Resistance, Vulnerability and Resilience: A Review of the Cognitive Cerebellum in Aging and Neurodegenerative Diseases

Abstract

In the context of neurodegeneration and aging, the cerebellum is an enigma. Genetic markers of cellular aging in cerebellum accumulate more slowly than in the rest of the brain, and it generates unknown factors that may slow or even reverse neurodegenerative pathology in animal models of Alzheimer’s Disease (AD). Cerebellum shows increased activity in early AD and Parkinson’s disease (PD), suggesting a compensatory function that may mitigate early symptoms of neurodegenerative pathophysiology. Perhaps most notably, different parts of the brain accumulate neuropathological markers of AD in a recognized progression and generally, cerebellum is the last brain region to do so. Taken together, these data suggest that cerebellum may be resistant to certain neurodegenerative mechanisms.

On the other hand, in some contexts of accelerated neurodegeneration, such as that seen in chronic traumatic encephalopathy (CTE) following repeated traumatic brain injury (TBI), the cerebellum appears to be one of the most susceptible brain regions to injury and one of the first to exhibit signs of pathology. Cerebellar pathology in neurodegenerative disorders is strongly associated with cognitive dysfunction. In neurodegenerative or neurological disorders associated with cerebellar pathology, such as spinocerebellar ataxia, cerebellar cortical atrophy, and essential tremor, rates of cognitive dysfunction, dementia and neuropsychiatric symptoms increase. When the cerebellum shows AD pathology, such as in familial AD, it is associated with earlier onset and greater severity of disease. These data suggest that when neurodegenerative processes are active in the cerebellum, it may contribute to pathological behavioral outcomes.

The cerebellum is well known for comparing internal representations of information with observed outcomes and providing real-time feedback to cortical regions, a critical function that is disturbed in neuropsychiatric disorders such as intellectual disability, schizophrenia, dementia, and autism, and required for cognitive domains such as working memory. While cerebellum has reciprocal connections with non-motor brain regions and likely plays a role in complex, goal-directed behaviors, it has proven difficult to establish what it does mechanistically to modulate these behaviors. Due to this lack of understanding, it’s not surprising to see the cerebellum reflexively dismissed or even ignored in basic and translational neuropsychiatric literature.

The overarching goals of this review are to answer the following questions from primary literature: When the cerebellum is affected by pathology, is it associated with decreased cognitive function? When it is intact, does it play a compensatory or protective role in maintaining cognitive function? Are there theoretical frameworks for understanding the role of cerebellum in cognition, and perhaps, illnesses characterized by cognitive dysfunction? Understanding the role of the cognitive cerebellum in neurodegenerative diseases has the potential to offer insight into origins of cognitive deficits in other neuropsychiatric disorders, which are often underappreciated, poorly understood, and not often treated.

Introduction

Several observations have made the cerebellum a tantalizing subject of study with respect to cognition, particularly as it may relate to symptom etiology in neuropsychiatric and neurodegenerative diseases known to result in cognitive dysfunction. In addition to its traditional role in the regulation of movement, the cerebellum has parallel interactions with the limbic components of the brain essential for cognition and emotional regulation. From an evolutionary perspective, expansions in neocortical regions in the brain are matched only by expansions of regions of cerebellar cortex and white matter associated with cognitive functions in primate, and potentially even recent hominid, evolution [1-3]. Humans have cerebellar activation during fear, reward, unpleasant, and activating emotions [4-8], as well as cognitive discrimination and sensory acquisition [9-11]. Cerebellar volume and function is correlated with cognitive function, attention, language, memory, hallucinations, thought organization, and affect in humans with schizophrenia and Alzheimer’s Disease (AD) [12, 13][14-18]. Focal lesions (e.g. strokes, tumors) in the cerebellum can cause non-motor cognitive and affective problems in the context of mild or no motoric abnormality [19]. These deficits include disturbances of executive function, impaired spatial cognition, anxiety, depression [20]. The cerebellar cognitive affective syndrome can also encompass personality changes with flattening of affect and disinhibited and inappropriate behavior, linguistic problems [21], and resemble clinical criteria of subjects presenting with autism and schizophrenia, diagnoses that are increasingly recognized as having associated cognitive deficits [12]. Two large-scale genome-wide association studies identified genetic loci associated with intelligence and cognitive function [22, 23]. Each study used methods to examine where in the brain these genes were expressed, and cerebellum was consistently at the top of the list for highest expression, rivalling only frontal cortex [22, 23].

The role of the cerebellum in higher cortical processing can be described to certain extent on an anatomic level. Cerebellar output pathways arise from the cerebellar nuclei, with non-motor efferents primarily arising from the dentate and interpositus nuclei. Cerebellar nuclei project to multiple regions of brainstem, midbrain, diencephalic nuclei involved in cognition, including the caudal tail of the ventral tegmental area (VTA) [24], laterodorsal tegmental nucleus, reticulotegmental nucleus of the pons[25-27]; and also basilar pontine nuclei, and multiple thalamic nuclei [28] [29]. Dentate nuclei also have prominent projections to the pedunculopontine nucleus [29, 30], and locus coeruleus [28], where it potentially modulates acetylcholine and norepinephrine release, respectively [29]. Electrical stimulation of different cerebellar regions results in increased electrical activity in multiple limbic areas, such as hippocampus [31, 32], anterior cingulate gyrus, and amygdala (though pathways between cerebellum and hippocampus or amygdala are not known) [33]. Stimulation of dentate nucleus of the cerebellum (DCN) increases release of dopamine in the medial prefrontal cortex (mPFC), via pathways through VTA, mediodorsal and ventrolateral thalamus [34]. The VTA projects to granule cells in Crus II of cerebellum [35] and also sends projections to the deep cerebellar nuclei, primarily the DCN and interpositus nuclei, but not the medial nuclei [35]. Lesions of the cerebellum result in increased dopamine D1 receptor levels in medial striatum, and decrease in dopamine transporter expression in dorsolateral striatum, indicating cerebellar modulation of phasic dopaminergic neurotransmission in striatum [36]. Reciprocal pathways connecting cerebellum with striatum have been demonstrated in primates and rodents [37-39]. Despite substantial evidence supporting a role of the cerebellum in higher cognitive function, the cerebellum remains poorly understood. One striking observation that may shed some light on its role in development of cognitive dysfunction is that the pattern of cerebellar involvement differs significantly in age-related dementias such as AD (in which it may play a role in maintaining cognitive homeostasis) compared to early onset dementia in diseases such as TBI-associated dementia and CTE (in which cerebellar integrity may actually be compromised). Here we explore theoretical frameworks for understanding cerebellum’s role in higher cognitive function and pathogenesis of cortical dysfunction, and review the literature of cognitive cerebellum in aging and dementia.

