Cerebellar aging is spatially heterogeneous and supports cognitive resilience in later life

Nature Neuroscience

Cerebellar aging is spatially heterogeneous and supports cognitive resilience in later life

The cerebellum contains most of the brain's neurons and supports many functions, yet how it changes with age remains unclear. Here we used three brain imaging studies spanning forty-seven thousand adults and examined how different parts of the cerebellum age and their relation to cognition. We characterized cerebellar aging using volumetry and the T-one-weighted over T-two-weighted ratio, and corroborated these findings with quantitative magnetic resonance imaging in an independent sample. We show a spatially heterogeneous pattern of aging in which specific association and motor-related regions show steeper relationships with age than other lobules. Greater cerebellar volume was associated with higher cognitive scores with increasing age, suggesting that cerebellar structure may provide brain reserve that helps maintain function despite aging. In patients with Alzheimer's disease, cerebellar volume was linked to cognition in individuals with lower amyloid burden, especially in those carrying two copies of the APOE-epsilon four risk gene. This supports a threshold-reserve model, in which the cerebellum helps sustain cognition until pathology becomes widespread. These results show that the cerebellum has an active role in healthy cognitive aging and resilience.

Global population aging underscores the need to identify neural mechanisms supporting cognitive longevity. The number of individuals older than seventy years is increasing faster than that of younger adults. Although research has focused largely on the cerebral cortex, converging evidence indicates that the cerebellum, long recognized for sensorimotor control, also supports higher-order cognitive and affective functions. Through closed-loop circuits with cortical association areas and subcortical nuclei, the cerebellum is positioned to influence cognition across the lifespan. Posterior cerebellar regions linked to prefrontal and parietal networks exhibit protracted maturation, suggesting a potential role in late-life cognitive resilience.

The cerebellum is not spared by aging. Morphometric and postmortem studies show hemispheric and vermal shrinkage, and loss of Purkinje and granule cells; longitudinal imaging shows accelerating atrophy in advanced age. Aging effects appear particularly pronounced in posterior regions, and cerebellar volume correlates with executive function and processing speed in older adults. Yet, whether cerebellar variation contributes to cognitive reserve remains unclear because most prior studies used modest samples or coarse anatomical resolution, limiting assessment of spatial heterogeneity.

Reserve frameworks emphasize neural efficiency and adaptability as buffers against decline, but focus largely on neocortical and hippocampal systems. The 'dysmetria of thought' hypothesis proposes that cerebellar dysfunction disrupts higher-order cognition analogously to motor dyscoordination, implying that degeneration may produce cognitive dysmetria whereas preserved cerebellar integrity may confer protection. The cerebellum has also been implicated in Alzheimer's disease, but while neuroimaging studies suggest cerebellar involvement in Alzheimer's disease, whether it contributes to pathology or resilience remains unclear. Some work even reports slower cerebellar than neocortical structural decay, although those analyses treated the cerebellum as a single structure and may have overlooked regional heterogeneity. Reduced cerebellar volume across neurodegenerative disorders further underscores the need to clarify its role in aging.

In this study, we combine large-scale multimodal data from the Human Connectome Project, UK Biobank, and the Alzheimer's Disease Neuroimaging Initiative to model cerebellar aging and its relationship to cognition, amyloid burden and the APOE genotype. We asked: Is cerebellar aging spatially heterogeneous? Do individual differences in cerebellar aging relate to cognitive resilience that is consistent with reserve? How do these processes manifest in Alzheimer's disease dementia? By integrating normative and pathological aging, we position the cerebellum within broader frameworks of cognitive aging and reserve.

Results

Heterogeneous aging of cerebellar tissue and associations with cognition

In seven hundred eight neurotypical adults aged thirty-six to one hundred years, we examined age-related changes in cerebellar volume relative to the neocortex, controlling for biological sex, estimated total intracranial volume and years of education. Age was inversely associated to neocortical volume, Cohen's F equals one point two seven and cerebellar volume, Cohen's F equals zero point five four, but not with estimated total intracranial volume.

To assess spatial heterogeneity, we parcellated the cerebellum into eleven rostrocaudal lobules. A mixed-effects model of log-transformed lobular volume showed strong age-related volume associations and significant heterogeneity across lobules.

F of one, zero one four, eight two six equals fifty point zero seven, P is less than two point two times ten to the negative sixteen. Main effects were present for age, F of one, seven zero two equals one four seven point three one, P is less than two point two times ten to the negative sixteen, lobule, F of one, zero one four, eight two six equals sixty-five, seven five six point two six, P is less than two point two times ten to the negative sixteen, hemisphere, F of one one four, eight two six equals sixty-four point zero four, P equals one point three one times ten to the negative fifteen, sex, F of one, seven zero two equals seven point four eight, P equals zero point zero zero six, eTIV, F of one, seven zero two equals three two seven point zero two point zero two, P is less than two point two times ten to the negative sixteen, F of one, seven zero two equals twelve point three one, wise slope comparisons were Benjamini-Hochberg-corrected.

Allowing hemisphere-specific slopes did not alter lobular heterogeneity (age times lobule: F of one, zero one four, eight zero five equals fifty-one point three nine, P is less than two point two times ten to the negative sixteen). There was no overall hemispheric difference in age slopes (age times hemisphere: F of one one four, eight zero five equals one point five three, P equals zero point two one six) and no age times lobule times hemisphere interaction, F of one, zero one four, eight zero five equals one point two seven, P equals zero point two three nine.

