucleus complex, also project 1:1 onto the soma of the Purkinje cells (similarly to the mossy fibers, they project a collateral branch to the deep nuclei in a topographically specific way).The inhibitory axons of the Purkinje cells represent the only output from the cerebellar cortex and reach the deep cerebellar nuclei, which, in turn, project to numerous targets in the central nervous system mainly via the thalamus.Note that various types of astrocytes in the different layers provide trophic and metabolic support to the complex activities performed, such as brown, blue, and light orange cells.Based on (D'Angelo, 2018; Schmahmann, 2019).A. Antonioni et al. in the body's state, establishing the link between planned actions or goals and the motor instructions required to achieve them (Ito, 2013;Torricelli et al., 2023;Wolpert et al., 1998).
While the prevailing model of its operation is now considered the feed-forward one, it is evident that the cerebellum plays a crucial role in identifying the actions that align with the desired sensory outcome.Importantly, the MNS can also implement both hypothesised models of motor control.Specifically, the connections from STS to parietal cortex and, further, to MNs in the ventral premotor cortex could represent an inverse model, which converts the visual representation of observed actions into motor plans (Miall, 2003).Conversely, the reverse pathway from MNs in the ventral premotor cortex to parietal areas and, then, to STS could encode a forward model, leading to the conversion of the motor plans back into anticipated visual representations, which are compared with the actually observed visual images (Iacoboni et al., 2001).Thus, these dual streams potentially facilitate action imitation, since observed actions are initially transformed into potential motor commands by the inverse model, and the visual outcomes of these movements are then predicted by the forward model for comparison with reference visual images (Iacoboni et al., 1999).Given the cerebellum's extensive bidirectional connections with nearly all areas crucially implicated in both models, and its receipt of all requisite information to participate in both the inverse and feed-forward models, it seems reasonable that it could contribute to the function of the MNS in this regard (Caligiore et al., 2013;D'Angelo and Casali, 2012;Popa and Ebner, 2022).Indeed, as emphasised by Miall, research exclusively focusing on the involvement of neocortical areas in the MNS fails to acknowledge the significance of internal models produced by the cerebellum, which are arguably crucial in this context, and overlook its capacity to modulate motor commands through its connections with premotor areas and its reception of sensory input from parieto-temporal regions (Dum and Strick, 2003;Kakei et al., 2019;Miall, 2003).Moreover, importantly, the cerebellum also shows numerous bilateral connections between neocortical areas with mirror properties, another aspect that could support its involvement in these tasks (Dum and Strick, 2003;Errante and Fogassi, 2020;Middleton and Strick, 1997).
Crucially, this cerebellar function has found considerable mathematical support in Active Inference Theory (AIT), which argues that the CNS operates in a Bayesian fashion by updating expectations about the sensory consequences of an action based on contextual information and a priori hypotheses, and adjusting internal models when inconsistencies arise (Pezzulo et al., 2015).Hence, the motor and sensory domains are intertwined, constituting a loop that enables continual refinement of interactions with the environment (Kim, 2021).According to this interpretation, the cerebellum is crucial in constantly updating the internal models and in assisting the selection of the most suitable actions to minimise the disparity between expected and achieved outcomes (Friston et al., 2016).Importantly, within this framework, the MNS might play a pivotal role in implementing Bayes-optimal perception of both self-generated and observed actions, representing the neuronal substrate that enables the simulation and understanding of actions executed by oneself or others (Friston et al., 2011).
Furthermore, it is noteworthy that the involvement of the MNS extends beyond motor functions, as multiple lines of evidence have revealed its participation in the somatosensory domain.Specifically, MNs exhibit activity during sensory perception and when observing another individual performing actions with similar sensory outcomes (Avenanti et al., 2007;Gazzola and Keysers, 2009).In this context, somatosensory areas could support the perception of others and, above all, of their actions, given their connectivity with key motor components of the MNS (Jacquet and Avenanti, 2015;Keysers et al., 2010).Specifically, while motor areas of the MNS may encode the motor features of observed actions, somatosensory areas might process the magnitude of perceived proprioceptive and tactile feedback (McGregor et al., 2016;Valchev et al., 2017).Indeed, arguably, the AON also entails simulations of the anticipated somatosensory outcomes during AO, integrating somatosensory and motor representations and supporting the current perspectives on the sensorimotor nature of the motor system (Gallivan and Culham, 2015;Valchev et al., 2016).Consistently, studies have highlighted that the somatosensory outcomes of AO impact activation within AON (Morrison et al., 2013).Thus, the MNS allows for the internal simulation of motor actions and the perceptual aspects of experiences.It could, therefore, represent the substrate for mental simulation and evaluation of implications in different components of the sensorimotor loop.
Since the MNS predicts the motor and sensory consequences of observed actions, it is reasonable to assume that the cerebellum collaborates closely with the cortical regions of the AON to achieve this purpose (Maranesi et al., 2014;Sokolov et al., 2017;Thach et al., 1992).Consistently, beyond its well-established role in the motor network, significant cerebellar activation on the sensory side has also been demonstrated, as it receives input from virtually all sensory systems, is activated by tactile stimulation without movement, and the dentate nucleus is closely related to sensory acquisition and discrimination tasks (Fox et al., 1985;Gao et al., 1996;Tanaka et al., 2021).Furthermore, the cerebellum is also critically involved in the so-called "principle of reafference", which enables the subject to differentiate sensory consequences stemming from external events from those related to the

Table 1
The functional roles of specific cerebellar regions.In bold the most representative functions.Abbreviations: B: bilaterally; L: left; R: right.Based on (King et al., 2019;Schmahmann, 2019).subject's movements, likely due to the contribution of the efference copy (i.e. a strict spatiotemporal contingency between the motor command and sensory outcome), which is crucial in the internal simulation processes of observed actions (Blakemore et al., 1999;Lindner et al., 2006).
To summarise, a cerebellar forward model could employ an efferent copy of motor commands from M1 and the premotor areas to facilitate somatosensory and/or visuomotor adjustments to the motor representation stored in the parietal areas (Allen et al., 2005;Picazio et al., 2016;Pisotta and Molinari, 2014;Welniarz et al., 2021).Thus, the cerebellum, together with the posterior parietal cortex, could be the crossroads between the motor and sensory systems, in order to process the implications of AO and support the AON in this complex process (Gazzola and Keysers, 2009).This interpretative model could elucidate the observation that AO does not invariably result in imitation, reasonably due to the involvement of a motor inhibitory mechanism where connections between the cerebellum and PFC might be pivotal, as demonstrated by imitative behaviours in patients with lesions affecting both these regions, as cerebellar lesions could manifest "frontal-like" symptoms (Botez-Marquard and Botez, 1993;Mandolesi et al., 2001;Picazio et al., 2020;Schmahmann, 2023).Importantly, the complex interplay among various regions of the AON may account for diverse levels of encoding of AO, ranging from fundamental representations to those invariants regarding trajectory and, ultimately, to those indifferent to the effector employed (Wolpert et al., 2003).Taken together, these findings imply that AO involves a complex interplay of factors, often recently identified, and further studies are needed to deepen their implications (Kemmerer, 2021).While basic action characteristics, such as kinematics, influence AO in a bottom-up manner, top-down factors like purpose and semantics, shaped by interactions between the cerebellum and neocortex, refine strategies aimed at ultimately imitating observed behaviours (De Marco et al., 2020;Errante and Fogassi, 2019;Li and Mrsic-Flogel, 2020;Xu et al., 2023).Indeed, learning by imitation is not a mere copying of AO, but rather a transformation into an action as similar as possible in terms of the goal and the motor strategies (Torriero et al., 2007).Consistently, the observer's motor system is not merely triggered by visual aspects of AO, but also aids in filling in gaps, providing an implicit understanding of others' actions.Therefore, essentially, the MNS may have the ability to anticipate action outcomes and select the best response to the observed behaviour (Caggiano et al., 2011;Craighero et al., 2008).
Here, evidence on cerebellar activity in the context of tasks involving the AON, i.e.AO and motor imagery (MI), will be analysed.

The cerebellum and AO in healthy subjects
AO is crucial in everyday life, as it not only allows one to perform numerous common actions but also to improve social interactions by interpreting other body signals to understand emotional states and intentions (Blake and Shiffrar, 2007;de Gelder, 2006).Moreover, observational learning significantly contributes to the acquisition of motor skills.Thus, a better understanding of the mechanisms underlying this complex phenomenon may improve the development of techniques to enhance the rehabilitation of motor skills after brain damage (Buccino, 2014;Garrison et al., 2013;Porro et al., 2007;Ramsey et al., 2021).Indeed, repeated AO has been shown to induce experience-dependent changes in the motor network, reflecting the acquisition of finely-tuned and effector-specific representations of movement based on the involvement of brain areas implicated in encoding observed actions (Celnik et al., 2006;Stefan et al., 2005;Valchev et al., 2015).Consistently, the visual perceptual system of vertebrates shows a selective preference and high efficiency for biological movements (i.e.motor patterns related to motion in living organisms) and, indeed, humans are able to recognize a conspecific movement in a series of point-lights attached to the limbs of an actor moving in a way that simulates biological actions (e.g.running, dancing), but not under stationary conditions (Johansson, 1973;Vallortigara et al., 2005).This simple experimental paradigm allows the subject to discriminate the gender and identity of familiar individuals, walking direction, social dispositions, and complex actions (Cutting, 1978;Dittrich, 1993;Mather et al., 1992;McArthur and Baron, 1983).Importantly, recent evidence has confirmed the pivotal role of the MNS in performing these essential activities in everyday life (Troje and Chang, 2023;Ulloa and Pineda, 2007).Moreover, notably, recognising biological movements may serve as the foundation for imitation learning, where MNS is thought to play a crucial role (Huber et al., 2009;Iacoboni, 2009).Indeed, a study employing psychophysical and electromyographic assessments revealed enhanced learning of a new, complex motor skill when individuals observed another person learning the same task, compared to merely witnessing the task being performed or observing the learning of a different task (Mattar and Gribble, 2005).
Of note, it has been suggested that perception and recognition processes are based on the implicit understanding of motor rules, providing the basis for identifying biological movement and various evidence has pointed out that the cerebellum is an essential component of the neural circuitry involved in motion perception (Decety and Grèzes, 1999;Viviani and Stucchi, 1989).Indeed, the perception of a motor behaviour enacted by another individual likely triggers our prior knowledge about the action and its expected outcomes, effectively simulating its actual performance by engaging our action system through mental simulation, and the cerebellum is reasonably crucial in this regard (Savaki and Raos, 2019).Consistently, brain-cerebellar loops, employing the motor system as a forward model, could play a fundamental role in sensorimotor prediction and internal simulation of movement (Callan et al., 2013;Casiraghi et al., 2019).Unfortunately, functional magnetic resonance imaging (fMRI) investigations have yielded inconsistent findings regarding cerebellar activation during AO and, indeed, meta-analyses have shown highly varied outcomes, ranging from no (Caspers et al., 2010) or minimal (Molenberghs et al., 2012) to marked activity (Van Overwalle et al., 2014).However, an important study employing the 14 C-deoxyglucose autoradiographic method in the rhesus monkey showed that the AO of a reaching and grasping movement correlates with selective activation of the lateral cerebellar lobules V/VI (Raos and Savaki, 2021;Stamos et al., 2010).Consistently, in humans (Grèzes et al., 2001;Grossman and Blake, 2001;Servos et al., 2002), some studies have confirmed significant activation, both in the anterior, near the midline, and in the lateral cerebellar regions during the perception of biological movements (Grossman et al., 2000;Vaina et al., 2001).Indeed, brain MRI and positron emission tomography (PET) studies have demonstrated significant cerebellar activation during the AO of human walking (Grossman et al., 2000;Ptito et al., 2003;Vaina et al., 2001).Moreover, Vaina et al. documented activation of the lateral cerebellum during the biological movement recognition task of walking (Vaina et al., 2001).Coherently, the cerebellum plays a key role in motor activities and is closely connected to the MNS areas, such as the STS.Thus, it is likely involved in mentally simulating observed actions, also in order to understand others' actions and intentions (Iacoboni and Dapretto, 2006;Puce et al., 1998;Rizzolatti and Fabbri-Destro, 2008).However, it is important to point out that the two cerebellar hemispheres have cross-connections with brain areas, and this could justify the functional specialisation of each hemisphere (Jansen et al., 2005;Wang et al., 2013).