Theoretical frameworks for understanding the role of cerebellum in cognitive behavior

The cerebellum is thought to regulate cognitive functions primarily through its interactions with different cortical regions, as a function of its role in different intrinsic connectivity networks [40]. Intrinsic connectivity analyses tell us which parts of the brain are active during specific tasks, machine learning tells us how components may work individually or in concert, and multiple parallel memory systems theory helps to predict what may happen if a particular system breaks down or becomes dominant relative to other systems. We propose that the key to the understanding of cerebellum’s contributions to cognition, particularly in neurodegenerative diseases, is through the synthesis of these conceptual frameworks relating cerebellum as a component in brain connectivity networks to machine learning, and Multiple Parallel Memory System theory.

Intrinsic connectivity mapping in humans has shown that cerebellum is connected with nearly all regions of neocortex (including visual [41-44] and auditory [45] cortex ) (reviewed in [40]). Areas of cortex related to attention independently activate cognitive regions of cerebellum [46]. The cerebellar cortex has been shown in human studies to have specific patterns of activation related to specific functions and cortical regions [47, 48]. Cerebellum is activated in at least five different intrinsic connectivity networks [49]. These networks include the sensorimotor network, the default mode network, right and left executive control networks, and the salience network [49]. Specific connectivity networks appear to be vulnerable in specific neurodegenerative diseases [50], and cerebellar circuits which are part of these networks appear to be vulnerable, or at least show changes in activity, as well [16]. Cerebellar circuits in the default mode network show reduced activity in AD, and cerebellar circuits in the salience network show reduced activity in frontotemporal dementia [16].

Several observations at the circuit level support the idea that the cerebellum acts in this fashion in cognition, as cerebrocerebellar connections appear to be organized into long-range loops [51-59]. The cerebellar nuclei have analogous subdivisions, and both cerebellar cortex and nuclei receive inputs from distinct cerebrocerebellar pathways [60]. The reticulotegmental nucleus of the pons relays inputs from motor and sensorimotor areas of neocortex to medial nucleus, anterior interpositus, and the mediodorsal region of the lateral nucleus, while the basilar pontine nuclei relay inputs from visual and auditory sensory, cognitive and associative areas of neocortex to the lateroventral part of the lateral nucleus and posterior interpositus nucleus, with little overlap [60]. Additionally, cerebellum has multiple reciprocal anatomic connections with midbrain dopaminergic nuclei and striatal system [35, 37-39], amygdala [61], hippocampus [62-64], and with regions of motor and association cortex via thalamic nuclei (reviewed in [40, 47, 65]). Neocortical network activity (local field potentials and multiunit recordings) appears to drive cerebellar network activity, and this cerebellar activity is abolished when neocortical activity is blocked (but not vice versa in this study) [66]. Activity in cerebellar cortex, particularly in granule cells, Purkinje cells (simple and complex spiking), and Golgi cells activity correlated with this neocortical activity [66]. Conversely, electrophysiological and optogenetic stimulation of lateral nucleus neurons in rats results in modulation of dopaminergic neurotransmission in frontal cortex and dopamine dependent timing behavior [34, 67]. In the generation of voluntary movements, there is sequential activation of cerebrocerebellar loops, starting with more cognitive regions (prefrontal cortex and lateral cerebellum), and ending with motor regions (motor cortex and midline cerebellum) [68].

Cerebellar modulation of cortically-based cognitive functions is thought to occur through detection of pattern changes and errors in thought, providing adaptive feedback to the cerebral cortex [69], and by encoding internal models that reproduce essential properties of mental representations in the cerebral cortex, which may underlie implicit thoughts and intuition [70, 71], At a circuit level, the cerebellum is known for efficient processing of ‘efference copy’ or, corollary discharge signals [72, 73]. In motoric function, the efference copy is a copy of the self-generated motor (corticospinal) command. The cerebellum compares the expected feedback of this command to the actual state of the organism, which allows for short-latency feed-forward adjustments in real time to achieve optimal kinetic outcomes [71]. The cerebellar nuclei have feedback projections into cerebellar cortex [72, 73], which allow for maintenance of amplitudes in specific, adaptive behavioral responses in the face of uncertain neural inputs, and thus reducing functional decline [74]. Expert motoric function depends on a highly efficient comparator mechanism, to achieve corrections as quickly as possible. Sensory prediction error, that is, mismatch between ‘expected’ and ‘actual’ outcomes typically results in a short-latency corrective behavior. In the context of cognition, the outcome of successful cerebellar efference copy processing would be to reduce cognitive load, and maintain adaptive functional responses in performance of predictive tasks requiring attention [74] [75]. The cerebellum has been shown to play a role in predictive timing of perceptual and cognitive events [76-78]. This function would facilitate efficient decision-making and attention to salient stimuli, especially when real-time attention demands are high such as engaging in social interactions, language, and discrimination of salient social and environmental stimuli [70, 79], functions that are highly relevant to dementia. The question remains, how does the cerebellum facilitate this in cognitive function?