We next examined vermis regions using the same log-volume framework. Mixed-effects models revealed heterogeneous age slopes across vermis regions (age times vermis: F of forty-two, eight two zero equals four point five six, P equals zero point zero zero one), alongside a main effect of age, F of one, seven zero two equals seventeen point two zero, P equals zero point zero zero zero zero four. eTIV, F of one, seven zero two equals one six one point seven four, P is less than two point two times ten to the negative sixteen and education, F of one, seven zero two equals five point four eight, P equals zero point zero one nine were significant covariates, whereas sex was not, P equals zero point one six six. Most vermis regions declined by approximately zero point nine to one point five percent per decade. Vermis VI, VII, IX and X showed significant negative estimates after Benjamini-Hochberg correction, whereas vermis VIII was near zero.

To assess whether volumetric reductions reflected loss of fine-scale tissue structure, we analyzed cerebellar T one-weighted over T two-weighted ratio maps, a contrast sensitive to myelin and neurite density. Although T one-weighted over T two-weighted remains susceptible to cerebrospinal fluid partial-volume effects, it reduces intensity inhomogeneity and enhances tissue contrast. T one-weighted over T two-weighted declined with age across lobules: mixed-effects models (covariates: sex, eTIV, education and hemisphere) showed heterogeneous age slopes (age over lobule: F left parenthesis one, zero one four, eight two six right parenthesis equals fifty-three point seven five, P is less than two point two times ten to the negative sixteen, with main F left parenthesis one, zero one four, eight two six right parenthesis equals two, eight one nine point four three, P is less than two point two times ten to the negative sixteen, F left parenthesis one one four, eight two six right parenthesis equals one, eight six zero point nine two, P is less than two point two times ten to the negative sixteen, and education F left parenthesis one, seven zero two right parenthesis equals ten point seven nine, P equals zero point zero zero one. Sex and eTIV were not significant both P is greater than zero point eight. Percentage change per decade showed significant decline across all regions after Benjamini-Hochberg correction, with the largest decreases in crus one (three point one three percent), crus two (two point nine eight percent) and lobule nine (three point one four percent), and the smallest in the anterior motor regions, consistent with a cerebellar sensory-association gradient.

We next compared cerebellar aging to neocortical organization along the sensory-association hierarchy. For each cortical region, we related mean T one-weighted over T two-weighted in three hundred sixty young adults (a proxy of sensory-association rank) to its age-related T one-weighted over T two-weighted slope in the aging cohort. Regions with lower baseline T one-weighted over T two-weighted-typically association cortices-showed more negative age slopes, whereas highly myelinated sensory regions were relatively preserved.

Because T one-weighted over T two-weighted is a relative metric, we validated these patterns using quantitative magnetic resonance imaging in an independent cohort. We derived maps of the volume of interacting protons, sensitive to tissue water macromolecular interactions expected to decline with tissue loss. Mixed-effects models controlling for sex and hemisphere showed significant lobular heterogeneity (age over lobule: F left parenthesis one zero, four six two right parenthesis equals five point one zero, P equals four point four times ten to the negative seven, with main effects of lobule F left parenthesis one zero, four six two right parenthesis equals forty-eight point three one, P is less than two times ten to the negative sixteen and hemisphere F left parenthesis one, four six two right parenthesis equals four point nine seven, P equals zero point zero two six, but not age F left parenthesis one, two zero right parenthesis equals three point six one, P equals zero point zero seven two. Simple slopes showed reductions in crus one, crus two, seven b, eight a, eight b and nine (uncorrected P is less than zero point zero five). After Benjamini-Hochberg correction, seven b and seven la remained significant P is less than zero point zero five, with crus one and two, eight b and nine trending P is less than zero point one. A posterior (crus one and two) versus anterior (one to three, four) contrast confirmed faster decline in posterior cognitive lobules (estimate equals negative zero point zero zero two per decade, t equals negative four point three zero, P is less than zero point zero zero zero one, an approximately two percentage point greater decline per decade than the anterior regions. These quantitative magnetic resonance imaging findings replicate the posterior-biased pattern observed with T one-weighted over T two-weighted.

To clarify the biological interpretation of volume of interacting protons, we examined age associations for tissue volume fraction, water tissue fraction and volume of interacting protons across cerebellar lobules. Tissue volume fraction and water tissue fraction exhibited complementary age-related patterns consistent with shifts in macromolecular versus water content across the adult lifespan. Volume of interacting protons demonstrated partially overlapping but distinct age associations and several lobules showed volume of interacting protons age effects despite minimal tissue volume fraction or water tissue fraction slopes. These findings suggest that volume of interacting protons captures microstructural variation beyond simple tissue loss or water accumulation, providing a complementary and partial-volume-robust index of cerebellar microstructural aging.

To assess whether cerebellar structure was related to cognition, older adults completed the Montreal Cognitive Assessment, a battery sensitive to cognitive decline. Higher Montreal Cognitive Assessment scores were associated with greater neocortical volume beta equals zero point zero zero zero one, P is less than two point one times ten to the negative nine, cerebellar volume beta equals zero point zero zero zero one, P is less than three point one times ten to the negative ten, Cohen's F equals zero point two four performance beta equals zero point zero zero two, zero point zero one one, Cohen's F equals zero point one zero Figure eleven. Thus, greater global tissue volume, cortical and cerebellar, was related to cognitive