Furthermore, it seems that AO, in addition to automatically triggering the motor resonance underlying action recognition, is also crucial for emotion recognition by similar mechanisms (Carr et al., 2003;Iacoboni, 2005).Interestingly, healthy subjects can easily infer emotions, desires, and intentions only through body motion dynamics.It is reasonable that the cerebellum, with its key role in the motor network and its connections to limbic regions, is fundamental in this context (Chouchourelou et al., 2006;Heberlein et al., 2004;Jung et al., 2022).Similarly to the motor side, emotions are also characterised by an "unconscious drive" that prompts us to mimic the facial expressions of others instinctively, recognise the corresponding emotion and attribute A. Antonioni et al. it to the others (Dimberg et al., 2000;Leslie et al., 2004).This could represent the perceptual-motor resonance mechanism underlying the motor theory of empathy, in which the AON is crucial (Baird et al., 2011;Preston and de Waal, 2002;Rizzolatti and Craighero, 2005).Indeed, a specific cerebellar network activates during social mentalising (Van Overwalle et al., 2020, 2015a, 2015b).Notably, increased fMRI activation has been detected in the right STS, which shows a prominent role during visual processing of body motion, and the left lateral cerebellum (specifically crus I and lobule VIIB) (Sokolov et al., 2012), in particular when individuals detect false intentions during AO (Grèzes et al., 2004;Pavlova et al., 2007;Zacks et al., 2006).Coherently, a PET study has revealed an activation of the left lateral cerebellum, specifically during the identification of dynamic emotional facial expressions, as opposed to static ones, highlighting a cerebellar engagement in visual social cognition via body movement (Kilts et al., 2003).Similarly, as assessed through a fMRI evaluation, a left lateral cerebellar activation has been noted when perceiving dynamic emotional expressions of the entire body, in contrast to static ones (Grèzes et al., 2007).On the other hand, a PET evaluation has highlighted a right cerebellar activation during the observation of meaningless actions, perhaps because of its connections with the semantic areas of the left hemisphere to attempt to explain the observed behaviour (Grèzes, 1998;Murdoch, 2010).Indeed, a PET study has shown that observation of meaningless actions is associated with activations in the right cerebellar hemisphere, while meaningful actions mainly activate the left cerebellar hemisphere (Decety et al., 1997).This emphasises that the AON is critically modulated by higher-order aspects of the AO, i.e. emotional aspects and meaning of actions, and that the cerebellum confirms functional lateralisation based on its connections with brain hemispheres (Bidet-Ildei et al., 2020;Cuijpers et al., 2006;Schmahmann, 2019).Interestingly, recent fMRI findings showed that cerebellar involvement in processing stimuli with an emotional valence spans beyond the visual domain and extends to the auditory side, as highlighted by studies on vocal emotion recognition (Adamaszek et al., 2017;Uono et al., 2017).Consistently, cerebellar regions such as the vermis, lobule VI, and crus I are connected to the dorsolateral PFC and auditory temporal regions, including the amygdala, which are crucial in the context of emotion recognition (Stoodley and Schmahmann, 2009).Given the complexity of these activities and the inter-subjective variability in assigning value to emotional stimuli, it is not surprising that literature findings are heterogeneous, with cerebellar activations ranging from unilateral (both in the right and in the left hemisphere) to bilateral (Alba-Ferrara et al., 2011;Kotz et al., 2013;Wildgruber et al., 2005).Reasonably, individual differences, experimental paradigms, and task complexity may partly explain different results.However, these findings support the cerebellum's involvement in recognizing the emotional valence of stimuli, regardless of their modality.This highlights its role not only in understanding others' behaviors, but also in predicting others' actions and updating internal response models coherently, probably also with the contribution of the basal ganglia, providing a cost-benefit assessment to modulate responses and avoid resource wastage (Adamaszek et al., 2019).Indeed, understanding a message's emotional content allows for verifying prior hypotheses about the nature of the interaction, updating internal models, and anticipating the consequences of potential discrepancies from expectations.
Notably, a recent work evaluated the specific localization of cerebellar regions engaged during AO of a grasping movement, highlighting an activation in cerebellar lobules VI, VIIb, and VIIIa and supporting its active contribution in encoding the hand movements of others (Abdelgabar et al., 2019).This evidence aligns with other studies showing activation of these cerebellar regions, particularly during the performance of complex tasks, leading us to hypothesise that the involvement of motor control regions during AO suggests that cerebro-cerebellar circuits associated with precise hand movements might also play a role in analyzing subtle kinematic details during AO (Abdelgabar et al., 2019;Fuentes and Bastian, 2007;Schlerf et al., 2010).Indeed, a strong activation of lobules VI and VIIA is also noted during motor tasks, particularly when movements entail complex actions demanding higher levels of sensorimotor coordination beyond simple one-dimensional joint actions (Diedrichsen et al., 2005;Miles et al., 2006).Supporting this interpretation, Biagi et al. also demonstrated the activation of cerebellar areas during AO of varying difficulty, suggesting that it may support the computations of the parietal cortex particularly during the observation of complex actions, i.e. those in which feed-forward models are most involved (Biagi et al., 2010).These data are also in line with the findings of Schlerf et al., who pointed out that, while the anterior cerebellar regions are activated indifferently for both simple and complex tasks, the neocerebellum, on the other hand, shows greater activation as task complexity increases (Schlerf et al., 2010).Consistently, Errante & Fogassi highlighted that, both during AO and action execution, the activation patterns (as assessed by shared voxels in fMRI evaluation) are mainly bilateral in cerebellar lobules V, VI, crus I, VIIIa and VIIIb (Errante and Fogassi, 2020).Interestingly, this bilateral activation also seems to be modulated by the degree of task complexity, as more demanding tasks result in bilateral neocerebellar activation that may depend on a similar bilateral activation of connected brain regions, probably to coordinate multiple effectors during high-level motor planning (Hanakawa et al., 2005;Thach et al., 1992;Verstynen et al., 2005).Additionally, another recent study comparing AO and action execution by means of a grip task has demonstrated shared activation patterns in both the anterior and posterior cerebellar regions (Casiraghi et al., 2019).They align with a recent fMRI investigation within a neuropsychological test battery, revealing notable activation during AO in various cerebellar regions, notably the posterior hand motor area, and to a lesser extent, the anterior region linked to primary action execution (King et al., 2019).Notably, the activation pattern during complex movements supports the functional subdivision between anterior, linked to direct action execution, and posterior cerebellar regions, more involved in action planning and comprehension.Indeed, similarly, one study found that physical learning was most associated with activity in the left anterior/medial cerebellum, while observational learning led to an activation in the right posterior cerebellum (Cross et al., 2009).However, other evidence suggests that the left lateral cerebellum also participates in observational learning, probably depending on the type and complexity of the task and the subject's experience (Torriero et al., 2007).Furthermore, Calvo-Merino dissociated the motor aspects from the visual component of the AO, showing a greater activation in the MNS in subjects experienced in the observed complex action than in novices (Calvo-Merino et al., 2006).This correlation between cerebellar activity, the complexity of the observed task, and the subject's experience could be attributed to various non-mutually exclusive mechanisms.On one hand, it is plausible that complex actions require higher computational demands to process the action and predict its sensorimotor consequences.This task becomes increasingly demanding with the complexity of the movements (both executed and observed), and the cerebellum might be pivotal in this regard (Avanzino et al., 2015;Popa and Ebner, 2019).On the other hand, complex tasks activate a wide, sometimes even bilateral, range of cortical areas involved in sensorimotor integration and cognitive functions (such as attention and mental flexibility) (Gohil et al., 2015;Puglisi et al., 2018).Reasonably, the cerebellum actively supports these cortical areas by identifying the internal models that best match the observed action (Streng et al., 2022).Furthermore, recent evidence suggests that the cerebellum is crucial in interpreting the so-called submovements, i.e. the elementary units composing seemingly fluid and continuous movements (Houk et al., 2007;Langlois et al., 2024).Indeed, although actions may appear similar in terms of movement and outcome, it has been shown that the combinations of submovements are numerous and depend on a complex interaction among various factors, including the subject's experience and ability to achieve optimal efficiency in effector combinations (Torricelli et al., 2023).Therefore, it is conceivable that, as a subject's experience in a particular context increases, the cerebellum can detect and compute differences in patterns of submovements compared to its own repertoire, A. Antonioni et al. evaluating finely the sensorimotor differences between various internal models (Habas et al., 2020;Langlois et al., 2024;Streng et al., 2022).Since movements may appear similar, the ability to perceive differences in submovement patterns is reasonably influenced by the subject's experience and, consequently, by the number of available internal models (Langlois et al., 2024).Therefore, a proportional cerebellar activation based on the subject's experience seems to be consistent with this interpretation.This is particularly relevant considering that submovements are now deemed fundamental in mediating interpersonal synchronization and coordination, essential for effective communication (Nazzaro et al., 2023;Tomassini et al., 2022).This evidence and the AON's crucial role in social interaction further support a connection between the cerebellum and the MNS, bolstered by these kinematic findings (Schmidt et al., 2021).Notably, the same correlation regarding cerebellar activation, task complexity, and subject experience is also observed in the context of MI, further supporting this perspective (see Section 3).Moreover, consistently, cerebellar regions connected with the dorsal visual pathway could encode the kinematic features of observed actions.Together with the representation of the action goal, this process triggers an automatic motor resonance dependent also on the motor expertise of the observer (Calvo-Merino et al., 2006;Errante et al., 2023;Errante and Fogassi, 2020;Fukui and Inui, 2013).Similarly, instrumental and expressive contrasts have shown that, among the brain regions particularly associated with recognising gestures as opposed to simply observing them, the cerebellum demonstrates significant activation in the right cerebellar hemisphere, likely within the crus I or II regions (Gallagher and Frith, 2004).This confirms recent evidence suggesting a profound link between semantic comprehension and gesture recognition and highlights the cerebellum's involvement through its crossed connections between the right cerebellar regions and the language network, lateralized in the left hemisphere (Proverbio et al., 2015;Schmahmann, 2019).
Of note, shared activation during both AO and action execution is also detected in the dentate nucleus, which is the main output of the lateral cerebellum and is critically involved in the modulation of visuomotor tasks in a effector-specific way (Errante and Fogassi, 2020;Miall et al., 2001;Strick, 1985).Indeed, the cerebellar activity patterns are similar during both the observation and execution of hand, mouth, and foot actions, suggesting a segregation of activities in the MNS that also depends on the specific effector used (Errante et al., 2023;Gazzola and Keysers, 2009).Moreover, the ventral region of the dentate nucleus projects to the posterior parietal and the dorsal PFC which represent fundamental stations for cognitive functions (e.g.working memory) and for the "dorsal stream" of visual processing (Dum and Strick, 2003).This suggests the importance of this circuit in forming and detecting high-level goals and intentions (Caligiore et al., 2013).Additionally, a noteworthy observation is that a distinctive characteristic of the human MNS, compared to that of the monkey, is its ability to encode the temporal dynamics of observed actions.Given the cerebellum's significant involvement in temporal perception, it could potentially play a pivotal role in this aspect (Gangitano et al., 2001;Koch et al., 2009;Mioni et al., 2020).Indeed, cerebellar processing of temporal expectations utilises timing cues to enhance motor or perceptual efficiency (Avanzino et al., 2015;Nobre et al., 2007).Taken together, these findings suggest that the motor system, especially during AO of complex behaviour, internally simulates the action also involving subcortical structures, as highlighted by the activation of the dentate nucleus and thalamus, which indicates a closed-loop simulation reasonably to achieve a prediction of sensory outcomes.Indeed, this cortico-cerebellar loop, akin to actual execution, updates cortical hand motor representations based on AO outcomes (Bijsterbosch et al., 2011;Errante and Fogassi, 2020).
However, it is interesting to note that areas active during both AO and execution of actions, including the cerebellum, display a significantly greater activity when the subject must later reproduce the observed action, perhaps because the modulation from higher brain regions influences pre-existing motor prototypes that represent individual actions or interactions between the subject and the environment (Frey and Gerry, 2006;Johnson-Frey et al., 2003).Specifically, Frey and Gerry showed a heightened bilateral cerebellar activation during passive AO.Crucially, this cerebellar activation is further increased by the subsequent intention to repeat and/or to learn the observed sequence of actions (Frey and Gerry, 2006;Grafton et al., 1998Grafton et al., , 1996)).This is particularly relevant as recent evidence has linked the cerebellum not only to motor learning processes but also to creativity, i.e. the ability to generate ideas fluidly and to devise alternative solutions (Adamaszek et al., 2022).Consistently, motor learning is not merely a motor repetition but often involves the integration of complex sequences and their adaptation in creative ways, suggesting a link between motor learning and creative thinking in which the cerebellum might be essential (Spampinato and Celnik, 2021;Sunavsky and Poppenk, 2020).
Notably, it has been recently suggested that the development of these cerebellar activities might have played a crucial role in driving technological advancement in humans, providing intriguing insights into the cerebellum's role from an evolutionary perspective (Vandervert et al., 2024).
Moreover, Decety highlighted that observing with the intention to recognise leads to an activation in structures involved in memory encoding, while observing with the intention to imitate is linked to activity in the dorsolateral PFC, the SMA, and the cerebellum (Decety, 1996).Importantly, a study has highlighted that bilateral cerebellar activation occurs specifically during AO intended for imitation, whereas mere observation without the subsequent intention to imitate is associated with significantly fewer activation patterns (Chaminade et al., 2002;Decety and Grèzes, 1999).Therefore, cerebellar activity, together with a broad network that also includes various cortical areas of the AON and subcortical regions such as the thalamus and basal ganglia, also seems relevant in the strategy used for imitation, reinforcing the postulated role in optimising the selection of actions to achieve the desired sensory outcome or, perhaps, in inhibiting the execution of the mentally simulated action until the proper moment to perform it (Grèzes, 1998;Li and Mrsic-Flogel, 2020).While these hypotheses are not mutually exclusive, the latter possibility seems particularly coherent with the strong correlation between the cerebellum and the dorsolateral PFC.Indeed, the inhibitory role of the latter in motor control and its significance in planning and selecting optimal strategies is well-established (Habas, 2021;Oldrati et al., 2021).This suggests that the cerebellum strictly interacts with this neocortical area to select the optimal behavior for achieving goals in a context-specific manner (D'Mello et al., 2020).Finally, it is interesting to note that disinaptic pathways have been observed linking the cerebellum and the basal ganglia, suggesting a joint effort between the two subcortical regions to select the best action (the former) and reinforce it through the dopaminergic reward circuits (the latter) (Bostan et al., 2013(Bostan et al., , 2010;;Hoshi et al., 2005;Manto et al., 2024).Table 2 (first section) summarises this evidence.