The concept of the cerebellum as a neuronal machine was introduced in 1967, in which the precise geometry of the cerebellum’s modular microanatomical structure was described, with implications for its functions [80, 81]. This set the stage for cerebellar network modeling [81, 82], as well as the sensory-feedback machine learning model proposed by Kawato [83]. Machine learning is a field that aims to develop computer algorithms that improve with experience [84], and the concepts of different styles of machine learning are applicable to our understanding of different regional brain functions. Doya proposed a framework in which cerebellum, as a specific type of learning machine (supervised), interacts with other regions of the brain with different machine learning styles [79]. Supervised machine learning algorithms learn with data containing training examples, containing correct labels (in the case of the cerebellum, error coding by climbing fibers projecting to Purkinje cells) [71, 81-83, 85-87], and are good at learning continuous values or classifications and making certain kinds of predictions, but may require large amounts of training sets to be useful [84]. Advantages of supervised learning include fast response times and high performance after learning, but may have only limited transferability or generalizability, and thus a limited set of responses. The cerebral cortex and basal ganglia are specialized for unsupervised learning and reinforcement/reward learning, respectively [88]. Data processed by unsupervised learning algorithms is unlabeled, i.e., no training sets are provided, and algorithm learns how data is clustered or associated, similar to a principal component analysis [84]. In the mammalian brain, this type of learning is based on Hebbian plasticity in cortical neurons and reciprocal connections within and between cortical areas [79, 88]. Reinforcement learning does not occur with training examples, instead, there is a learning agent that learns from experience with the goal of maximizing reward (and/or minimizing punishment), and is encoded by dopaminergic fibers in the midbrain [79]. Model-free and model-based reinforcement systems are both present in the brain and play different roles in certain kinds of decision making processes [89]. Advantageous outcomes of reinforcement learning models include increasing the rate of learning via a process of updating (i.e., not requiring large sets of training examples for learning complex data), as well as mechanisms for temporal and probabilistic discounting that aid associative learning; flexible, relational learning and generalizability [89]. Houk and Wise proposed a model of these systems interacting in arrays of distributed processing modules, and proposed several functions for these systems acting in tandem [85, 90]. Similarly, these machine learning styles likely collaborate in multiple domains, including motor function, sequence learning, timing and rhythm, and cognitive processing [79]. While this view of different components of the nervous system is likely oversimplified (e.g., prediction error processes are likely active in many regions of the cerebral cortex [91]), it is useful and illustrative of how the cerebellum, via its numerous reciprocal anatomical connections, participates in network function in different learning domains. Schmahmann has argued that the role of the cerebellum is likely similar in its interactions with all brain regions it interacts with “because cerebellar anatomy is essentially uniform throughout the structure, the basic work that cerebellum does in the nervous system should be constant as well. This we have referred to as the universal cerebellar transform, characterized as the cerebellar modulation of behavior, serving as an oscillation dampener maintaining function automatically around a homeostatic baseline and smoothing out performance in all domains” [21]. The analogies of specific brain structures (particularly cerebellum) to types of machine learning rely on reference to our understanding of local anatomical microstructure. But given the rich interconnections the cerebellum has with other regions of the brain, its role as recognized in Houk and Wise’s distributed processing model is particularly important [85, 90]. If the analogy to machine learning systems to brain networks is viable, then it stands to reason that multiple brain systems coordinating would result in a hybrid machine learning system. Thus, the cerebellum is perhaps best understood as a part of a larger hybrid machine learning system (or many hybrid learning machines), with reinforcement and unsupervised components. Hybrid machines benefit from advantages of each style of machine learning, and as such have high performance, generalizability and may show rates of learning over shorter time scales, with emergent abilities not available to each individual component [92]. Emergent abilities that hybrid machines have include anomaly and novelty detection, and making predictions [93]; which are functions classically associated with cerebellum [94].

What would it look like to lose the cerebellar supervised learning machine in cognition? This question in the context of cerebellar-basal ganglia-cortical network interactions is particularly important in the context of aging and neurodegenerative disease. Before we attempt to answer, we will place the question in the context of what is known about the cognitive cerebellum in aging.

Cognitive cerebellum in aging: morphology, physiology, and functionality

The cerebellum appears to have a specific pattern of aging. In a stereological study of human neocortex across age, neuron numbers declined by 9.5%, volume decreased by 12.3%, white matter volume decreased by 28%, total brain weight decreased by 12%, while other cortical volumes decreased by 28% [95, 96]. Cerebellum shows similar changes with age: total cortical volume decreased by 10.8% over the lifespan, total cerebellar volume decreased by 16%, and total cerebellar white matter volume decreased by 25.9% [97]. Cerebellar subregions show differential losses in volume [97]. The posterior lobe (which is more associated with cognitive functions) only loses about 10.6% of its volume (and about 27% of its white matter volume), while the anterior lobe (a region more associated with motoric functions) loses almost 29% of its volume (and about 36% of its white matter volume) over the lifespan, much of it attributed to granule and Purkinje cell loss [97].

Brain imaging studies show similar results, and offer the ability to look at cerebellar morphology throughout the lifespan. Cerebellar development shows close coupling with cerebral cortical development [98-101]. Changes in volumetric measurements of cerebellum vary regionally through the lifespan, with cognitive areas showing an “inverted U” like pattern from adolescence through late middle age [98]; total cerebellar volume peaks at about mid-adolescence [98, 101], with anterior and vermal regions peaking in adolescence, and cognitive areas showing a peak at about age 30 [98], similar to prefrontal cortex [102]. One study found that the cerebellar hemispheres and vermis lose approximately 2% of their volume per decade between the ages of 20 and 80 years of age [103]. Another study using diffusion tensor imaging found that microstructure organization of the superior cerebellar peduncles (the main output tract of the cerebellum) increased with age, and strongly correlated with IQ (as opposed to all other white matter tracts) [104].

Several observations in post-mortem tissue have led to the idea that the cerebellum ages more slowly than other parts of the brain, and may perhaps be resistant to developing pathological markers of neurodegeneration. These include changes in DNA markers of aging such as 1) that cerebellar cortical volume negatively correlates with age, but not cerebellar leukocyte telomere length [105], 2) cerebellum accumulates fewer mitochondrial DNA deletions [106], 3) has less oxidative damage to mitochondrial and nuclear DNA [107], and 4) shows fewer changes in gene expression with age relative to cortex [108]. Furthermore, the cerebellum shows decreased expression of markers of metabolic and oxidative stressors, SIRT1 and SOD-1 [109], and has less age-related DNA methylation [110].

Imaging studies have also allowed characterization of cerebellar physiology as a function of aging. In young adults, cerebellum has the lowest amount of regional glucose metabolism (at approximately 70% of cortical regions), but also shows the lowest amount of change in glucose metabolism over the lifespan [111]. Cerebellum also shows relatively low amounts of aerobic glycolysis [112], an energetic process which supports development and synaptogenesis, but which may also put brain at risk for oxidative damage [113]. Interestingly, Crus of the cerebellum are part of a recently identified network identified which develops late in adolescence and shows accelerated degeneration in old age, and have heightened vulnerability in illnesses such as schizophrenia and Alzheimer’s disease, suggesting a link between developmental and neurodegenerative processes [114].

