Three major dimensions of human brain cortical ageing in relation to cognitive decline across the eighth decade of life

The study addresses how patterns of cortical brain ageing unfold within individuals during the eighth decade of life and how these changes relate to cognitive decline. Prior work has often relied on cross-sectional data, which conflates age-related changes with lifelong structural differences and cannot capture dynamics of within-person change. The authors emphasize the need for longitudinal analyses to reveal coordinated regional atrophy patterns and their cognitive significance. They investigate whether regional cortical volumes change together over time, whether distinct spatial dimensions of cortical ageing exist, and how these brain change factors relate to declines in general cognition and the domains of memory, visuospatial ability, and processing speed in community-dwelling older adults scanned at ~73, ~76, and ~79 years.

Cross-sectional studies have shown that cortical regions cluster based on morphometric correlations, particularly noting greater age effects in later-developing, association cortices linked to complex cognition. However, cross-sectional designs cannot resolve the dimensionality and time course of ageing-related changes. A prior multivariate longitudinal study in an age-heterogeneous clinical sample with amnestic MCI (N = 317) identified five coordinated atrophy patterns: posterior default mode, prefrontal, medial temporal, spared (sensorimotor/occipital), and a diffuse global pattern. It remains unclear how such patterns manifest in generally healthy older adults and how cross-sectional structural correlations relate to shared longitudinal change.

Participants were from the Lothian Birth Cohort 1936, a longitudinal study of brain and cognitive ageing. They were assessed at Wave 2 (~73 years, N = 866), Wave 3 (~76 years, N = 697), and Wave 4 (~79 years, N = 550). Structural brain MRI was collected at Waves 2–4 using a 1.5T GE Signa HDxt scanner with an 8-channel head coil. T1-weighted 3D volumes (1 mm isotropic) were acquired, assessed by a neuroradiologist for stroke, and processed in FreeSurfer v5.1 using the longitudinal pipeline with Desikan-Killiany parcellation into 34 cortical ROIs per hemisphere. Outputs were visually checked; errors were corrected or scans excluded. The analyses included 1376 MRI scans (Wave 2: 629; Wave 3: 428; Wave 4: 319). Cognitive testing targeted three domains based on prior hierarchical SEM models: Visuospatial ability (WAIS-IIIUK Matrix Reasoning, Block Design; WMS-IIIUK Spatial Span total), Processing speed (WAIS-IIIUK Symbol Search, Digit Symbol; Inspection Time; Four-Choice Reaction Time), and Verbal memory (WMS-IIIUK Logical Memory total, Verbal Paired Associates total; WAIS-IIIUK Digit Span Backwards); MMSE was also administered. APOE ε4 status was determined by TaqMan genotyping of rs7412 and rs429358. Statistical analysis proceeded as follows: for each ROI per hemisphere, growth curve models (latent intercept and slope) were fit in an SEM framework using lavaan in R with full information maximum likelihood; slopes were expressed as percent change per annum. The latent correlation matrices for intercepts and slopes were extracted and analysed separately. Exploratory factor analysis using a Schmid–Leiman transformation (minres extraction, oblimin rotation) was applied to intercept and slope correlation matrices to identify a general factor and group factors, focusing on slope structure (2–4 factor solutions evaluated via eigenvalues, loading patterns, and RMSR). Replicability across hemispheres was assessed with Pearson’s r and factor congruence coefficients. Sensitivity analyses excluded participants with subsequent dementia diagnosis or MMSE <24, and separately those with radiological evidence of stroke, to test robustness of factor structures. A confirmatory factor analysis then imposed the EFA-derived slope factor structure (a bifactor model with an uncorrelated general factor and correlated group factors). Non-significant ROI slope loadings were set to zero. Multivariate SEMs extended this model to relate cortical change factors to trajectories of cognitive change at the level of general cognitive ability (g) and the three domains, using Factor of Curves growth models for cognition with test-specific intercepts and slopes loading onto domain-level latent intercepts and slopes. Cognitive test intercepts and slopes were corrected for sex; cortical factors showed no sex differences. Residual correlations were included between slopes of spatially contiguous ROIs; negative residual variances were constrained to zero for convergence. Multiple comparisons across brain–cognition associations were controlled using FDR. Analyses were conducted in R 3.5.0 and Mplus 8.2.

- Longitudinal changes across cortical ROIs were highly correlated (average r = 0.805, SD = 0.252), substantially exceeding cross-sectional correlations among ROI volumes (average r = 0.350, SD = 0.178). - A broad, cortex-wide general factor of cortical atrophy explained 66% of variance in longitudinal changes across ROIs. - Two additional, orthogonal group factors—fronto-temporal and occipito-parietal—together explained an additional 20% of variance in change. - Overall cortical volume declined by 0.87% per annum (~3475 mm³/year relative to baseline). Regional declines were heterogeneous: greatest in frontal and temporal poles, parietal cortex, lateral occipital, and lateral frontal regions; smaller in insula, cingulate, and pre/postcentral areas. - The general cortical change factor was associated with steeper declines in cognition: general cognitive ability g (r = 0.431, p < 0.001), visuospatial ability (r = 0.415, p = 0.002), processing speed (r = 0.383, p < 0.001), and memory (r = 0.372, p < 0.001). - Participant retention decreased across waves; individuals returning for later waves had modestly higher baseline cognitive scores (and slightly lower cortical volume at Wave 3), indicating potential attrition bias.

The findings demonstrate that within-person cortical atrophy during the eighth decade is organized along three major dimensions: a dominant, cortex-wide factor and two regionally specific fronto-temporal and occipito-parietal factors. The much higher correlations among longitudinal changes than among cross-sectional levels underscore the value of longitudinal designs for capturing coordinated ageing processes. The general cortical atrophy factor’s association with declines across general cognitive ability and key domains indicates that widespread cortical degeneration is a principal neural substrate of cognitive ageing in this cohort. The identification of additional, orthogonal regional factors suggests heterogeneity in ageing trajectories that could reflect distinct neurobiological pathways. Robustness checks excluding participants with potential dementia or stroke aimed to ensure that the observed factor structures are not driven by clinical pathology. Overall, the work clarifies how ageing-specific brain changes, rather than lifelong structural differences, relate to cognitive decline.

This study delineates three principal dimensions of cortical ageing in generally healthy older adults: a dominant general atrophy factor and two orthogonal fronto-temporal and occipito-parietal factors. The general factor is strongly linked to declines in general cognitive ability and in visuospatial, processing speed, and memory domains. By leveraging longitudinal MRI and cognitive data, the study distinguishes ageing-specific patterns from cross-sectional associations and highlights the importance of longitudinal approaches for understanding neurobiology of cognitive ageing. Future work should investigate determinants and biological substrates of the identified factors, examine their predictive value for clinical outcomes, and test generalizability across diverse populations, imaging platforms, and earlier or later life stages.

- Potential attrition bias: participants who returned for later waves had modestly higher baseline cognitive scores (and slightly lower cortical volume at Wave 3), which may affect generalizability. - Although dementia-free at baseline, some participants may have developed cognitive impairment or dementia during follow-up; sensitivity analyses excluding such cases were conducted to assess robustness. - Exclusion of scans with uncorrectable segmentation errors and reliance on a single 1.5T scanner and FreeSurfer v5.1 may limit generalizability to other acquisition/processing pipelines. - The cohort comprises generally healthy, community-dwelling older adults from a specific region, potentially limiting applicability to more diverse or clinical populations.