The cerebellum and MI in healthy subjects
MI, i.e. a dynamic phase in which the mental representation of a specific motor action is internally rehearsed without any observable motor output, represents a fundamental simulation mechanism when observing the actions of others (with the aim of imitating them or not) (Decety and Grèzes, 1999;Lotze and Halsband, 2006;Munzert et al., 2009).Indeed, similar to other types of memory, it is believed that individual motor acts are stored, modified and recalled according to a precise hierarchical representation to compose specific sequences of actions (Sheahan et al., 2016;Stringer and Rolls, 2007).Importantly, however, MI entails a vivid mental simulation of movements and is associated with a kinesthetic sensation (Hurst and Boe, 2022;Jeannerod and Frak, 1999).Since all these processes require two phases, namely one of representing the body as a force-generator and a subsequent one A. Antonioni et al. of encoding the goal of the programmed action, the neural substrates largely overlap with those of motor planning and, therefore, with the AON (Grèzes and Decety, 2001;Kim et al., 2017;Stephan et al., 1995).Studies across psychology, electrophysiology, and neuroimaging suggest that performing a task and imagining it engage overlapping neural circuits both in the sensory (e.g.visual) and motor domains, leading to a subthreshold activation of descending motor pathways (Dijkstra et al., 2017;Pilgramm et al., 2016;Porro et al., 1996;Rizzolatti et al., 2002).Indeed, MI has been shown to activate a network of areas, including the SMA, premotor and parietal cortex, and the cerebellum (Jueptner et al., 1997;Parsons et al., 1995;Savaki and Raos, 2019).Consistently, a recent review has documented a specific cerebellar activation in the right lobule VI, which contains multiple body representations, during motor execution and, to a lesser extent, during MI (Buckner et al., 2011;Hardwick et al., 2018).It is possible that the cerebellum, together with other neocortical regions, analyses the efference copy from a motor and perceptual perspective to simulate the action and its most reasonably expected sensory consequences and, indeed, it appears that its activity during MI increases proportionally with the complexity of the task (importantly, as already discussed for AO) (Boecker et al., 2002;Cebolla et al., 2015;Grush, 2004;Kuhtz-Buschbeck et al., 2003;Yi et al., 2013).Indeed, MI of complex movement sequences leads to a significant cerebellar activation, as movement execution and AO do (Decety et al., 1990;Jokisch et al., 2005;Krüger et al., 2020;Luft et al., 1998;Ryding et al., 1993).Coherently, there is substantial evidence to show that the cerebellum is activated during sensory-motor imagery tasks, such as producing simple speech utterances, generating verbs to noun stimuli or performing a finger tapping, highlighting a functional specialisation of the right cerebellar hemisphere for linguistic tasks and of the left A. Antonioni et al. cerebellar hemisphere for spatial tasks, reasonably due to the cross-connections with corresponding brain areas (Ackermann et al., 1998;Hanakawa et al., 2008;Petersen et al., 1989;Raichle et al., 1994;Stoodley, 2012).Interestingly, the cerebellum shows extensive connections with the dorsolateral PFC, which is pivotal in working memory and could, indeed, play a role in the verbal rehearsal loop by comparing subvocal articulation with phonological representations through a MI speech mechanism, potentially correcting errors to maintain the integrity of rehearsed items (Deschamps et al., 2020;Desmond and Fiez, 1998;Fiez and Raichle, 1997).This seems similar to the proposed role of the cerebellum in episodic memory, wherein it could aid in self-initiated retrieval or the generation of 'candidate responses' during retrieval (Almeida et al., 2023;Bäckman et al., 1997;Cabeza et al., 1997).Consistently, subcortical structures, including the cerebellum, contribute to the spatiotemporal functions and programmes for self-generated movements (Deecke and Kornhuber, 1978;Kilteni and Ehrsson, 2020).Indeed, a simple imagination of a graphic movement (i.e. writing a number by simulating the action in the first person and 'feeling' the hand writing) leads to the activation of a network of areas that also includes the cerebellum, and the same occurs if one imagines a tennis practice (Decety et al., 1988;Decety and Ingvar, 1990).Importantly, a recent meta-analysis has confirmed the activation of several cerebellar regions during MI tasks (Hétu et al., 2013).Therefore, the cerebellum is involved not only during AO, but is also crucial when a subject makes use of one's internal (imagined) voice or sensation to represent, maintain, and process task-relevant information and conscious reflections, supporting its critical role in regulating internal models in order to achieve goals profitably and accurately (Ito, 2008;Strick et al., 2009).
Indeed, internal models can be conceptualised as forms of sensorimotor imagination that allow precise error detection based on the synchronous input from distinct representation maps (de Solages et al., 2008;Fuentes and Bastian, 2007;Gold and Ciorciari, 2021).Consistently, also in the context of MI, cerebellar activity is mainly located postero-laterally, supporting that this is not a simple motor act but rather a cognitive task that relies on interplay with other brain areas (Allen et al., 1997;Grami et al., 2022;Hanakawa et al., 2003).Some authors have suggested that, during MI, cerebellar activation indicates an inhibitory process that blocks efferent impulses, preventing their transmission to the medullary and muscle levels (Berthoz, 1996;Cengiz and Boran, 2016;Decety and Ingvar, 1990;Leonardo et al., 1995).Coherently, Sakai et al. reported a maximal activation of the lateral cerebellum (lobule VII) during non-go motor tasks, and it is reasonable that the same mechanism occurs during MI (Sakai et al., 1997).This aligns with evidence indicating that MI places greater demands on inhibitory mechanisms related to AO, attributed to the internally generated sensations of movement, and some authors interpret MI as involving conscious access to sensory-motor action contents, while the preparation of the movement remains subconscious until the action is initiated (Crammond, 1997;Lotze et al., 1999).Interestingly, cortical areas of the MNS are activated in a somatotopically specific fashion when a subject reads words related to the movement of particular body parts, and it appears that the cerebellum also intervenes in this internal simulation (Fuentes and Bastian, 2007;Hauk et al., 2004;Leslie et al., 2004).Furthermore, when examining studies regarding the MI of gait, seminal work has shown that it also critically involves the cerebellum, in particular the anterior lobe for the kinematic simulation of gait and lobules VI and VII (crus I and II) in relation to the vividness of MI and the complexity of the performed task (Cabaraux et al., 2023).Interestingly, increased bilateral activation of crus I and II distinguishes those with better MI of gait performance from poor "imagers" (van der Meulen et al., 2014).
Thus, mentally simulating motor behavior could activate the motor programme, by engaging various parts of the CNS that execute different aspects of the imagined action (Decety, 1996;Ladda et al., 2021).
Supporting the interpretation that neural processes involved in executing specific motor tasks are also recruited when mentally simulating those movements, it has been shown that the time to act overlaps with that required for the MI of the same task (Decety et al., 1989;Decety and Michel, 1989).However, mentally simulating and physically executing simple actions, like grasping and picking up an object, activate cortical areas that exhibit only partial overlap (Decety et al., 1994;Grafton et al., 1996).Moreover, directly comparing mental rehearsal to physical performance has revealed partial involvement of distinct brain regions (Gerardin et al., 2000;Hanakawa et al., 2005Hanakawa et al., , 2003)).This suggests that, even if the same network is involved in action execution, AO and MI, the degree and specific locations of activation depend on the particular activity performed (Decety et al., 1997;Papitto et al., 2020;Van der Lubbe et al., 2021).Specifically, there may be a gradient of activation for areas within the AON, including the cerebellum, particularly its lateral-posterior areas, ranging from MI, to AO and, finally, to execution itself (Hardwick et al., 2018;Henschke and Pakan, 2023;Jeannerod, 2001;Lotze et al., 1999;Nedelko et al., 2012).According to some authors, given the relevant role of the cerebellum also in processing sensory input, it is possible that the lower activation during MI and AO compared to the execution itself also depends on the lack of sensory feedback (Baumann et al., 2015;Gao et al., 1996;Jueptner et al., 1997).However, interestingly, other authors have, instead, found a similar activation pattern between MI and execution in the cerebellum, probably justified by its involvement in motor planning (Kuhtz-Buschbeck et al., 2003), whereas yet other works have highlighted a higher cerebellar activation during MI than during execution, and have hypothesised that this may be due to the inhibitory mechanisms already examined, the computational load due to sustained motor representation or the retrieval of sensory impressions from memory, necessary to compute the feedforward model in the absence of external input (Annett, 1996;Berthoz, 1996;Cengiz and Boran, 2016;Jeannerod and Decety, 1995;Luft et al., 1998).This could also justify the possibility of using MI to perform perceptual tasks, e.g. a mental rotation task (Butson et al., 2014;Vingerhoets et al., 2002).A summary of these findings is provided in Table 2 (second section), while Fig. 2 shows an overview of the brain-cerebellar loops active during the execution of movements, AO, and MI.

The cerebellum, AO and MI in pathological conditions
A seminal demonstration of the importance of the cerebellum in the implicit learning by AO of a visuomotor sequence stems from animal model studies: it has been demonstrated that, when a rat sees another animal exploring a water maze, the observation has the same important enhancing effect in improving subsequent behaviour as the exploration performance itself (Petrosini et al., 2003).The cerebellum has been hypothesized to support the rehearsal of cortical action processes without overt movement (Leggio et al., 2000).However, crucially, if damage to the cerebellum is experimentally induced, the animal loses its ability to learn from observation while retaining access to declarative memory to improve performance (Graziano et al., 2002;Molinari et al., 1997;Petrosini et al., 2003).Of note, in this model, rats observed behaviours without an apparent goal, thus the cerebellum results fundamental in imitation learning by means of AO irrespective of their goal (which is probably more mediated by dopaminergic reward circuits) (Hart et al., 2024).Importantly, a human study that employed an inhibitory protocol of repetitive cerebellar TMS (rTMS) highlighted an inability to learn new visuomotor tasks through AO, without affecting established motor memories.This underscores its vital contribution to the initial stages of observational learning, specifically during the observation and execution matching phase, and emphasises its role in combining sensory information (derived from environmental observation) and motor information (assessed by exploratory behaviour) during AO to generate internal models that underpin learning by imitation (Caligiore et al., 2013;Torriero et al., 2007;Wolpert et al., 1998).
Furthermore, the specificity of the cerebellum during the acquisition phase is highlighted by its activity decline following prolonged practice, indicating optimized internal models and the resolution of discrepancies between predicted and actual outcomes (Penhune and Doyon, 2005).Reasonably, the cerebellum learns to associate specific contexts with consciously initiated movements leading, through repetition, to connections between contexts and movement patterns, enabling automatic action triggering when the context reappears (Kirsch and Cross, 2015;Thach, 1998).Theoretical models also support this evolution of brain activity, as the reduction in cerebellar function leads to so-called 'sensorimotor habits', i.e. the strengthening of the control process for actions that are habitually performed (Hua and Houk, 1997;Synofzik et al., 2008).
Further evidence is derived from diseases affecting human patients.Indeed, data from subjects affected by cerebellar lesions (e.g.atrophy, tumour, stroke, or an autoimmune injury) have shown an important role of the posterior cerebellar regions (particularly the vermis) in judging the speed of moving stimuli (Ivry and Diener, 1991).The authors also emphasised that this impairment is unlikely to depend primarily on disconnection from the cortical areas, not only because primates, which are characterised by a limited amount of connections from visual cortical areas to the cerebellum, show the same deficits in similar experimental contexts, but also because single-cell recordings in these regions demonstrate that cellular activity is best predicted by the speed of external stimuli (Glickstein et al., 1985;Suzuki and Keller, 1988).An important study by Christensen et al. examined this action-perception coupling in patients with chronic cerebellar lesions following tumour removal (Christensen et al., 2014).Interestingly, they demonstrated a dissociation between, on the one hand, the impairment of the sensory-motor representation following damage to the motor representation of the arm in the right anterior and posterior cerebellar cortex (lobules V, VI and VIII), especially in lobule VI, which is crucial for sensory encoding and for predicting and updating the sensory state (Miall et al., 2007;Schlerf et al., 2012); on the other hand, impairment of multisensory integration, depending mainly on lesions of lobule VII and the ventral portion of the dentate nucleus, which are involved in the cerebellar-parietal loop that is crucial for analysing information from different domains (Christensen et al., 2014;Greger et al., 2004;Strick et al., 2009).Consistently, Abdelgabar et al. compared the performance in recognising fine kinematic details of AO between healthy subjects and patients affected by spinocerebellar ataxia type 6 (SCA6, a neurodegenerative disorder that selectively affects cerebellar circuits), demonstrating that the latter show specific deficits in AO tasks (Abdelgabar et al., 2019).Indeed, SCA6 is characterised by a specific loss of Purkinje cells and a selective loss of grey matter in the lobule VI (Du et al., 2013;Rentiya et al., 2017).The authors provided further evidence supporting the notion that AO entails a cortical-cerebellar loop, converting sensory input into motor control strategies (inverse models) and anticipating sensory feedback from motor commands (forward models).Specifically, they attribute the observed deficits to damage of lobules VI and VIIb/-VIIIa, aligning with the examined physiological models (see Section 2).Furthermore, Sokolov et al. emphasized the cerebellum's region-specific activity during AO, particularly in its lateral regions, as evidenced by the fact that tumours affecting the left lateral cerebellum significantly impair visual detection of human movement, whereas lesions in the medial regions have minimal impact on the perception of human locomotion (Sokolov et al., 2010).Of note, the cerebellar role in processing visual information, despite its relatively limited connections with the visual cortical areas, supports evidence pointing to its multiple sources of visual afferences.Specifically, some visual information can rapidly access the cerebellum through a fast pathway originating from the superior colliculus in the midbrain, reaching the cerebellum (particularly the lobules VI and crus I and II) via pontine nuclei (Voogd et al., 2012).Additionally, several cortical areas (both strictly visual and, more prominently, associative in the temporo-parietal regions) provide the cerebellum with processed and integrated visual data through the pontine nuclei (Voogd et al., 2012).This dual source likely allows the cerebellum not only to obtain updated information on the sensory modality most relevant to humans but also to constantly compare bottom-up information from the superior colliculus (i.e.rapid and unprocessed, reflecting the current environmental condition and essential for faster actions such as adjusting eye movements for fast-tracking) with top-down data from cortical areas (i.e.reasonably integrating expectations based on priors and past experiences) (Krauzlis et al., 2013;Strick et al., 2009;Voogd et al., 2012).
It is also interesting to note that patients suffering from autism spectrum disorders (ASD), which are known to be characterised by reduced social skills, manifest not only peculiar alterations during AO and action imitation, suggesting a specific AON dysfunction, but also cerebellar abnormalities, both neuroanatomically and in terms of connectivity (e.g. a reduction in cerebellar white matter and changes in the function of Purkinje cells) (Allen, 2006;Catani et al., 2008;van der Heijden et al., 2021;Yang and Hofmann, 2016).Of note, neurocomputational models of ASD provided further support to the involvement of cerebellar abnormalities in the genesis of characteristic behavioral alterations (Trimarco et al., 2021).In addition, these patients also show alterations in the perception of biological movements (which correlate with alterations in the left superior cerebellar peduncle, closely connected with the right STS), as well as a hypersensitivity to sensory stimuli that may stem from difficulties in attenuating predicted stimuli due to alterations in spatio-temporal matching, a process which may be significantly influenced by cerebellar abnormalities, as previously discussed (Blake et al., 2003;Blakemore et al., 2006;Robertson and Baron-Cohen, 2017;Rogers et al., 2003).Importantly, a specific MNS dysfunction, underpinning a large amount of behavioural and motor alterations in ASD patients, has been suggested, as also demonstrated by fMRI studies showing reduced MNS activation during AO and imitation of facial expressions (Dapretto et al., 2006;Emanuele et al., 2021;Fishman et al., 2014;Williams et al., 2004).