Age-related changes are found in classical cerebellar functions including timing, classical conditioning of eyeblink, vestibulo-ocular reflex, optokinetic reflex, and postural control (reviewed in [115]), however, there are very few studies looking at relationships between cognitive cerebellum and aging. In one study on the cognitive effects of aging, cerebellar gray matter volume (but not frontal cortical areas) completely accounted for general cognitive abilities (which include non-verbal reasoning, short- and long-term memory, and speed of information processing) with stronger correlations in males than in females [116]. Posterior and lateral changes may account for this, as another study showed no correlation between changes in cerebellar vermis volumes and cognitive changes with respect to age [117].

Pathological changes in cognitive cerebellum in neurodegenerative diseases

One region of the cerebellum recognized to have cognitive function is the dentate nucleus of cerebellum. DCN activation increased 3 to 4 fold during attempts to solve a pegboard puzzle compared to movements only in a functional MRI study [118]. DCN regions related to cognitive function show evolutionary expansions mirroring those in cognitive regions of the cerebellar cortex and frontal cortex [3]. DCN shows specific connections and activations related to cognitive functions in monkeys and humans [119-122] [123]. Untargeted optogenetic stimulation and genetically-targeted chemogenetic inhibition experiments in rodents have shown sufficiency and necessity, respectively, of DCN cells for cognitive functions such as timing, working memory, spatial navigation memory, social recognition memory, and response inhibition [29, 67]. DCN is known to accumulate iron as a function of aging, though it is not clear if this correlates with changes in cognitive function [124, 125]. Several neurodegenerative diseases show pathological changes in DCN (which may account for associated cognitive deficits), including: changes in cell cycle marker expression in AD [126], Lewy bodies in Parkinson’s Disease Dementia and Dementia with Lewy Bodies [127, 128]; cell death in frontotemporal dementia [129]; tau accumulation in progressive supranuclear palsy [130]; calcification in bilateral striopallidodentate calcinosis (‘Fahr’s Disease’) [131-133]; and pathological iron deposition in Friedrich’s ataxia [134, 135]. Studies investigating the role of the cognitive cerebellum in patients with cerebellar degeneration with event related potentials implicate the lateral cerebellum (which includes the dentate nucleus) in response preparation and selection, and cognitive processing speed [136, 137]. Other neurodegenerative diseases with more general cerebellar pathology, cognitive dysfunction and increased rates of dementia include Huntington’s disease [138], spinocerebellar ataxias [139], cerebellar degeneration [140]; and essential tremor [141-146]. DCN appears to be an understudied locus in studies relating cognitive function with aging or neurodegeneration.

Cerebellum in Alzheimer Disease: Resistant to AD pathology?

While it is clear that cerebellum has altered morphology and activation in AD, it is unclear whether these features are causal of symptoms or simply reflective of dysfunction elsewhere. Earlier research showed that amyloid deposition in cerebellum occurred late in the disease with rare occurrence of neurofibrillary tangles [147, 148]. Another study showed that cerebellum showed less synapse loss than hippocampus and frontal cortex [149]. These observations have led some to interpret that cerebellum may actually be protected from or resistant to developing AD pathology. For example, cerebellum is resistant to the neurotoxic effects of soluble amyloid-beta (Aβ) [150]. This may be mediated by a variety of mechanisms. The receptor for advanced glycation end products (RAGE), which when it binds Aβ, leads to microglial activation, is elevated in hippocampus and superior frontal gyrus of patients with AD relative to controls. Curiously, it shows no elevation in cerebellum, suggesting that this mechanism of Aβ–mediated toxicity is not as active there [151]. Microglia form a network of cells that spans the CNS, including the cerebellum, and express TREM2 [152]. TREM2 is a gene implicated in CNS resilency to AD pathogenesis, regulates neuro-immune mediated clearance of Aβ and is expressed in cerebellum [153]. Wild-type mice show increasing numbers of microglia in the cerebellum as a function of age (which is not seen in Trem2 knockout mice), suggesting Trem2-dependent neuro-immune clearance capacity may be greater in the cerebellum than in other brain regions [154]. A Trem2 mutant associated with frontotemporal dementia has less age-dependent microglia activity in cerebellar white matter in mice [155]. Mitochondrial DNA deletion levels remain low in the cerebellum in AD, but are quite high in other brain regions in AD patients, earlier in the disease [156]. Microglia mediate synaptic remodeling via the complement system, which can be over-activated by Aβ production leading to neurotoxicity [157] and neurodegeneration [152, 158]. Curiously, complement factor C1q is lowest in expression in cerebellum relative to other brain structures in non-demented individuals, but is present in cerebellum of patients with AD [159], and full complement activation is not seen in AD cerebellum [160]. Metabolites derived from cultured cerebellar media, when infused into hippocampus and neocortex of a mouse with Aβ deposits, induce expression of the Aβ degrading enzymes, insulin degrading enzyme (IDE) and neprilysin (NEP) [161]. NEP and IDE decrease with age in hippocampus, whereas their levels stay the same or even increase in cerebellum with aging [162].

Aberrant protein deposition is affected in part by neurotransmitter signaling, and vice versa. Norepinephrine promotes degradation of Aβ via promoting microglia to upregulate expression of IDE [163]. When left unchecked, Aβ deposition reduces both DA and dopamine receptors in frontal and temporal cortex [164, 165]. Similarly, the principle site of norepinephrine synthesis, the locus coeruleus (LC), has also been implicated as one of the first sites of tau accumulation in AD [166, 167]. Norepinephrine levels are decreased in temporal cortex in AD, and correlate with cognitive impairment [168]. Cerebrospinal fluid concentration of norepinephrine increases with age, and inversely correlates with cognitive performance on tasks engaging attention and working memory [158]. The lowest amount of neuronal loss in LC in AD occurs in the caudal LC, where LC-cerebellar projections arise [169-173]. Taken together, these results suggest that cerebellum may not be affected by pathological Aβ accumulation until later in AD.

Cerebellum in Alzheimer Disease: Conferring resilience when AD pathology is present?