Similarly, it is noteworthy that individuals with schizophrenia exhibit diminished attenuation of self-generated compared to external tactile stimulation, reflecting a re-afference impairment, which may stem from inadequate prediction of movement outcomes, a process in which the cerebellum plays a pivotal role (Shergill et al., 2005).These data are also supported by recent evidence demonstrating altered cerebellar-brain connectivity in schizophrenic patients that could account for their cognitive impairments (Cao et al., 2023;Cattarinussi et al., 2024;Kim et al., 2021;Saito et al., 2018).Importantly, epigenetic investigations have revealed an abnormal aging trajectory in the cerebellum of individuals with both ASD and schizophrenia.This finding may explain the shared anomalies among ASD and schizophrenia and their common alterations in social cognition, which could represent the neurophysiological underpinnings of their deficits in the perception of biological movements (Kim et al., 2005;Klin et al., 2009;Koldewyn et al., 2010;Liu et al., 2023).As suggested by Sokolov et al., these data are also supported by the altered functional connections between the right STS and the left cerebellar lobule Crus I, which are common to both pathologies (Kaiser et al., 2010;Sokolov et al., 2012).Consistently, schizophrenic patients show alterations in the emotion perception in dynamic facial expressions linked to the activity of the left cerebellar hemisphere (Kilts et al., 2003;Kohler et al., 2000).
Furthermore, interestingly, patients affected by neurodegenerative ataxia or generic cerebellar injury show a deficit in acquiring and automating the comprehension of social action sequences related to individuals' mental states during AO, a finding that supports its critical role not only in sequence learning but also in social cognition (Leggio and Molinari, 2015;Van Overwalle et al., 2019, 2014;Vandervert et al., 2024).Nevertheless, most of the cases investigated so far concern chronic and multifaceted conditions that evolve over extended durations, thus allowing for the potential involvement of compensatory mechanisms.Consequently, attributing a definitive causal role to specific brain structures in the development of observed alterations becomes challenging (Draganova et al., 2021).In contrast, stroke is characterised by a sudden onset and, especially in some cases, selectively affects specific brain regions, making it possible to assess the functional consequences of an injury.Indeed, importantly, it has been shown that patients affected by median cerebellar lesions (in the vermis) show a deficit in motion direction perception not only in the acute phase but also in the chronic timeframe, without defective fixation and eye movements and despite the regression of other typical symptoms due to cerebellar lesions (Nawrot andRizzo, 1998, 1995).On the other hand, Cattaneo et al. examined a group of chronic stroke cerebellar patients using a task in which they were asked to identify the intruder in a sequence of biological or inanimate actions, demonstrating marked impairment in both categories (Cattaneo et al., 2012).Of note, the task required them to mentally reconstitute the correct sequence, compare it with the options presented, and identify the inconsistent option.Hence, the authors proposed that the cerebellum, with its internal feed-forward models, plays a critical role in decomposing the sequence into basic motor acts, which are then arranged in a chronobiologically coherent way under normal conditions, whereas cerebellar damage prevented the patients from performing this task correctly (Cattaneo et al., 2012;Molinari et al., 2008).However, in another investigation involving post-acute cerebellar stroke patients, a selective deficit in the discrimination of movement direction but a sparing in the perception of biological movements have been observed, leading the authors to hypothesise that neocortical regions, such as the STS, might play an exclusive or compensatory role in this function, while the cerebellum might enhance efficiency but not necessarily be essential for performance (Bower, 1997;Jokisch et al., 2005).Indeed, some theories suggest that the cerebellum is useful but not indispensable for many brain activities.This notion finds support in the observation that its engagement increases proportionally with task complexity, particularly beneficial when cortical functions require additional support (Salmi et al., 2010;Xiang et al., 2003).However, it is possible that various factors, including the limited sample size, the precise lesion site, and the time from stroke onset (for the intervention of compensation mechanisms) may account for the observed differences.Furthermore, a specific deficit in the recognition and discrimination of cues from both facial and vocal expressions with basic emotional valence has been highlighted in a group of patients affected by cerebellar stroke, predominantly located in the dorsomedial parts of crus I and II (Adamaszek et al., 2019).Further studies have confirmed a selective impairment of vocal emotion recognition following stroke in lobules VIIb, VIII, and IX and in crus I and II, more frequently in the right cerebellar hemisphere (Thomasson et al., 2021(Thomasson et al., , 2019)).These findings support the cerebellum's role in recognizing emotional stimuli across various sensory domains, a crucial aspect of its contribution to the activities of the MNS and the regulation of social interactions.Similar to AO, the lesion of cerebellar regions active during MI also impairs the performance and especially the learning of the same cognitive tasks (Fiez et al., 1992;Thach, 1998).Importantly, a study on patients affected by a unilateral cerebellar stroke showed that not only, as expected, the excitability of contralateral M1 was reduced (due to the well-known phenomenon of diaschisis), but also that motor facilitation during MI and the pre-movement facilitation were reduced, leading the authors to hypothesise a deficit due to the cerebellar lesion to prepare and imagine sequential movements (Battaglia et al., 2006).This result is A. Antonioni et al. also consistent with another finding showing that unilateral cerebellar damage involving the dentate nucleus disrupts kinesthetic imagery of a complex motor task (González et al., 2005).Interestingly, another work compared healthy subjects with Parkinson's disease (PD) patients and highlighted that, while in the former MI led to a cerebellar activation regardless of the imagined hand, in the latter MI using the right ("akinetic") hand was characterised by a different motor network activity with a lack of cerebellar activation (Thobois et al., 2000).This evidence underscores that cerebellar lesions not only hamper execution but also affect AO and MI, reinforcing the notion that these activities rely on the MNS and that the cerebellum makes a crucial contribution to its functions.Table 3 summarises these data.

Neurophysiological and rehabilitation implications
Based on the findings above, the interest in AO and MI has increased in the last decades, both from a prognostic and, above all, therapeutic point of view (Antonioni et al., 2024a;Sun et al., 2016;Zhang et al., 2023).Indeed, reasonably, both MI, which promotes mental rehearsal, and AO, by activating representations within one's motor network corresponding to the observed behaviour, may elicit subthreshold activation of the motor system, thus enhancing performance through sensory-motor neuroplastic mechanisms, as evidenced by the heightened cortico-spinal excitability induced by both (Aoyama et al., 2020;Cochin et al., 1999;Mizuguchi and Kanosue, 2017;Romano-Smith et al., 2018;Tremblay et al., 2004).Consistently, both AO and MI facilitate subsequent motor performances, albeit to a different degree, as suggested by a study in which AO prevented the corticomotor depression induced by immobilization, but not MI (Bassolino et al., 2014;Chye et al., 2022;Gatti et al., 2013;Mulder, 2007).Specifically, multisensory AON enables individuals to (re)learn impaired motor functions by activating internal action-related models (Fadiga et al., 1995;Rizzolatti et al., 2009;Sale and Franceschini, 2012).
This evidence has sparked considerable interest in MI and Action Observation Treatment (AOT) from a neurorehabilitation perspective.These methods offer a non-invasive means to engage the MNS, even for patients unable to execute actions correctly, and by leveraging sensorimotor resonance mechanisms, they hold promise for enhancing performance.Indeed, their potential has been demonstrated across various pathological conditions, from stroke to PD (Binks et al., 2023;Caligiore et al., 2017a;Mao et al., 2020;Mulder, 2007;Small et al., 2012).Agosta et al. intriguingly implemented an AOT protocol in PD patients exhibiting freezing of gait, demonstrating its efficacy in enhancing motor performance and balance.Notably, only the experimental group (i.e. the AOT group) exhibited reduced activity in the cerebellar vermis and bilaterally in cerebellar lobule IV and V compared to the control group, underscoring its therapeutic significance (Agosta et al., 2017).Similarly, in stroke patients, the integration of MI with physiotherapy emerges as a viable and efficacious approach to enhance rehabilitation outcomes (Schuster et al., 2012).Significantly, recent research in severe stroke patients bolstered this crucial application in neurorehabilitation since AO has been shown to impact action prediction time, underlining the involvement of the MNS, even in cases where the affected hand is unable to perform the action (Craighero et al., 2023).Notably, another study on PD patients showed that AOT yields not only motor benefits but also cognitive ones, particularly on verbal and visuo-spatial memory and on attentional abilities (Caligiore et al., 2019).Moreover, these improvements were sustained even a month after the treatment, suggesting a long-term effect.This strengthens the idea that AOT acts influencing multiple cognitive-motor areas and the cerebellum, given its complex and relevant connections with various cortico-subcortical networks, could be pivotal in this context to explain the common positive effects across different domains.
Importantly, neurophysiological studies support the cerebellar role in neurorehabilitation, particularly concerning stroke patients.Indeed, a great deal of research underscores the significant role of the cerebellum in motor network reorganization during stroke recovery (Olafson et al., 2021).Furthermore, it is noteworthy that animal research has demonstrated that stimulating cerebellar-cortical networks can enhance functional recovery after stroke (Machado et al., 2013;Park et al., 2015).Additionally, fMRI studies in stroke patients have revealed that heightened activity in the contralesional cerebellum correlates with improvements in gait ability (Luft et al., 2005).Stroke survivors frequently need to acquire novel motor strategies, and it stands to reason Table 3 Summary of evidence on cerebellar involvement in AO (first section) and MI (second section) in pathological conditions.Abbreviations: AO: action observation; BM: biological motion; fMRI: functional magnetic resonance imaging; MI: motor imagery; MRI: magnetic resonance imaging; NS: not specified; PD: Parkinson's disease; PET: positron emission tomography; PICA: posterior-inferior cerebellar artery; SCA: Spinocerebellar ataxia; SPECT: Single Photon Emission Computed Tomography; TMS: transcranial magnetic stimulation.

Pathological condition Reference
Goal A. Antonioni et al. that the cerebellum plays a pivotal role in this regard (Dahms et al., 2020;Hermsdorf et al., 2020).Hence, given its critical involvement in motor and non-motor functions, the cerebellum is a key target for non-invasive brain stimulation (NIBS) techniques (Baroni et al., 2024).Consistently, research exploring the physiology of the cerebellar-thalamo-cortical pathway through TMS protocols indicates that stimulating the cerebellum activates Purkinje cells in the superior cerebellum.This activation leads to inhibition of the dentate nucleus, reducing its facilitatory drive on the contralateral motor cortex via the ventral lateral thalamus.Consequently, this process suppresses the contralateral motor area, known as cerebello-cerebral inhibition (CBI) (Fernandez et al., 2018;Manto et al., 2022;Ugawa et al., 1997Ugawa et al., , 1994)).Indeed, paired-pulse NIBS protocol involving a TMS pulse applied over the cerebellum (conditioning stimulus, CS), followed by another to the M1 after 5-8 ms (test stimulus, TS), results in a decrease in the amplitude of motor evoked potentials (MEPs) compared to a single pulse applied to M1 with the same intensity (Fernandez et al., 2018).This suggests a decline in the excitability of motor regions, likely resulting from the suppression of excitatory signals originating from the ventrolateral thalamus to neurons located in cortical layers I, III, V, and VI (Daskalakis et al., 2004).Importantly, thalamic modulation through cerebellar stimulation does not only act on M1 directly but also indirectly through an action on the premotor cortex, which is, in turn, closely connected with M1 (Spampinato et al., 2023).Likewise, transcranial electrical stimulation (tES) targeting the cerebellum modulates connections between the dentate nucleus and motor areas.For instance, a recent investigation has demonstrated that a specific type of cerebellar tES, transcranial alternating current stimulation (tACS) at theta frequency (5 Hz), enhances CBI during the stimulation period.This effect is likely due to the theta rhythm aligning with the natural resonance frequency of granule cells, resulting in an "entrainment" phenomenon that enhances the modulation of the cerebello-cortical connection pathway (Spampinato et al., 2021).
Importantly, cerebellar stimulation allows the modulation of complex interneuronal circuits within cerebral cortex regions linked to the cerebellum through thalamic relay pathways.Indeed, paired-pulse protocols provide insights into neurophysiological measures reflecting the functional status of excitatory and inhibitory cortical circuits, typically assessed at M1 due to its accessibility and the availability of neurophysiological markers for use as outcome measures.For instance, short-and long-interval intracortical inhibition (SICI and LICI) describes the activity of ionophoric GABA-A and metabotropic GABA-B currents, respectively, reflecting the activity of cortical inhibitory interneurons.In contrast, intracortical facilitation (ICF) outlines glutamatergic excitatory activity (Ferreri et al., 2011).These metrics, assessed through the amplitude of MEPs at M1, offer insight into the modulation of cortical interneurons shaping motor responses (Klomjai et al., 2015).Significantly, a recent form of rTMS, known as theta-burst stimulation (TBS), utilises a high-frequency protocol able to induce long-lasting changes through neuroplastic mechanisms and shows shifts in cortical excitability across both motor and non-motor regions (Huang et al., 2005;Rocchi et al., 2016).Crucially, a seminal work applied TBS on the lateral cerebellum and highlighted, by means of a continuous TBS (cTBS), a reduction in SICI and an increase in LICI over M1.In contrast, an intermittent TBS (iTBS) led to a decrement in LICI (Koch et al., 2008).These changes have been ascribed to the ability of this cerebellar NIBS protocol to modulate intracortical GABAergic circuits and underscore the cerebellum's relevance in regulating the circuitry, especially inhibitory pathways, of interconnected brain areas (Koch et al., 2008).Additionally, a recent investigation paired cerebellar TMS stimulation with electroencephalography (EEG) recordings to evaluate TMS-evoked potentials (TEPs), representing a gauge of overall cortical activation prompted by TMS across both M1 and parietal regions.This approach aimed to delve into the impact of cerebellar TBS on the contralateral hemisphere (Casula et al., 2016).Importantly, the focus was on the initial 200 ms components of TEPs, which rely on GABA-A activity within the first 15-60 ms post-TMS, while the subsequent phases are influenced by GABA-B modulation (Casula et al., 2014;Rogasch and Fitzgerald, 2013).By analysing TEPs, this study confirmed that cerebellar iTBS decreases LICI, whereas cTBS increases it, while cTBS also decreases SICI (Casula et al., 2016;Koch et al., 2008).Moreover, the authors also significantly underscored a TBS-induced modulation of M1 oscillatory activity.Specifically, they showed that cerebellar cTBS decreases TMS-evoked alpha (8-12 Hz) activity, while iTBS enhances the beta (13-20 Hz) activity (Casula et al., 2016).As posited by the authors, these changes in oscillatory patterns likely mirror the influence of this particular NIBS protocol on GABAergic interneurons, potentially at the thalamic or cortical levels or both.Therefore, it is reasonable to propose that an increase in GABA-mediated inhibition prompted by cTBS might have diminished the power of M1's intrinsic frequencies, whereas iTBS had the converse effect (Habas et al., 2019;McCormick et al., 2015).This emphasises not only the pivotal role of the cerebellum in modulating the activity of interconnected intra-cortical circuits, particularly inhibitory ones, but also its apparent involvement in regulating brain rhythms.