Given that the cerebellum is relatively unaffected in early AD and its involvement may accelerate disease progression, it begs the question of whether the cerebellum may also play a role in compensation for AD related deficits, maintaining brain resilience when other parts of the brain are challenged with AD-related pathology. Cerebellar activation changes have been monitored in studies looking at AD progression, and this subject was recently reviewed in detail elsewhere [174]. Low connectivity in the cerebellum is found in patients with mild cognitive impairment and AD [174-177], with increasing connectivity correlating with higher levels of cognitive function [178, 179]. Better memory was associated with increased cerebellar activity in a cohort of AD patients studied longitudinally [180]. Observations of increased fMRI activity in cerebellum in patients with Parkinson’s disease have led to a similar interpretation that the cerebellum may play a compensatory role in this neurodegenerative disease [128, 181]. These data suggest that while the cerebellum may be protected from AD pathology in early disease, the spread of degenerative changes to the cerebellum likely heralds late stage disease and may contribute to cause of death associated with AD. Furthermore, in cases where the cerebellum is spared, it may play an active role in neuroprotection.

Cerebellum in AD and other dementias: Vulnerability to disease progression?

In this review we have so far primarily examined cerebellum in Alzheimer’s disease from the perspective of having homeostatic changes in disease progression in an age dependent manner. In later sections, we will contrast this with cerebellum in CTE, in which cerebellar tissue integrity is compromised, though not in an age-dependent manner. In this subsection, we compare and contrast role of cerebellum in AD (in which we are not sure whether there are age dependent changes in tissue integrity/function) and another dementia, frontotemporal dementia, in which cerebellum’s integrity and function is known to be affected in an age dependent manner.

Volume changes in two areas of cerebellum have been used to help differentiate genetic variants of frontotemporal dementia [182]. Tan et al. examined patients diagnosed on the amyotrophic lateral sclerosis (ALS) - behavioural variant frontotemporal dementia (bvFTD) spectrum with cognitive, neuropsychiatric and functional evaluations as well as structural imaging [183]. They found specific patterns of gray matter atrophy in ALS and bvFTD patients, specifically, that ALS patients had cerebellar atrophy in inferior lobules (correlating with changes in motor function) and bvFTD patients had atrophy in cerebellar crus and the superior lobule (and associated with cognitive and neuropsychiatric symptoms) [183]. Patients with both ALS and bvFTD symptoms had both patterns of atrophy [183].

It has long been noted that the AD pathology occurs in a cascade, with various disease markers occurring in a spatially and temporally ordered fashion (reviewed in [184]). It is not well understood how disease markers in specific brain regions correlate with specific cognitive and neuropsychiatric symptoms in the clinic [184]. When clinical symptoms are correlated with pathological changes in the entire brain, neurofibrillary tangles (NFTs) are more closely related to cognitive impairment than amyloid plaques [185, 186]. Furthermore, the appearance of pathological markers of AD such as amyloid plaques and NFTs in subcortical structures such as brainstem and cerebellum coincides with the overt clinical dementia syndrome, while earlier in the disease cortical deposition of these markers is present, but without clinically detectable behavioral changes [187]. However, neurodegeneration, as measured by neuron and synapse loss, is robustly related with clinical cognitive symptoms [188, 189]. AD patients with aggression were found to have increased expression of adrenergic receptors (particularly alpha-2 receptor) in cerebellar cortex, but not frontal cortex or hypothalamus [190]. Cerebellar cortex shows reduced tyrosine-hydroxylase positive fibers relative to controls in AD [191]. In a study that carefully correlated neuropsychiatric symptoms with brain-region specific concentrations of monoamines and monoamine metabolites, concentrations of serotonin in cerebellum and temporal cortex, and concentration of 5-HIAA in hippocampus (and not several other brain regions) closely correlated with performance on the mini-mental status exam (MMSE) [192, 193]. Furthermore, the most significant correlation in the study was between specific neuropsychiatric symptoms and cerebellar markers of catecholaminergic neurotransmission (and, surprisingly, none in cortical regions) [192]. In a study tracking 600 patients with mild cognitive impairment or AD over the course of developing delusions, gray matter atrophy was detected in several brain regions, including anterior cerebellum (culmen) and areas of frontal cortex [194]. Notably, a study utilizing data from the Alzheimer’s Disease Neuroimaging Initiative has shown that changes in right cerebellar cortical volume is one of 15 predictors of conversion from MCI to AD [195]. Other studies using Alzheimer’s Disease Neuroimaging Initiative data have shown that increases in cerebellar metabolism utilizing FDG-PET imaging is associated with conversion from MCI to AD [196], and that cerebellum shrinks in size in patients with AD, but not MCI, and is that AD-associated decrease in volume is not associated with education or APOEe4 genotype [197]. Significant reduction of total cerebellar volume by almost 13% has been observed in patients with AD compared to age-matched controls (but no significant reduction in numbers of granule cells or Purkinje Cells, nor white matter volume) [198]. In one small study, neuronal loss was seen in cerebellar cortex and inferior olive, as well as gliosis and atrophy, despite there not being any amyloid plaques or NFTs [199]. In another study, volume and cell losses were seen in all layers of cerebellar cortex, as well as presence of amyloid plaques [200]. Among biomarkers in living patients with AD, FDG-PET shows good correlation with synaptic dysfunction in dementia [184, 201]. The aging cerebellum shows about 12% loss of FDG uptake relative to younger subjects, while the aging supratentorial brain shows approximately a 23% loss in FDG uptake relative to younger subjects [202]. Atrophy patterns in cortical structures progress in a recognized pattern, which in turn correlate to measures of cognitive function [184, 203, 204]. However, FDG uptake in cerebellum also correlates significantly with measures of cognitive function including the MMSE and the Dementia Severity Rating Scale [205]. It is thus apparent that changes in cerebellar neurotransmission, particularly monoaminergic neurotransmission, are at least reflective of, if not causally related to, neuropsychiatric and cognitive symptoms in AD.

AD is a progressive, and ultimately fatal, neurodegenerative disorder. Cerebellar glucose metabolism shows significant decreases in severe/advanced Alzheimer disease (severity of illness determined by performance on MMSE) [206]. Observations of cerebellar vulnerability raise the possibility of cerebellar pathology playing a role in the ultimate outcomes of living with the disease. AD related deaths are often attributed to aspiration and pneumonia, suggesting brainstem dysfunction associated with swallowing and breathing in later stages. Cerebellar modulation of the musculature involved in swallowing is recognized [207]. In AD, the vestibulocerebellar system is particularly vulnerable. In the aging brain, the flocculonodular lobe, part of the vestibulocerebellar system, loses 22.5% of its total volume, and nearly 60% of its white matter volume [97]. A morphological analysis of the vestibulocerebellar system in AD compared with normal controls by Baloyannis and colleagues revealed extensive synaptic alterations in multiple components of the granule cell layer of the vestibulocerebellar system, including mossy fibers, granule cell dendrites, parallel fibers and Purkinje cell dendritic spines [208]. Furthermore, granule and Golgi cell numbers and mossy fiber-granule cell dendrite synapses were also decreased. Although pathologic examination revealed only a limited number of neuritic plaques and minimal NFTs, synapses containing limited polymorphous synaptic vesicles and numerous atypical mitochondria and dense bodies were observed mostly in the mossy fiber terminals. Loss of synapses and morphological alterations of the Golgi apparatus were also seen in medial and lateral vestibular nuclei neurons [208]. This pattern of neurodegeneration is reminiscent of a fatal breathing dysfunction attributed to vestibular nucleus dysfunction seen in a mouse model of Leigh syndrome (knockout of the gene Ndufs4, NADH dehydrogenase (ubiquinone) iron-sulfur protein 4), a neurodegenerative disorder in humans that results in death in infancy [209]. Total knockout of Ndufs4 results in degeneration of the vestibulocerebellar system [209]. Selective inactivation of Ndufs4 in the vestibular nuclei (VN), one of the principle sites of gliosis, led to breathing abnormalities and premature death [209]. Vestibular nuclei get rich input from the cerebellum, and learning and plasticity in the reflexes that VN supports appears dependent on cerebellar input [210]. Conversely, Ndufs4 restoration in the VN corrected breathing deficits and prolonged the life span of knockout mice [209]. While these observations are not related to cognitive functions, they illustrate that cerebellum may still play a very important role in pathophysiological mechanisms in AD—though very few studies (as reviewed above) have looked at synaptic changes in different regions of the cerebellum, particularly in cognitive regions. It’s possible that even though cerebellum shows minimal expression of classical markers of AD, more subtle cerebellar neurodegenerative changes leading to synaptic and cognitive dysfunction could still be important in the clinical progression of the disease, especially in subtypes of Alzheimer’s disease, which may have various patterns of atrophy in various regions of the brain, including cerebellum [16, 211]. Elucidating this may reveal novel therapeutic targets for cognitive symptoms in various stages of AD. In order to clarify how it may be relevant as a vulnerable structure in age-related neurodegenerative diseases such as AD, we next review data from chronic traumatic encephalopathy literature, which has a more specific focus on cerebellar function.

Cerebellum in CTE and TBI: Accelerated dementia?

TBI can lead to chronic traumatic encephalopathy (CTE) and increases the risk of dementia, with the risk of AD increasing by as much as two-fold in patients with a history of TBI [212-216] and some studies showing the correlation with vascular dementia to be even stronger [217]. Risk of dementia has been shown to be closely correlated with increased severity [217, 218] and number of TBIs [217, 219] in a dose-dependent manner. TBI may also accelerate dementia onset, as reviewed in detail elsewhere [220]. While the correlation between TBI and dementia is well documented, mechanistic causality has yet to be elucidated. One observation that may provide some mechanistic clues lies in gross anatomic observations of the cerebellum in dementia patients. In contrast to age-related neurodegeneration where the cerebellum remains relatively untouched until late stages of disease, the cerebellum appears to be one of the first brain regions affected in CTE and other types of dementia following TBI. Cerebellar atrophy appears to be common in human patients post-TBI [221], even when the initial insult is not directed at the cerebellum itself [222, 223]. One of our goals with this review is to explore the literature behind cerebellar dysfunction following TBI to help us better understand whether and how the cerebellum may be involved in the pathogenesis of dementia, which may also provide clues behind the contribution of the cerebellum to overall cortical health and function.

Cerebellum in CTE and TBI: Observations in humans

Investigation of clinical brain structure and function using various imaging techniques reveals vulnerability of the cerebellum months to years following TBI. Axonal injury in the cerebellum was found in the blast-exposed veterans using diffusion tensor imaging (DTI), especially in the middle and superior cerebellar peduncle [224-226]. The middle cerebellar peduncle carries projections from the pontine nuclei to the cerebellum; disruption of this tract in animals has been shown to result in deficits in motor learning tasks [227]. Cerebellar damage may very well be secondary to TBI-induced shear stress on these cerebellar input tracts. Decreased metabolic activity has been used as a marker for damaged brain regions, and indeed PET imaging of brains of blast-exposed veterans has also revealed that the cerebellum was less metabolically active compared to other brain regions following blast-induced TBI, an observation that was consistent across two separate studies [224, 228]. Meabon and colleagues further demonstrated that there was an inverse dose-dependent relationship between number of blast exposures and metabolic activity [224]. As reviewed in detail elsewhere [229], evidence also suggests that axonal injury caused by TBI may lead to amyloid plaque formation, which may further contribute to cerebellar dysfunction. Recent data using PET imaging from Mohamed and colleagues revealed higher amounts of amyloid beta formation in the cerebellum of veterans exposed to TBI compared to controls, whereas little difference was seen across other brain regions [230]. These human data paint a picture that implicates the cerebellum in pathogenesis of post-TBI brain injury, with much of the evidence pointing toward axonal injury leading to downstream cerebellar damage with Aβ formation and decreased metabolism.