This holds particular significance as a specific EEG metric, namely the mu rhythm (8-13 Hz), was used to assess MNS activity from a neurophysiological standpoint.Indeed, the attenuation of the mu rhythm occurs both during AO and during action execution and is, indeed, regarded as a surrogate for MNS activation (Fox et al., 2016).Moreover, its suppression is most evident during the observation of a precision grip and when object interaction occurs, namely conditions characterised by a relevant and specific activation of the AON (Muthukumaraswamy and Johnson, 2004;Muthukumaraswamy et al., 2004).Of note, the mu rhythm is formed by the variable contribution of both alpha-and beta-band activity, and also the latter is suppressed by both action execution, MI and AO.Thus, both the alpha and the beta-band mu components appear to be modulated by MNS activity (Babiloni et al., 2002;Hobson and Bishop, 2016;Pozzo et al., 2017).Importantly, the two EEG frequency bands implicated in these processes align with those modulated by cerebellar NIBS (Casula et al., 2016).Moreover, crucially, a seminal study combined the spatial resolution of fMRI with the temporal precision of EEG, offering pivotal evidence that, during AO tasks, mu rhythm modulation correlated negatively with the blood oxygenation level-dependent (BOLD, reflective of neural activity) signal in the MNS areas, prominently involving the bilateral cerebellum (Braadbaart et al., 2013).Taken together, this evidence not only reinforces the cerebellar involvement in the MNS but, more importantly, suggests a potential role in modulating the cortical activity within this functional system.
Further support for this functional connection stems from neurophysiological investigations.Specifically, an analysis of various regions within the AON in a monkey study revealed insights into the temporal sequencing of motor events during AO.The discharge patterns of neurons with mirror properties, especially in the area F5 and, to a lesser extent, in the pre-supplementary area (F6), were found to align with profiles typically associated with inhibitory interneurons (Ferroni et al., 2021;Gold et al., 2006;González-Burgos et al., 2005).Importantly, also human studies have shown that AO influences the activity of the intra-cortical inhibitory circuitry of M1, notably by modulating SICI.Indeed, extensive research has demonstrated that AO can downregulate SICI in various contexts, including joint actions and during the observation of errors in others' actions.Thus, GABA activity serves as a marker for the inhibition necessary for rapid motor adaptation and the regulation of motor functions via both fast and slow modulatory signaling pathways, thereby amplifying the adaptability of motor behaviours (Cardellicchio et al., 2020(Cardellicchio et al., , 2018;;Neubert et al., 2011;Patuzzo et al., 2003;Tritsch et al., 2016).Moreover, the so-called "beta-rebound" in EEG activity, occurring subsequent to both AO and execution, is consistently considered as reflecting an active inhibition of motor activity (Jurkiewicz et al., 2006;Koelewijn et al., 2008;Pfurtscheller et al., 2005).Notably, a recent study confirmed the modulation of SICI in the context of AO and demonstrated its potential utility in predicting AO-induced behavioural improvement (Nuara et al., 2023).Indeed, it is widely acknowledged that intracortical inhibitory circuits are crucial for achieving optimal movement speed and accuracy since they play a pivotal role in selecting precise neural representations required for executing the desired movements while preventing undesired motor outcomes (Beck and Hallett, 2011;Greenhouse, 2022).Hence, the balance between cortical excitation and inhibition is paramount for executing precise movements, particularly in visuomotor tasks where the cerebellum plays a pivotal role (Ebbesen and Brecht, 2017;Giboin et al., 2021;Spampinato and Celnik, 2021).Furthermore, the thalamus plays a critical role in shaping the excitatory and inhibitory activity of the cortex, with the cerebellum emerging as one of the main modulators of thalamic activity (Habas et al., 2019;Taub et al., 2013).Therefore, it is possible that, after an initial phase of increased corticospinal excitation produced by the AO due to the sensory-motor resonance mechanism already discussed, the subsequent modulation of inhibitory activity (as assessed by the modification of SICI) selects the most useful motor representations to shape a "meaningful behavioural outcomes" and, indeed, the greater the inhibitory modulation, the better the subsequent performance in the same observed behaviour (Byblow and Stinear, 2006;Giboin et al., 2021;Hannah and Aron, 2021;Naish et al., 2014;Nuara et al., 2023).
However, a role for GABA-B has also been highlighted in this context, evidenced by LICI reduction when the observed action does not correspond to the concurrent executed action (Cardellicchio et al., 2020).Of note, this evidence could also provide some insight into the origin of the MNs found in the AON.In fact, some authors claim that MNs acquire their ability to match observed actions through a typical process of sensorimotor learning since they initially act as simple motor neurons during action execution and, later, undergo associative learning through repeated experiences of observing and performing the same actions (Heyes and Catmur, 2022).Thus, reasonably, through this correlated experience, motor neurons establish robust connections with perceptual (e.g.visual) neurons specialised in similar actions.However, other researchers have recognised the importance of visuomotor learning in the context of MNS, but they have suggested a genetic predisposition for their formation or an innate nature that can be traced back to motor experiences from birth or even during gestation (Del Giudice et al., 2009;Gallese et al., 2011Gallese et al., , 2009)).Although a genetic predisposition for MNs, given their crucial role in many fundamental activities that characterise evolutionarily higher-order organisms, seems plausible, a great deal of evidence has not only shown that sensorimotor learning is crucial to achieve their formation, but has also revealed that 'counter-mirror' training can reverse the mirror response and that mirror and counter-mirror responses follow the same time course (Brunsdon et al., 2020;Catmur et al., 2011;Cavallo et al., 2014;Guidali et al., 2020;Zazio et al., 2019).Consistently, in 7-month-old infants, sensorimotor mu rhythm suppression was heightened when they had received more sensorimotor experience in previous training sessions, despite their inability to walk at that age (de Klerk et al., 2015).Similarly, children born with a severe bilateral cataract, surgically treated only after several years, show a lower propensity to imitate the observed behaviour, which supports the hypothesis that the sensorimotor matching is crucial in shaping and modulating MNS activities (McKyton et al., 2018).
Based on this evidence, although some motor neurons might have a predisposition to develop mirror properties, the sensory-motor matching provided by visual-perceptual learning could be crucial and, reasonably, the cerebellum is instrumental in sculpting these circuits.Indeed, in the early stages of motor learning, the cerebellum displays greater activity, likely driven by significant disparities between anticipated and actual outcomes, but also identifies, via its connections with neocortical regions implicated in motor planning and execution, the most effective patterns to minimize such discrepancies (Penhune and Doyon, 2005;Spampinato and Celnik, 2021).Given, as already discussed, the substantial activation observed in the cerebellum during AO, mirroring that observed during actual execution, it is highly plausible that a similar mechanism is at play in shaping observational learning.Indeed, evidence from animal models has shown that a cerebellar lesion prevents learning by observation but not the use of already established memories or the performance of exploratory motor behaviour (Graziano et al., 2002;Molinari et al., 1997;Petrosini et al., 2003).Of note, Koch et al. showed, using a cerebellar TBS protocol, that it can modulate the characteristics of learning in a visuomotor adaptation task.This modulation, dependent on whether it is facilitatory (iTBS) or inhibitory (cTBS), correlates with alterations in activity across interconnected sensorimotor regions, likely attributable to shifts in the excitation/inhibition balance (Koch et al., 2020).Furthermore, given the crucial modulation of GABA-A interneurons (measured via SICI) by AO, and the predictive nature of this modulation on subsequent performance during executed behaviour, alongside the demonstrated significant impact of cerebellar NIBS techniques on intra-cortical GABAergic circuits, it seems reasonable to hypothesise that the cerebellum modulates, throughout motor learning mechanisms, the activity of intra-cortical inhibitory interneurons, presumably the same ones influenced by AO and known to exhibit mirror properties (Cardellicchio et al., 2020;Casula et al., 2016;Ferroni et al., 2021;Koch et al., 2008;Nuara et al., 2023).This would also justify the different pattern of activity between the cerebellum, which is more active in the early phases of learning to sculpt the visuomotor association in the intra-cortical circuitry, and the neocortical areas, which are more active in the later phases, reasonably because the sensory-motor matching process has been optimised and the motor circuits are able to select the most suitable activity patterns without a cerebellar intervention (Spampinato and Celnik, 2021).Moreover, some evidence has suggested that anodal tDCS applied to the lateral cerebellum during the learning phase of a new temporal motor skill may also facilitate its retention due to a priming effect, likely by enhancing the consolidation of new information once training is completed (Wessel et al., 2016).This is particularly intriguing when considering that the cerebellum not only plays a role in motor learning associated with AO but also in learning that integrates auditory information to guide motor actions (i.e., audio-motor learning, such as in musical practice) (Kohler et al., 2023).These findings suggest the potential of cerebellar NIBS to improve motor learning in a cross-modal manner, potentially overcoming specific deficits in one sensory modality by leveraging others to achieve improvement, which is a relevant aspect for tailoring therapeutic interventions according to the peculiar impairments of each patient.
Thus, although predictive coding hypotheses suggest that the motor system computes the difference between expected and observed information, this evidence seems to indicate that, depending on the degree of correspondence between the predicted and the achieved outcome, the involvement of specific areas of the AON is different, and this also has implications from a clinical point of view (Kilner et al., 2007;Urgen and Miller, 2015).Of note, this interpretation aligns with findings observed post-stroke.Indeed, a randomised clinical trial conducted on stroke patients demonstrated that a combined intervention of cerebellar iTBS and physiotherapy resulted in greater improvements in balance and gait compared to sham-iTBS and physiotherapy alone.This suggests that neuroplastic changes in the cerebellum may influence the functioning of the fronto-parietal network through modulation of intra-cortical GABAergic circuits via dentate-thalamus connections (Koch et al., 2019(Koch et al., , 2008)).Consistently, after stroke, patients often exhibit deficits linked to alterations in excitatory and inhibitory circuits within the motor network, which are crucial for motor control and the generation of voluntary movements (Hummel et al., 2009).Specifically, it is well-established that SICI decreases just prior to movement initiation, increases just before movement termination, and subsequently decreases again, returning to baseline levels (Zaaroor et al., 2003).Importantly, a number of studies has shown that, in the acute and subacute stroke timeframes, resting SICI is reduced in the damaged M1 (Bütefisch et al., 2008;Cicinelli et al., 2003;Liepert et al., 2000).
A. Antonioni et al.Interestingly, one study documented that even LICI was abnormally reduced in the affected hemisphere in the acute phase and it was the only intra-cortical neurophysiological index correlating with functional status and predicting recovery at 3 months (Swayne et al., 2008).Furthermore, a study on chronic stroke patients (after 6 months from the event onset) showed not only a persistent reduction in resting SICI in those with a good recovery but also its abnormal persistence during paretic hand movement preparation (Hummel et al., 2009).Notably, this decline in the function of inhibitory intracortical circuits, coupled with typically normal intracortical facilitation (ICF), results in heightened excitability in M1, which is deemed a crucial mechanism for enhancing functional recovery, likely influenced by modulation from other areas within both the ipsilateral and contralateral motor networks (Bäumer et al., 2006;Cicinelli et al., 2003;Manganotti et al., 2002;Murase et al., 2004;Swayne et al., 2008;Talelli et al., 2006).Consistently, several studies revealed that a decreased inhibitory activity is crucial for creating a permissive environment for cortical reorganisation and the establishment of new cortical boundaries (Jacobs and Donoghue, 1991).Moreover, following a stroke, the observed disinhibition at rest in affected M1 might serve as a compensatory mechanism for the impaired ability to execute voluntary movements effectively.However, it could also impose constraints on the extent of pre-movement modulation in the affected hand (Hummel et al., 2009).Indeed, although the evidence is not always consistent, the most accredited models support this interpretation since numerous EEG studies have demonstrated that, when the injured hemisphere is able to reorganise its activity through diverse compensatory mechanisms, the prognosis tends to be more favorable.Conversely, if the healthy hemisphere needs to compensate, likely due to severe damage in the affected one, the recovery outcomes are markedly less satisfactory (Antonioni et al., 2023;Di Pino et al., 2014;Milani et al., 2022).These findings gain further support in a study revealing that intra-cortical inhibition undergoes an abnormal reduction in the acute stroke phase in both the injured and healthy hemispheres.Interestingly, one month post-stroke, restoration of intra-cortical inhibition occurs in the unaffected hemisphere among patients exhibiting good motor recovery whereas, in cases of poor recovery, this abnormality persists in the unaffected hemisphere even at this timeframe (Manganotti et al., 2002).Significantly, longitudinal tracking of the same cohort over several months revealed a consistent pattern (Manganotti et al., 2008).This supports the notion that hemispheric disinhibition mirrors its reorganisation and that the persistence of the involvement of the healthy hemisphere indicates a failure of recovery in the injured one, thereby suggesting a poorer prognosis.However, an alternative hypothesis posits that the reorganisation of the healthy hemisphere relies on neuroplastic mechanisms associated with heightened usage of the unaffected limb.Consequently, severe patients lacking functional capacity in the affected limb might exhibit a use-dependent modulation of the intra-cortical inhibitory circuits in the unaffected hemisphere (Liepert et al., 1998;Shimizu et al., 2002).