Cerebellum in CTE and TBI: Animal models and investigation of mechanism

Various animal models have been developed to further study mechanisms of brain injury post-TBI, reviewed in detail by Potts and colleagues [231]. These models have corroborated and expanded much of the clinical evidence for cerebellar vulnerability to TBI. Several animal studies of blast-induced TBI have demonstrated axonal degeneration [224, 232, 233], microglial activation [224, 234], Purkinje cell damage [224, 234, 235], and microvascular hemorrhages [236], and found these changes to be more pronounced in the cerebellum compared to other brain regions. In their mouse model of blast-induced TBI, Meabon and colleagues found cerebellar mechanisms of injury (axonal injury, amyloid precursor protein expression, microglial activation, Purkinje cell damage) that appeared dose-dependent, with increased markers of pathology correlating with increased number of blast exposures [224]. The time-course for changes on the cellular level were observed within hours to weeks. Regionally in the cerebellum, the ventral lobules appeared the most vulnerable to Purkinje cell loss. Interestingly, they also found that the blood brain barrier of the cerebellar molecular layer displayed prominent injury and microvascular extravasation into the parenchyma. These animal data reveal cerebellar pathology that confirm axonal injury as well as amyloid protein involvement observed in humans post-TBI, while also revealing mechanisms of further tissue injury and inflammation, including Purkinje cell damage and cell loss, microglial activation, and blood brain barrier compromise. Microglial activation may be particularly important, as deficits in cerebellar conditioned learning and fine motor coordination are observed after exposure to Tetrahydrocannibinol, and are dependent on microglial activation in cerebellar cortex [237].

Cerebellar Vulnerability after TBI

Based on these data, Meabon et al. proposed possible vulnerabilities particular to the cerebellum that may explain its particular susceptibility to damage from repeated TBI including anatomic weaknesses of its location in the posterior fossa, and on a cellular level, the high metabolic demands of Purkinje cells [224]. Evidence suggests that excitotoxicity may also play a role in neuronal dysfunction. Bullock and colleagues analyzed cerebrospinal fluid samples of patients post-TBI and found that levels of excitatory amino acids increase in the days following injury [238]. Cerebellar Purkinje cells are particularly prone to excitotoxicity due to strong and overwhelming glutamatergic input of climbing fibers from the inferior olive in which a single climbing fiber forms hundreds of synapses on Purkinje cell dendrites. Mechanisms of Purkinje cell excitotoxicity are reviewed in detail elsewhere [239]. Computational modeling suggests that the cerebellum may be exposed to some of the largest amounts of shear stress during blast-induced TBI compared to other brain regions [240], which likely contributes to susceptibility of cerebellar tracts to axonal injury. It is not unreasonable to suspect that deafferentation may play a role in leading to cortical dysfunction following TBI-associated axonal damage; however further investigation is necessary to elucidate the relationship between cerebellar injury and specific effects on other brain regions.

Several studies in humans with specific neurodegenerative diseases further illustrate some of these relationships between cerebellum and other structures in the brain, and that losing cerebellar function can be particularly problematic for acquiring new information. Procedural learning is generally thought to be intact in AD [94, 241-243] but affected in patients with Huntington’s [244] PD [245-247] and cerebellar degeneration [248]. Using a serial reaction time task, Pascual-Leone et al., examined patients with either PD or cerebellar degeneration to test procedural learning and translation of procedural learning into declarative knowledge [140]. PD patients were able to achieve procedural knowledge and translation into declarative knowledge, but required more repetition relative to healthy controls to achieve this. Patients with cerebellar degeneration never achieved performance improvement in procedural learning or translation into declarative knowledge. They infer that basal ganglia (BG) are required for acquisition of procedural learning, and that cerebellum is essential for learning any cognitive function involving sequences. Both BG and cerebellum interact with the prefrontal working memory buffer; BG for acquisition, and cerebellum for proper temporal sequencing. Individuals with PD, both on and off medication, were found to have normal acquisition and short-term retention of a non-motor procedural learning task (mirror-reading of new words) [140]. The authors speculate that the basal ganglia may be more important for retention of learning in the motor rather than non-motor domain [249]. Indeed, patients with AD have been found to have decreased performance relative to controls on learning sequences, with severity of dementia showing a direct relationship to the ability to acquire them [243].

Unraveling the role of the cognitive cerebellum in neurodegenerative disease

To summarize so far, we have discussed mechanisms for how the cerebellum participates in larger networks to facilitate cognitive function, and reviewed how cerebellum is affected in several cognitive disorders of aging: in some circumstances it appears either resistant or vulnerable to dementing processes. Multiple Parallel Memory Systems Theory describes how different memory systems, such as cerebellum (or the particular intrinsic connectivity networks in which it is part), can either facilitate or inhibit other memory systems in a cooperative or competitive manner [250-252]. Functionally, this results in parallel, independent memory systems converging to produce either cooperative facilitation of similar behavioral responses, or competitive facilitation of different behavioral responses in the same situation, and is illustrated in experiments designed to dissociate contributions of different memory systems in specific behavioral tasks [252]. McDonald and White performed a triple dissociation experiment by lesioning hippocampus, dorsal striatum or amygdala in rats to identify contributions of these regions in performance on a radial arm maze task [253]. Similar experiments have been performed in the context of instrumental learning [254-256], and working memory [257]. Others have shown similar dissociations of specific cell types in hippocampus and striatum with pharmacologic and transgenic approaches in mice in spatial navigation tasks [258, 259]. Lee and Kim showed differential contributions of amygdala, hippocampus, and cerebellum to eyeblink conditioning in rats in a similar dissociation experiment [260]. McDonald and White argue that no task is purely dependent on one learning system, and that these parallel systems facilitate learning different aspects of learning several kinds of behaviors, as would likely be encountered in the wild [252]. Disabling one system would likely restrict learning of a particular task or aspect of a task in the case of damage to a particular parallel system that is cooperative with others. In the case of disabling a memory system that is competitive with other systems, the remaining functional, competitive learning system(s) would exert greater control over behavior, and perhaps even result in supernormal acquisition and performance of a specific task [252].