Furthermore, it is worth mentioning that alterations in SICI have been documented in many disorders (e.g.schizophrenia) characterised by altered MNS functioning, whereas LICI alterations have been reported in cerebellar degeneration, highlighting a correlation between the cerebellum, GABAergic intra-cortical circuits, and MNS even in pathological contexts (Chen, 2004;Wessel et al., 1996).However, crucially, these findings are supported by studies in stroke patients indicating that EEG desynchronisation during AO follows a similar pattern since it occurs in the affected hemisphere among individuals with milder impairments and in the unaffected one among those with more severe deficits.Additionally, it has been demonstrated that the outcome of AOT can be predicted by mu suppression, which serves as a measure of MNS integrity (Antonioni et al., 2024a;Boni et al., 2023).Hence, we suggest that cerebellar NIBS enhances outcomes in stroke patients by modulating MNS activity.Through the regulation of intra-cortical GABAergic circuits, it enables the adjustment of the excitation/inhibition balance, which is essential for effective motor function.This modulation likely targets inhibitory interneurons with mirror properties, facilitating the restoration of optimal perceptual-motor associations necessary for achieving intended outcomes (Caligiore et al., 2013;Ferroni et al., 2021;Koch et al., 2019Koch et al., , 2008)).Indeed, it is likely that, when the lesioned hemisphere retains activity in its MNs, such as mu suppression during AO, it possesses sufficient elements with mirror properties.These elements could be subject to modulation by cerebellar influence, facilitating the recruitment of optimal representations to minimize discrepancies between predicted and realised outcomes.Consistently, cerebellar activity correlates with motor recovery post-stroke, underlining its significance in this context (Luft et al., 2005).Indeed, even if inhibitory cortical interneurons are fewer in number compared to excitatory neurons, their impact is disproportionately significant, likely owing to the strategic positioning of their synapses (Kätzel et al., 2011).In summary, we propose that cerebellar NIBS, by modulating the circuits of intra-cortical inhibitory interneurons with mirror properties, restores the correct MNS response observed during both AO and action execution, leading to better motor performance.Consequently, we speculate that cerebellar NIBS might enhance mu suppression during AO in stroke patients with diminished responses by shaping novel visuomotor associations crucial for neuronal activity within this network.
And what about the cerebellar contribution during MI? Research indicates that, similar to AO, MI elicits heightened corticospinal excitability and, importantly, diminishes SICI akin to the pattern observed during mild voluntary contraction.This implies that while both conditions similarly activate the motor programme, its peripheral execution is impeded during MI (Abbruzzese et al., 1999;Rossini et al., 1999).However, in this case, there are no external stimuli to activate the corresponding motor representations, and it is reasonable to assume that subjects use their own memories to activate the AON and mentally simulate the action.Consistently, we suggest that the fundamental component for MI involves neurons with mirror properties within the hippocampal circuitry -although, to the best of our knowledge, no single-cell recording studies are currently available to prove this correlation, thus this interpretation remains speculative as it is based on indirect evidence.Indeed, a seminal study performed extracellular recordings of individual neurons in frontal and temporal regions among patients with drug-resistant epilepsy and demonstrated the existence of MNs in areas including the hippocampus, parahippocampal gyrus, and entorhinal cortex.Importantly, these neurons exhibited responses to hand-grasping actions, facial emotional expressions, and associated acoustic signals (Mukamel et al., 2010).Of note, recent research has revealed the close interconnection between the cerebellum and the hippocampus, suggesting their pivotal role in spatial navigation, which was traditionally associated solely to the hippocampus (Lefort et al., 2015;Moser et al., 2017).For example, the dentate nucleus is implicated in the acquisition phase of learning the Morris water maze task, particularly in mastering the navigation toward a hidden platform.However, its involvement does not extend to the retention of spatial memory (Locke et al., 2018).Similarly, a self-motion-based motor sequence is not correctly acquired following damage to the dentate nucleus (Gaytán-Tocavén and Olvera-Cortés, 2004).In particular, during a navigation task, activation has been observed in cerebellar crus I, along with the contralateral hippocampus and either the medial PFC or medial parietal cortex, underscoring its significance in the cognitive facets of spatial navigation (Rondi-Reig et al., 2022).Consistently, functional connectivity between the hippocampus and specific cerebellar regions, i.e.Crus I, lobules VI, IV/V, IX, and X, plays a crucial role in both sequence learning and execution (Babayan et al., 2017).This connectivity pattern, observed during sequence-based navigation, mirrors findings in early motor sequence learning and accurate prediction of finger movements (Doyon et al., 2003;Onuki et al., 2015).Additionally, interactions between the hippocampus and the cerebellum facilitate not only precise movement prediction but also the perception of visual-spatial changes (Hauser et al., 2019).Taken together, these A. Antonioni et al. findings suggest that the hippocampus and cerebellum cooperate in tasks requiring integration of spatial and temporal dimensions (Rondi-Reig et al., 2022;Yu and Krook-Magnuson, 2015).
Furthermore, crucially, compelling evidence suggests that the hippocampus assumes a pivotal role in navigation tasks by encoding relationships among diverse elements, regardless of their nature, e.g.spatial, sensory, or motor, a process akin to episodic memory (Eichenbaum and Cohen, 2014).Moreover, research indicates that the cerebellum contributes to integrating sensory inputs that impact the activity of hippocampal cells (Lefort et al., 2015;Rochefort et al., 2011).It is intriguing that communication between the cerebellum and hippocampal structures occurs within the theta rhythm (5-7 Hz), which aligns with the intrinsic resonance of cerebellar granule cells.Indeed, this rhythm also appears to significantly influence the connectivity between the cerebellum and motor areas, raising the hypothesis that NIBS techniques employing this frequency may facilitate interactions with the hippocampus, potentially yielding significant implications for the observed positive outcomes (De Zeeuw et al., 2008;Hoffmann et al., 2015;Koch et al., 2020;Spampinato et al., 2021).It is plausible to speculate that these connections between the cerebellum and hippocampus could be relevant in the context of MNS since subjects might leverage them as a valuable short-term memory repository for prediction and postdiction concerning action outcomes (Friston et al., 2016;Friston and Buzsáki, 2016).Indeed, memory is no longer regarded as a static information storage but rather as a dynamic process that facilitates the recollection of actions performed and the associated sensations.From this perspective, MNS could hold significance both for the encoding of memories (e.g.associating words with gestures that activate the corresponding motor representation) but also for retrieving, as recalling actions aligned with one's motor repertoire (overlapping with those already performed or observed) appears more effortless than unfamiliar ones (Dijkstra et al., 2007;Macedonia, 2014;Yee et al., 2013).In line with Rondi-Reig et al.'s proposition, it also appears that the hippocampus might provide contextual cues, such as recalling past actions, to the cerebellum, which could execute a learning mechanism based on model-free reinforcement learning driven by sensory predictions (Babayan et al., 2017;Rondi-Reig et al., 2022).This notion finds support in the pioneering work of Buzsáki, positing that the CNS gives meaning to actions by probing their sensory outcomes and encoding the resulting sensory-motor loop through a "chunking" process, namely by combining already existing neuron sequences (linked to familiar aspects of the experience) to novel activity combinations (Buzsáki, 2019).Crucially, in this context, the pivotal role of the hippocampus emerges, with the cerebellum potentially accessing these experiential segments, reasonably guided by sensory cues, suggesting it might have the ability to combine them to explore the best pattern to achieve the outcome, ultimately minimising the disparity between predicted and actual results (Buzsáki and Tingley, 2018;Friston et al., 2016;Friston and Buzsáki, 2016).Therefore, drawing from this evidence, we propose that the MNs within the hippocampal circuitry provide the cerebellum with the mnestic fragments necessary to implement the mental representation of the motor programme during MI and that the cerebellum, in evaluating the incongruity between the anticipated and realised sensory outcomes in internal simulation, due to its close connection with fronto-parietal areas, implements adjustments by trying new combinations of hippocampal-encoded sequences.This mechanism may underscore the functional connectivity changes, enhancements in performance, and notably, the generalization of these enhancements, as the newly encoded segments from cerebellar simulation for optimizing performance become accessible for other tasks (during MI, AO, and execution because the AON shares these representations), solely through mental practice (Gentili et al., 2006;Ladda et al., 2021;Marins et al., 2019).
Moreover, of note, since the cerebellum-hippocampal connections appear to be fundamental for voluntarily recalling previous experiences but not for improving motor performance, this could justify why severe hippocampal lesions (e.g., the notorious H.M. patient) preclude voluntary recall of previous motor and non-motor experiences but do not obtaining the benefits associated with repeated practice, which instead depend more on the cortico-striatal-cerebellar circuitry (Behrendt, 2013;Krakauer and Shadmehr, 2006;Squire, 2009).Indeed, if memory traverses the realm of recorded experiences, then MI and predictive coding could be likened to exploration within the kingdom of potentialities.In both domains, the hippocampus and cerebellum appear pivotal.
Finally, additional support for this interpretation also arises from another pathological model, namely phantom limb pain (PLP), a condition in which the patient has undergone a traumatic or surgical amputation of a limb yet persists in experiencing pain from the territory of the removed body region.This phenomenon potentially stems from the brain's endeavor to retain the cortical territory associated with that body region, even in the absence of peripheral input, and such maladaptive plasticity often engenders severe and intractable pain (Antonioni et al., 2024b;Nikolajsen and Christensen, 2015).Once more, several studies have intriguingly revealed an alteration in the activity of the intra-cortical inhibitory circuit, akin to mechanisms previously discussed in stroke cases, which may facilitate cortical reorganisation (Candido Santos et al., 2020).Additionally, there appears to be an alteration in the circuitry of the MNS, as the internal simulation of actions involving the absent limb fails to align with reality, and this discrepancy between expected and actual outcomes might potentially give rise to severe pain in affected individuals (Antonioni et al., 2024b;Chen et al., 2013;Collins et al., 2018;Schwenkreis et al., 2000;Teixeira et al., 2021).This could also explain why many effective therapies for PLP, e.g.box-mirror therapy and AOT, are based on MNS activation: it is conceivable, indeed, that AO and MI activate the sensory-motor representations of the loop, which the subject, naturally, can no longer effectively utilise and, by mitigating this discrepancy, it may inhibit maladaptive plasticity and the ensuing pain (Beaumont et al., 2011;Kuffler, 2018;Suso-Martí et al., 2020).Importantly, several studies have documented significant cerebellar activation in the context of pain perception, likely linked to the processing of sensory information and the regulation of frequently provoked motor responses (Coombes and Misra, 2016;Moulton et al., 2010;Peyron et al., 2000).Consistently, recent research has shown that a cerebellar NIBS protocol can reduce pain symptoms in PLP patients (Bocci et al., 2019).Indeed, although the evidence is not yet conclusive, the application of cerebellar NIBS techniques appears to be a promising strategy for pain modulation, both from the perceptual, motor-response preparation, and emotional-cognitive perspectives (Manto et al., 2022;Schmahmann, 2019).We speculate that this phenomenon occurs not solely due to the restoration of the excitation/inhibition balance in interconnected cortical areas, reducing maladaptive plasticity and enhancing the critical functioning of the MNS to alleviate pain, but also because cerebellar stimulation is likely to facilitate the updating of the internal model, thereby reducing the discrepancy between predicted and actual outcomes and, reasonably, correcting the sensory-motor representation of the lost effector in the vast network of the AON.

Neuroanatomical and neurophysiological underpinnings of cerebellar modulation on the MNS
Notably, while the cerebellar output is solely inhibitory to the DCNs, provided by the axons of Purkinje cells, the other cerebellar cells can exert a complex influence that may significantly modulate the overall response (D'Angelo, 2018).For instance, cerebellar inhibitory interneurons targeting Purkinje cells can modify their activity to enhance or reduce the inhibitory signal directed to the DCNs.Interestingly, a great deal of research highlighted distinct types of long-term neuroplastic changes within the cerebellar cortex, both in terms of potentiation (Long-Term Potentiation, LTP) and depression (Long-Term Depression, LTD) (D'Angelo, 2014).Of note, these changes depend on different molecular mechanisms and the specific cell types involved, and they have been linked to behavioral changes associated with learning (Cheron et al., 2016).This suggests that performance improvement derives from the possibility of finely modulating and adapting cerebellar computations at the molecular and cellular level.
It is also important to note that these LTP/D effects are considered the neurobiological correlates of modulations induced by NIBS techniques (Cirillo et al., 2017).Indeed, neuromodulation can induce long-term changes in neurophysiological measures and in various behavioral outcomes by acting on different molecular mechanisms, such as calcium signaling or neurotransmitter receptor activity (Huang et al., 2017).This is particularly relevant in the cerebellum, as LTP/D effects have been shown to underpin various forms of learning, and the Fig. 3. Schematic representation of the hypothetical model of the connections between the cerebellum and cortical/subcortical areas with inhibitory interneurons displaying mirror properties.On the left, an example of a cerebellar module is depicted (refer to Fig. 1 for more details).The deep cerebellar nuclei project primarily to the thalamus (on the top), which in turn projects mainly to the primary motor area (M1) (on the right).This influence affects the output (i.e., the pyramidal neurons located in the cortical layer V) not only directly but also through excitatory (orange cells) and inhibitory cortical (blue cells) interneurons.Other inhibitory interneurons (grey cells located in cortical layer VI) provide a modulation on projections from the thalamus.Some of these interneurons reasonably display mirror properties (highlighted by a dashed red circle) and are responsible for the cerebellar influence on measures related to the Mirror Neuron System (MNS) activities, such as electroencephalographic measures (e.g., mu rhythm) and non-invasive brain stimulation techniques (e.g., short and long interval intracortical inhibition, SICI and LICI).These inhibitory interneurons are primarily found in cortical layers II and III in M1, where they can modulate the excitability of pyramidal neurons (and consequently the motor output) both directly and through a complex regulation of the excitation-inhibition balance involving other interneuronal circuits across various cortical layers.Notably, interneurons with similar properties and subject to cerebellar influence also appear to be present in prefrontal and premotor regions, as well as in associative parieto-temporal areas (on the right), justifying the cerebellum's participation and in various aspects of MNS regulation.During the initial phases of learning, the cerebellum's influence on these interneurons may contribute to shaping the circuits by creating new visuo-motor (or audio-motor, depending on the specific areas) associations to optimize performance.When the discrepancy between predicted and actual outcomes is minimized, cerebellar influence becomes less relevant, and its activity diminishes.These interactions likely explain the performance improvements obtained not only by executing a task but also by merely observing it or listening to related significant stimuli.Furthermore, the cerebellum engages in bidirectional communication with the hippocampus (in the centre), both directly and through the thalamus and, to a lesser extent, the Median Raphe nucleus, a serotonergic brainsteam structure, which project to the medial septum and the nucleus of Broca's diagonal band, which provide access to the hippocampus.This allows the cerebellum to retrieve mnemonic traces during motor imagery (MI) to test the hypothetical sensory consequences of motor action (also using available information from cortical networks, which provide essential data on planning, strategy, multi-sensory associations, and a priori hypotheses from already established models), thereby refining interneuronal patterns to minimize discrepancies between predicted and expected outcomes.Conversely, once the models are optimized, the cerebellum strengthens hippocampal circuits that retain the optimized patterns, reasonably explaining the performance benefits associated with MI.Notably, these new internal models are available for other activities in which the MNS is involved, such as execution and AO, reasonably justifying the similar benefits derived from these different activities.Axons terminating bifidally indicate excitatory synapses, while those ending with a straight line indicate inhibitory ones.Based on (Bonini et al., 2022;Caligiore et al., 2017b;Casula et al., 2016;Daskalakis et al., 2004;Dum and Strick, 2003;Koch et al., 2008;Rolls, 2023;Rondi-Reig et al., 2022;Spampinato et al., 2023;Zeidler et al., 2020).Abbreviations: AO: action observation; LICI: long interval intracortical inhibition MI: motor imagery; MNS: Mirror Neuron System; M1: primary motor cortex; SICI: short interval intracortical inhibition.numerous involved cell types allow for extremely complex and refined modulation (D'Angelo, 2014).
For example, although it is not associated with long-term neuroplastic changes, tACS is able to influence specific cell types through a frequency-specific "entrainment" effect, which is thought to depend on the intrinsic resonance frequency (i.e. the characteristic activity frequency) of the targeted cell (Elyamany et al., 2021).Consistently, several studies suggested that theta-tACS predominantly affects granule cells, while gamma-tACS (30-80 Hz) mainly influences Purkinje cells, with effects due to, reasonably, modulation of the computational activity performed by the specifically targeted cell type (Miyaguchi et al., 2018;Spampinato et al., 2021).Additionally, TBS involves triplets of pulses in the gamma-band interspersed at the theta rhythm, offering the potential for even more extensive and complex network effects, hypothetically reflecting the interaction between different cell types (Koch et al., 2008).This is crucial, as the cerebellum exhibits one of the broadest range of activity frequency in the CNS, from slower rhythms, such as delta (< 4 Hz), likely attributable to the inferior olivary nucleus and its climbing fibers, and theta, likely associated with granule cell activity, to faster ones, like beta, associated with Golgi cell activity, gamma, hypothetically linked to the network of Purkinje cells and molecular layer inhibitory interneurons, and even ultra-rapid frequencies, whose origin is not yet fully understood (De Zeeuw et al., 2008).Therefore, cerebellar NIBS techniques might be fundamental for modulating cerebellar output and, consequently, its complex influence on numerous targets within the CNS.
At the brain level, as shown in Fig. 3, the DCNs modulate various target areas through a thalamic relay, primarily involving the complex fronto-parietal network, but also temporal regions (Baumann et al., 2015;Casula et al., 2016;Dum and Strick, 2003;Mukamel et al., 2010).
Specifically, in M1, the thalamic projections modulated by the cerebellum influence not only the pyramidal neurons of layer V (i.e. the output of the motor cortex which forms the descending motor pathways) but also, notably, inhibitory interneurons, mainly located in cortical layers II and III (Opie et al., 2022;Rolls, 2023;Spampinato, 2020).Based on the discussed evidence, these interneurons could be the neurobiological substrate of EEG (i.e., mu rhythm) and NIBS (i.e., SICI/LICI) measures whose activity has been associated with the MNS and, therefore, justify the cerebellum's crucial modulation of this network (Cardellicchio et al., 2020;Casula et al., 2016;Farina et al., 2020;Ferroni et al., 2021;Nuara et al., 2023).Notably, the position of these inhibitory interneurons with mirror properties (dotted red circles in Fig. 3) in M1 is critical, as it allows them not only to inhibit pyramidal neurons directly but also to finely modulate excitatory and inhibitory interneuronal circuits distributed across the different cortical layers (Chen, 2004;Yi et al., 2014).Indeed, these circuits likely enable increasingly accurate and precise motor outcomes by balancing excitation and inhibition to select the best motor patterns to achieve the desired motor outcome (Cardellicchio et al., 2018;Nuara et al., 2023).It is reasonable to hypothesize that cerebellar influence on these circuits is greater when there is a high disparity between predicted and actual outcomes, as seen in the early stages of learning.Conversely, as practice (whether actual or related to AO and MI) reduces this disparity, reflecting a proper shaping of visuomotor relationships, cerebellar activity diminishes, likely because the efference copy and peripheral signals do not report errors in internal models (Spampinato and Celnik, 2021).
However, as previously discussed, inhibitory interneurons displaying mirror properties have also been found in prefrontal, premotor, parietal, and to a lesser extent, temporal regions (Bonini et al., 2022).All these areas are strictly interconnected with M1, allowing for further modulation of its activity (Ferrer and De Marco García, 2022;Kinnischtzke et al., 2014;Rolls et al., 2023).Indeed, inhibitory interneurons with mirror properties in prefrontal and premotor areas might affect higher-order aspects of behavioral strategies and motor planning, where cerebellar modulation could play a critical role in selecting optimal activities based on specific goals (Balsters et al., 2010;Schmahmann, 2019;Tzvi et al., 2020).In contrast, cerebellar modulation of temporo-parietal areas, crucial for multisensory integration, might contribute by fine-tuning the action of inhibitory interneurons showing mirror properties that influence motor regions (Jakobs et al., 2012;Kohler et al., 2023).This process might leverage various sensory channels (i.e.visual and auditory), allowing learning not only through execution but also through AO or listening to meaningful stimuli, which represent a perspective with important clinical implications.
Crucially, the cerebellum is bidirectionally connected also with the hippocampus, both directly and, more importantly, indirectly, via other thalamic nuclei and brainstem structures, such as the Median Raphe nucleus (Rondi-Reig et al., 2022).Animal model studies have demonstrated that stimulation of this serotonergic brainstem nucleus leads to a motor inhibition associated with strong suppression of hippocampal theta rhythm (Bland et al., 2016).Considering the role of the cerebellum in motor inhibition and its extensive anatomic-functional connections (likely in the theta band) with the hippocampus, it is possible that these effects are also dependent on an interplay between the cerebellum and hippocampus mediated by this serotonergic system (Beliveau et al., 2015;Hoffmann and Berry, 2009;Peterburs and Desmond, 2016;Rondi-Reig et al., 2022).Both these thalamic nuclei (specifically, the supramammillary nucleus and nucleus incertus) and the Median Raphe nucleus access the hippocampus via the medial septum/diagonal band of Broca, a region that modulates various neurotransmitter systems, most crucially the GABAergic one (Damborsky and Yakel, 2021;Takeuchi et al., 2021).Therefore, it could represent a key gateway for the cerebellum to modulate hippocampal inhibitory interneurons displaying mirror properties (Kang et al., 2021).Reasonably, the hippocampus provides the cerebellum with memory traces to access context-dependent internal models (Mukamel et al., 2010;Zeidler et al., 2020).Furthermore, these connections might be crucial for enabling the cerebellum to test hypothetical sensory consequences and generate internal models in new contexts.This process is shaped by simulating actions within the AON and utilizing multisensory and strategic information from previous experiences stored in interconnected cortical networks (Caligiore et al., 2013;Yu and Krook-Magnuson, 2015).Since the hippocampus might provide memory traces to the cerebellum independently of available sensory stimuli, it is plausible to hypothesize that the refinement of internal models could occur solely through MI.Finally, since these connections are bidirectional, the cerebellum might strengthen hippocampal patterns matching to models minimizing the discrepancy between predicted and actual outcomes, likely modulating the activity of the same inhibitory interneurons that facilitate the recall of stored models during memory retrieval (Rondi-Reig et al., 2022).These patterns then become available for future use but might still be susceptible to modification if MI leads to the hypothesizing of different scenarios.
While these interactions are effective and optimized in healthy subjects, CNS lesions may impair these complex activities.For instance, direct cerebellar damage might impair the shaping of visuo-motor associations due to the loss of influence on cortical inhibitory interneurons, hindering learning and, consequently, performance optimization (Petrosini et al., 2003).Conversely, lesions in different brain areas could disrupt the already discussed specific contributions to this complex computational model (Bonini et al., 2022).Importantly, in this context, NIBS techniques might be harnessed to promote recovery.Indeed, cerebellar neuromodulation techniques offer the opportunity to modify numerous aspects of cerebellar computation.This includes modulating membrane excitability in either an excitatory or inhibitory manner, such as through tDCS, or promoting the activation and synchronization of specific cell types using tACS by selecting the relevant frequencies (Ferrucci et al., 2015;Wessel et al., 2023).Moreover, other techniques, such as iTBS or cTBS, can induce long-lasting neuroplastic changes, like LTP/D effects, in various regions (depending on the application site) and in different cell types, based on the technique A. Antonioni et al. employed (Koch et al., 2008).All these modulations might help shift the cerebellar network effect, aiding in the recovery of compromised activities.For example, in pathologies characterized by abnormal cortical excitability, the inhibitory effect of the cerebellum on connected areas can be enhanced, and conversely in conditions where cortical excitability is reduced (Manto et al., 2022).Indeed, as previously discussed, the modulation of this balance between excitation and inhibition is crucial for performance optimization and this is likely attributable to the specific cerebellar influence on cortical inhibitory interneurons with mirror properties, potentially creating new visuomotor associations that enable the recovery of lost functions using available resources despite the damage (Vattikonda et al., 2016).Additionally, protocols combining cerebellar and cerebral neuromodulation might offer a crucial tool to bypass disrupted anatomical connections.Indeed, by leveraging rhythm-based coordination, these protocols could enable the cerebellum to create new visuo-motor associations to improve performance (Miyaguchi et al., 2018).Specifically, the gamma rhythm is thought to be essential for coordinating activities between spatially distant neuronal populations, and studies employing this rhythm have already demonstrated improved cerebellar-motor interactions and behavioral outcomes (Buzsáki and Schomburg, 2015;Miyaguchi et al., 2020).Crucially, it has been suggested that gamma rhythm arises from the activity of inhibitory interneurons, which enable fine regulation within functionally integrated circuits (Cardin, 2018).Therefore, it seems reasonable that protocols combining gamma-band cerebellar and brain stimulation modulate, on the one hand, the network formed by the Purkinje cells and the interneurons in the cerebellar molecular layer (De Zeeuw et al., 2008).On the other hand, they might influence the circuitry of cortical inhibitory interneurons, promoting the shaping of sensorimotor associations that are essential for learning and, notably, fundamental to the functioning of the MNS (Bonini et al., 2022;Miyaguchi et al., 2020;Nowak et al., 2018).Given the number of cell types involved in cerebellar computations, further studies will be fundamental to deepen their different contributions to these complex interactions.
Furthermore, despite the topographical organization of the cerebellum reflects the cerebellar-cortical loops from an input and output perspective, the horizontal extension of Purkinje cell dendritic arborizations and the collateral branches of mossy and climbing fibers allow communication between non-adjacent modules (Apps et al., 2018;Shinoda and Sugihara, 2022).This could be a useful tool for neuromodulating healthy areas to bypass lesioned sites because of various local and remote pathologies.Lastly, although our discussion has primarily focused on the relationship between the cerebellum, cortical networks, and the hippocampus, cerebellar output can also modulate the spinal cord and basal ganglia, providing additional benefits and likely exerting a highly complex influence on the MNS (Arber and Costa, 2022;Caligiore et al., 2017b).Further studies and future research will be needed to delve deeper into these influences.

Future clinical perspectives
Given the burden of neurological diseases and the expected increase in their prevalence in the coming years, interventions that enable effective recovery while reducing the burden on healthcare facilities are increasingly required (Feigin et al., 2020).Indeed, as recently highlighted by the COVID-19 pandemic, current diagnostic and therapeutic services depend critically on hospital admissions, increasing both direct and indirect medical costs and leading to significant risks, especially for the most vulnerable patients (Pujolar et al., 2022).In this context, there is a considerable interest in the potential for effective therapies that do not require hospital visits but can be administered directly at local facilities or even remotely (e.g. through video recordings or home-based NIBS treatments), enabling a safe and cost-effective treatment of patients (Caumo et al., 2023;Sale and Franceschini, 2012).Therefore, AOT might be a fundamental tool for improving the functional outcomes of patients with neurological disorders.Indeed, it could be systematically used not only in the acute phases of the disease, when patients often show motor deficits preventing the effective execution of exercises, but also in post-acute settings, thus integrating current clinical neurorehabilitation practices.As cerebellar NIBS enhances motor learning not only online but also through a priming effect for the retention of subsequently acquired skills, the combination of AOT and cerebellar neuromodulation might significantly improve clinical outcomes (Cantarero et al., 2015;Wessel et al., 2016).This is likely through a common modulation of the MNS, which might enrich patients' motor repertoire and, consequently, their daily autonomy.