Behaviorally, most of these learning systems (including cerebellar-based learning) are to some degree dissociable, and their interactions facilitate both motor and reinforcement learning. Izawa and Shadmehr had human subjects learn a novel motoric task (a reach adaptation with a robotic arm) which could be learned with or without sensory feedback [261]. While adaptation of motor commands could be driven by reward prediction errors when sensory feedback was lost, only learning from sensory prediction errors appeared to generalize and resulted in remapping of the neural system which predicts consequences of motor commands [261]. Dissociating negative and positive reinforcement from a similar motoric adaptation task, Galea et al. found that negative feedback/reinforcement accelerated learning and relearning of a movement, while positive feedback resulted in increased retention of the motor memory when sensory feedback was withdrawn [262]. Increased activations of cerebellar cognitive regions were seen in learning the reversal of a reward-based task with negative feedback (monetary loss) [263]. Initiating the correct or accurate action in operant tasks had a marked influence on dopamine release in the nucleus accumbens of rats, and varying the size of the reward elicited a significant statistical interaction between reward prediction error and action initiation [264]. In contrast, in a series of papers, sensory prediction error was found to have no significant effect on reinforcement learning in humans [265, 266]. Ablation of the hippocampus facilitates acquisition of eyeblink conditioning [260], while inhibition of the central nucleus of the amygdala during delayed eyeblink conditioning severely impaired acquisition of this cerebellar based learning [61,267]. Reversible inhibition of the interpositus nucleus abolishes retention of fear conditioned memory [268], and fear induces long term changes in synaptic strength in vermis Purkinje cells [269]. Using genetic tools to selectively impair long term depression (one of the primary mechanisms of cerebellar memory formation) in Purkinje Cells, Rochefort and colleagues demonstrated the necessity of cerebellar cortical plasticity for generation of normal hippocampal place cell properties and pathway integration (dead reckoning) in a water maze without external visual cues [63]. Collectively, these studies show that cerebellum plays an intimate role in learning of several approach and avoidance behaviors classically associated with other, more “cognitive” regions.

While these examples support the idea that cerebellum is primarily a bidirectional cooperative player amongst other learning systems, it’s also possible that cerebellar modules, or rather, the intrinsic connectivity networks they reside in, may compete with other memory systems/connectivity networks. For example, in a small study of individuals with spinocerebellar ataxia type 14, affected individuals performed significantly better than intrafamilial controls in verbal learning and memory, despite having some executive functioning deficits, and relatively normal performance on others tests of cognition [270]. In a study with children with autism spectrum disorder (ASD), children with ASD were found to have much stronger association between motor commands and proprioceptive feedback after learning a novel motoric task (a robotic reach adaptation), and that the stronger the proprioceptive-driven generalization pattern of trained movements to novel movements, the greater the impairment in overall motor function, social interaction and responsiveness, and imitation behaviors [271]. This kind of imbalance between networks as a basis for disease is not without precedent. Take, for example, the default mode network (DMN). It plays an important role in perpetuating maladaptive ruminative thinking styles in depression, and appears to be overactive relative to another network, the “task-positive network” (prefrontal and parietal structures) which is less active, in depression patients vs. controls [272, 273]. A recent meta-analysis found brain regions with altered function and connectivity common to many psychiatric disorders, and that they support executive functions [274]. The authors found that gray matter loss converged across diagnoses in 3 regions: the dorsal anterior cingulate, right insula, and left insula [274]. These regions are part of the DMN, and formed a tightly interconnected network during tasks and at resting; lower gray matter in this network was associated with poor executive functioning [274], and altered connectivity here is associated with AD [16, 50, 275, 276].

The roles of different learning machine types in a given system are now potentially testable in silico, and may adequately model observations from humans with specific cerebellar deficits, in the context of their roles in intrinsic connectivity networks such as the DMN. For example, it may be possible to model such a hybrid network with supervised, unsupervised and reinforcement components. Techniques such as DropOut [277] or DropConnect [278] have been invented to prevent overfitting of data to machine learning models. They could potentially be used in the context of hybrid machines to see if specific combinations of network units (say, just the supervised or unsupervised units) can independently extract features from their inputs, and compare performance to an entire hybrid machine, or more finely test which features are extracted by which units in a hybrid machine.

It may be possible to conceptualize cerebro-cerebellar loops as hybrid learning machine arrays in the brain. We previously noted a study identifying at least 5 intrinsic connectivity networks that include parts of cerebellum [49]. This number may be conservative when considering how only relatively few cerebellar nuclear output neurons are required for cerebellar modulation [279], thus potentially multiplying the number of possible hybrid learning machine arrays of which the cerebellum is part.

While higher level cortical processing tends to be thought of as a separate entity from motor learning and coordination, which tends to be viewed as taking place subcortically, there is much evidence that these entities are much more integrated than is generally appreciated. Specific anthropometric measurements, grip strength and walking speed, are well-known and practical phenotypic markers of aging, functional decline, and mortality (reviewed in [280]). Gray matter volume in cognitive cerebellum (but not vestibular cerebellum) was associated with faster gait speeds and information processing in a population-based study of aged adults [281]. Hand motor function is negatively associated with cognitive impairment and dementia, which is reviewed in detail elsewhere [282]. In a study looking at neurological correlates of grip strength in aging, activity in thalamus, putamen and cerebellum were higher in the aging group [283]. Notably, the posterior “cognitive” cerebellum had particularly high activation in the aged group [283]. Patients with Parkinson’s Disease develop abnormally high grip forces [284], suggesting that cerebellum may perhaps be more important in decreases of grip strength in cognitive decline and dementia. Decreases in grip strength are also observed in neuropsychiatric illnesses known to have cerebellar and cognitive abnormalities, including schizophrenia and autism [285-287]. As we continue to investigate neural circuitry involved in cognition and learning, it will be important to continue to recognize these complexities. The literature reviewed here highlights a role for the cerebellum in higher level cortical processing and suggest a potentially protective role from development of cognitive dysfunction. Further investigation of these complex circuits amongst brain regions at all levels will be essential in furthering our understanding of cognition and may allow us further insight into mechanisms behind cognitive dysfunction at the highest levels.

Highlights.

• The role of the cognitive cerebellum in neurodegenerative disease is unknown

• We review cognitive cerebellar findings in aging and neurodegenerative disease

• Machine Learning and Memory Systems Theory inform cognitive cerebellar roles

Abbreviations:

All abbreviations are identified in their first instance throughout. Genes and proteins are given standard notations.

AD

Alzheimer’s Disease

PD

Parkinson’s Disease

CTE

chronic traumatic encephalopathy

TBI

traumatic brain injury

FTD

Frontotemporal Degeneration

DCN

Dentate Nucleus of the cerebellum

LCN

Lateral cerebellar nuclei

VTA

ventral tegmental area

mPFC

medial prefrontal cortex

BG

basal ganglia

MMSE

mini-mental status exam

Aβ

amyloid-beta

NFT

Neurofibrillary Tangles

DMN

default mode network