Notably, the utility could extend beyond patients with motor deficits, including the large group of those suffering from various pain-related issues, which are extremely common in the general population (Rice et al., 2016).Indeed, as recently suggested, pain may also result from a mismatch between the sensory and motor information in the internal models due to different circuit alterations (Antonioni et al., 2024b).Therefore, remodulating the sensorimotor associations characterizing internal models could significantly improve patient's quality of life.This is particularly relevant when considering that pain therapy often requires the combination of numerous medications, which are burdened by significant adverse effects and frequently incomplete efficacy (Siebenhuener et al., 2017).Furthermore, it could represent an important therapeutic aid for patients who, due to their clinical conditions, often cannot undergo traditional rehabilitative treatments, such as those in palliative care.This is not only because of these techniques' safety and minimal invasiveness but also because they can be administered in any care setting (Chowdhury et al., 2020).Thus, combining cerebellar NIBS and AOT might be crucial in this respect.Considering the recent evidence on chronic pain syndromes post-COVID-19, these issues are expected to become even more relevant in the next years (Baroni et al., 2023).Hopefully, the evidence discussed in this review will provide insights for improving the complex management of various patient categories, leveraging the fundamental synergy between the cerebellum and cortical networks reasonably mediated by the MNS.

Limitations and possible alternative interpretations
Here, we have sought to gather and analyze the best evidence about the cerebellar involvement in the MNS.Although extensive research suggests a crucial role for cerebellar modulation in this context, there are relevant aspects to consider.Firstly, unfortunately, much of this evidence is indirect, as the cerebellum has not been adequately studied by older brain MRI protocols (Talairach and Tournoux, 1988).Indeed, this established atlas sets a standard for neuroimaging reporting, but it is tailored for the cerebral cortex and lacks validation for the cerebellum.Furthermore, most neuroimaging studies must choose between a larger spatial coverage or an increased task sensitivity (i.e. a shorter acquisition time), leading researchers to favour a smaller field of view and, thus, to remove the cerebellum in many analysis protocols (Abdelgabar et al., 2019).Moreover, to support the involvement of the cerebellum in MNS activities, several meta-analyses have been included (Caspers et al., 2010;Hétu et al., 2013;Molenberghs et al., 2012;Van Overwalle et al., 2014).While they are important references in the literature due to their high quality, it is well known that meta-analyses can have several potentially significant limitations, chiefly the heterogeneity of the data collected in terms of sample size, data analysis techniques, study population, and potential confounding factors (e.g., age, gender, medication use) (Jennings and Van Horn, 2012;Müller et al., 2018).This is particularly relevant when considering that the evidence stems from healthy subjects and pathological contexts characterized by highly heterogeneous pathophysiology and various study designs, each characterized by its own advantages and disadvantages (Colditz, 2010).Furthermore, the data have been collected using different neuroimaging techniques, namely PET and fMRI.While the former is characterized by relatively low temporal resolution, the latter boasts excellent spatiotemporal definition but is often affected by artifacts and physiological variations (Wang et al., 2021).Moreover, the difference in the aspects investigated (i.e.brain metabolism versus blood flow variations) potentially complicates direct comparison (Jamadar et al., 2021).In addition to introducing considerable qualitative variability among the different studies, these aspects can lead to potential inconsistencies in the interpretation of results, which is crucial in meta-analysis of neuroimaging data (Metelli and Chaimani, 2020).In addition, there is often a publication bias in these contexts, as studies with significant results are more likely to be published than those with negative or inconclusive results, which could distort the reality of the available evidence (Page et al., 2021).Consistently, as previously discussed (see Section 2), the results presented by the included meta-analyses are often conflicting.However, it is encouraging to observe that the most recent meta-analyses, compared to earlier ones, confirm significant cerebellar activity during AO and MI (Hétu et al., 2013;Van Overwalle et al., 2014).This suggests that, thanks to more recent evidence, with more up-to-date and higher quality data (also due to advancements in neuroimaging atlases), the cerebellum's role is emerging more clearly and consistently despite the aforementioned limitations.
Furthermore, a direct recording of the cerebellar rhythms, both in healthy controls and patients, is severely limited by numerous technical issues that have constrained the application of EEG in studying cerebellar activity.Indeed, the cerebellum is a deep structure and the distance between the EEG signal source and the recording electrode decreases the signal-to-noise ratio, reducing the reliability of the recorded data (Andersen et al., 2020).Moreover, the distance of the cerebellum from the main source of the EEG signal, the contamination caused by myogenic activity, and its architecture provide further limitations for EEG applications (Muthukumaraswamy, 2013;Samuelsson et al., 2020).However, emerging research indicates that, under specific conditions, cerebellar activity can be detected using EEG, and it is likely that much more neurophysiological data on its involvement in various functions will be available in the future (Andersen et al., 2020;Samuelsson et al., 2020).
Finally, although this review has compiled numerous pieces of evidence on the involvement of the cerebellum in the context of MNS activities, as previously highlighted, many of the discussed findings are indirect.Thus, the interpretations presented here require further research to confirm their correctness.Indeed, it is important to consider that the results on cerebellar activity during AO and MI allow for additional interpretative approaches.For example, the cerebellum is critically involved in motor coordination tasks and the fine-tuning of movements, so its activation during tasks related to MNS functioning may just reflect a motor coordination aspect rather than a direct role in the imitation, MI or understanding of AO (Miall, 2022).Furthermore, as suggested by some authors, it is possible that the cerebellum merely encodes observed movements to make them more efficient and automatic, leaving the primary functions attributed to the AON in different brain areas (Balsters and Ramnani, 2011).It is also conceivable that the cerebellum's role is limited to processing data that supports the activities of cortical and subcortical regions (e.g., the basal ganglia), enhancing their computational capacity and, consequently, the subject's performance without being an integral part of the AON (Raymond and Medina, 2018).This might align with evidence showing increased cerebellar activation with general task complexity (Katz and Knops, 2016;van Dun et al., 2022).Additionally, the cerebellum might support high-level cognitive functions, such as attention and working memory, perhaps without a specific role in MNS activities (Schmahmann, 2019).However, studies demonstrating a strong correlation between AO, EEG biomarkers of MNS involvement (i.e.mu rhythm desynchronization), and cerebellar activation (as assessed through fMRI), together with evidence on the loss of observational learning due to AO following cerebellar lesions suggest a specific cerebellar role rather than mere co-activation or support to other key areas (Braadbaart et al., 2013;Petrosini et al., 2003).It is hoped that future studies will delve deeper into its precise contribution to this complex set of activities and clarify its specific interaction with other structures involved in shaping the significant effects of MNS-related activities (Caligiore et al., 2017b(Caligiore et al., , 2013)).

Conclusions
The evidence suggests that the cerebellum is critically involved in MNS activities, i.e.AO and MI, and that it may participate by modulating the activity of cortical inhibitory interneurons with mirror properties in visuomotor matching.During motor learning and movement observation, this phenomenon would reduce the discrepancy between what is expected and what is achieved.Indeed, it is reasonable to hypothesise that the cerebellum reinforces patterns that minimise, from a sensorimotor perspective, the difference between predicted and obtained outcomes, both during execution and during AO.Furthermore, through its connections with the hippocampus, it is reasonable that the cerebellum, during MI, participates in internal simulations of motor programmes, exploring numerous possibilities and reinforcing those most useful for achieving desired outcomes.However, further studies are needed to confirm this hypothesis.This aspect is innovative and crucial, as different NIBS techniques might also be used for therapeutic purposes, using cerebellar stimulation to promote motor recovery in patients.Consistently, cerebellar NIBS modulation appears to improve its impact on MNS activity by regulating the inhibitory circuitry of frontoparietal regions via the thalamus, and applications in neurorebilitative terms are steadily increasing.In conclusion, the MNS and the cerebellum seem to be strictly interconnected in their function, aimed at enhancing individual motor functioning, even in a pathological setting, a crucial point from a motor recovery perspective.
Fig.1.Schematic representation of the complex cerebellar circuitry and its main components.The vertical lines on the right show the subdivision into layers: the molecular layer (blue line), the Purkinje cell layer (yellow line), the granule cell layer (red line), and the layer of white matter fibers and the deep cerebellar nuclei (green line).On the left, examples of cerebellar glomeruli are shown, formed by the synapse between mossy fibers (in gray, originating from pontine nuclei that transmit information from cortical areas and also send a copy of this information to the deep cerebellar nuclei), the Golgi cell axon (violet cell), and the granule cell dendrites (green cell).In the same context, additional modulation is provided by the unipolar brush cells (light blue cells).The axons of the granule cells reach the molecular layer, where they give rise to parallel fibers, providing excitatory input not only to the extensive and branched dendritic arbors of the Purkinje cells (dark pink cells) but also to the inhibitory interneurons that feed-forward inhibit the Purkinje cells.Specifically, the stellate cells (red interneurons) mainly inhibit the apical portions of the Purkinje cell dendrites.In contrast, the basket cells (yellow interneurons) exert control primarily over the basal portion of the dendrites and the cell body of the Purkinje cells, which are located in the layer named after them, along with the soma of the Golgi cells and the Lugaro cells (dark orange cells).The climbing fibers, originating from the inferior olivary nucleus complex, also project 1:1 onto the soma of the Purkinje cells (similarly to the mossy fibers, they project a collateral branch to the deep nuclei in a topographically specific way).The inhibitory axons of the Purkinje cells represent the only output from the cerebellar cortex and reach the deep cerebellar nuclei, which, in turn, project to numerous targets in the central nervous system mainly via the thalamus.Note that various types of astrocytes in the different layers provide trophic and metabolic support to the complex activities performed, such as brown, blue, and light orange cells.Based on (D'Angelo, 2018; Schmahmann, 2019).Figure created with BioRender.com.[Color should be used in print].

Fig. 2 .
Fig. 2. Activation patterns in cortical and cerebellar regions during movement execution, AO, and MI.The schematic representation displays cortical (upper panel) and cerebellar (lower panel) activation patterns during movement execution (left), AO (middle), and MI (right).Stronger relative activation is depicted by darker colors, while striped regions indicate areas with highly task-dependent activation patterns.Abbreviations: AO: action observation; MI: motor imagery; M1: primary motor cortex; PMC: premotor cortex; S1: primary somatosensory cortex; SMA: supplementary motor area.Reproduced in accordance with the copyright policy by Frontiers in Systems Neuroscience from (Henschke and Pakan, 2023), © 2023 Henschke and Pakan.[Color should be used in print].
Fig.3.Schematic representation of the hypothetical model of the connections between the cerebellum and cortical/subcortical areas with inhibitory interneurons displaying mirror properties.On the left, an example of a cerebellar module is depicted (refer to Fig.1for more details).The deep cerebellar nuclei project primarily to the thalamus (on the top), which in turn projects mainly to the primary motor area (M1) (on the right).This influence affects the output (i.e., the pyramidal neurons located in the cortical layer V) not only directly but also through excitatory (orange cells) and inhibitory cortical (blue cells) interneurons.Other inhibitory interneurons (grey cells located in cortical layer VI) provide a modulation on projections from the thalamus.Some of these interneurons reasonably display mirror properties (highlighted by a dashed red circle) and are responsible for the cerebellar influence on measures related to the Mirror Neuron System (MNS) activities, such as electroencephalographic measures (e.g., mu rhythm) and non-invasive brain stimulation techniques (e.g., short and long interval intracortical inhibition, SICI and LICI).These inhibitory interneurons are primarily found in cortical layers II and III in M1, where they can modulate the excitability of pyramidal neurons (and consequently the motor output) both directly and through a complex regulation of the excitation-inhibition balance involving other interneuronal circuits across various cortical layers.Notably, interneurons with similar properties and subject to cerebellar influence also appear to be present in prefrontal and premotor regions, as well as in associative parieto-temporal areas (on the right), justifying the cerebellum's participation and in various aspects of MNS regulation.During the initial phases of learning, the cerebellum's influence on these interneurons may contribute to shaping the circuits by creating new visuo-motor (or audio-motor, depending on the specific areas) associations to optimize performance.When the discrepancy between predicted and actual outcomes is minimized, cerebellar influence becomes less relevant, and its activity diminishes.These interactions likely explain the performance improvements obtained not only by executing a task but also by merely observing it or listening to related significant stimuli.Furthermore, the cerebellum engages in bidirectional communication with the hippocampus (in the centre), both directly and through the thalamus and, to a lesser extent, the Median Raphe nucleus, a serotonergic brainsteam structure, which project to the medial septum and the nucleus of Broca's diagonal band, which provide access to the hippocampus.This allows the cerebellum to retrieve mnemonic traces during motor imagery (MI) to test the hypothetical sensory consequences of motor action (also using available information from cortical networks, which provide essential data on planning, strategy, multi-sensory associations, and a priori hypotheses from already established models), thereby refining interneuronal patterns to minimize discrepancies between predicted and expected outcomes.Conversely, once the models are optimized, the cerebellum strengthens hippocampal circuits that retain the optimized patterns, reasonably explaining the performance benefits associated with MI.Notably, these new internal models are available for other activities in which the MNS is involved, such as execution and AO, reasonably justifying the similar benefits derived from these different activities.Axons terminating bifidally indicate excitatory synapses, while those ending with a straight line indicate inhibitory ones.Based on(Bonini et al., 2022;Caligiore et al., 2017b;Casula et al., 2016;Daskalakis et al., 2004;Dum and Strick, 2003;Koch et al., 2008;Rolls, 2023;Rondi-Reig et al., 2022;Spampinato et al., 2023;Zeidler et al., 2020).Abbreviations: AO: action observation; LICI: long interval intracortical inhibition MI: motor imagery; MNS: Mirror Neuron System; M1: primary motor cortex; SICI: short interval intracortical inhibition.Figure created with BioRender.com.[Color should be used in print].

Table 2
Summary of evidence on cerebellar involvement in AO (first section) and MI (second section) in animal models and healthy human subjects.Abbreviations: AO: action observation; BM: biological motion; DCM: dynamic causal modelling; fMRI: functional magnetic resonance imaging; MI: motor imagery; NS: not specified; PET: positron emission tomography; SPECT: Single Photon Emission Computed Tomography.