Researchers identified four frequently recurring topological turning points in human brain structural networks across the lifespan by analyzing manifold projections of diffusion imaging data from birth to 90 years of age. The study, published in Nature Communications, reveals that brain network topology followed distinct developmental trajectories that shift at approximately at ages 9, 32, 66, and 83 years old, defining 5 organizational epochs.  Each epoch showed a characteristic direction of topological change (increasing vs decreasing integration) and a different set of metrics that best predicted age.

The analysis drew on 12 graph theory metrics from 4,216 participants across 9 datasets. Using Uniform Manifold Approximation and Projection (UMAP), researchers generated three-dimensional manifold spaces across 968 different parameter combinations to ensure robustness. Turning points were defined as ages most frequently identified across all manifolds—ages where the overall trajectory of topology shifts, not just peaks or valleys in individual metrics.

To harmonize data, researchers applied ComBat across nine diffusion imaging datasets. Structural networks were derived using generalized q-sampling (GQI) imaging with deterministic tractography and 5 million streamlines per participant. UMAP projections varied by nearest neighbors (2 to 89) and minimum distance (0.1 to 1.0). Turning points that appeared consistently across projections were identified as major lifespan shifts.

Density and Strength Patterns

To understand connectivity changes, variable-density networks (preserving 70% of raw network density)) showed nonlinear age associations, with denser networks near birth and around 30 years, and sparser networks around ages 10 and 80. Node strength increased nearly linearly across the lifespan, reaching lifetime peaks at age 90.  However, to isolate pure topological changes from density differences, the main analyses used density-controlled networks thresholded to exactly 10% density, the highest density achievable for all participants. After density control, average network strength still increased with age, though with different dynamics.

Epoch-Specific Topology

For density-controlled networks at exactly 10% density, researchers examined how integration, segregation, and centrality measures varied across the 5 identified epochs.

Epoch 1 (0 to 9 years)

Eight organizational measures correlated significantly with age, characterized by decreasing global integration, increasing local segregation, and stable centrality.  Although small-worldness showed the strongest correlation with age  regularized LASSO regression identified clustering coefficient as the strongest predictor of age, significant 55 of 90 brain regions after false-discovery rate (FDR) correction.

This turning point aligns with puberty onset (8 to 13 years for females, 9 to 14 for males), a period of hormonal changes, rapid cognitive development, and increased risk for mental health disorders. The shift from childhood to adolescence is marked not only by topological reorganization but also by synaptic elimination and cortical thickness stabilization (peaking around age 7).

Epoch 2 (9 to 32 years)

All topological measures showed significant age correlations. Small-worldness was both the strongest predictor and most strongly correlated metric. Integration increased while global segregation decreased, and local-level segregation continued to rise.  Modularity decreased across this period. 

While adolescence traditionally ends before age 20, the data suggest that topological maturation continues until around age 32. The transition to adulthood is influenced by cultural, historical, and social factors, but brain network organization follows a consistent trajectory across this entire epoch.

Epoch 3 (32 to 66 years)

Ten topological measures correlated significantly with age, showing decreasing integration, rising segregation, and stable centrality.  Clustering coefficient showed the strongest correlation with age, but LASSO retained local efficiency as the most informative predictor. These highly correlated measures were significant in 71 and 74 regions respectively after FDR correction, indicating that the age-topology relationship was distributed across the majority of the brain.

The 32-year turning point coincides with white matter maturation peaks.  Fractional anisotropy peaks around 29 years, mean diffusivity reaches a minimum around 36 years, and radial diffusivity reaches a minimum around 31 years. This shift from increasing to decreasing efficiency may reflect the transition from structural maturation to the onset of age-related degeneration.

At age 32, 5 topological metrics showed significant directional changes: global efficiency increased while, characteristic path length, small-worldness, modularity, and betweenness centrality all reversed their trends.  This turning point marked the largest shift in trajectory across the lifespan, with significant differences in Principal Component (PC) 1 and 2 scores between epochs 2 and 3.

Epoch 4 (66 to 83 years)

Only four topological metrics showed significant age correlations. Modularity emerged as both the strongest predictor and largest correlation, accompanied by decreasing integration and increasing centrality. The regularization of the LASSO had to be weakened for any predictors to survive, suggesting a decline in the strength of the age-topology relationship.

The 66-year turning point aligns with the onset of hypertension and dementia in high-income countries. This epoch marks accelerated white matter degeneration and increasing modularity, consistent with the shift from healthy aging to age-related decline.

Notably, at age 66, significant Principal Component Analysis (PCA) score differences occurred across all 3 principal components, and dynamic time warping showed this epoch transition had the most dissimilar trajectory pattern of any turning point—despite no directional changes in individual metrics. This suggests a fundamental reorganization of how topological features relate to one another, even if their individual trends remain stable.

Epoch 5 (83 to 90 years)

Only subgraph centrality maintained significantly association with age and was the strongest predictor though significance in only 10 regions. Statistical power was lowest in this epoch compared to earlier periods (mean power = 0.16 vs. 0.72–0.97 in earlier epochs).

The 83-year turning point was the second most frequently identified across UMAP projections (111 times ). While the small sample size (n=93) limits statistical power, the consistent detection of this turning point across manifolds suggests it reflects a real shift. The decline in significant age-topology relationships may indicate that individual variability in aging trajectories increases substantially in late life, obscuring population-level patterns.

Turning Point Characteristics

PCA across 11 topological metrics, revealed 3 principal components that explained 76.6% of variance:

• PC 1 loaded primarily on local segregation measures (clustering coefficient, local efficiency) and strength.

• PC 2 loaded on integration metrics (global efficiency, characteristic path length) and betweenness centrality.

• PC 3 loaded on centrality measures (subgraph centrality) and segregation metrics (core/periphery structure, small-worldness, modularity).

Significant shifts in PC 1 and PC 2 occurred at both the 9-year and 32-year turning points,  while 66-year turning point was unique in showing significant shifts across all 3 PCs, despite no directional correlation changes. The 83-year turning point showed significant changes only in PC 2. 

Sex-stratified analyses showed similar turning points by sex with some variation around the 66-year turning point. However, observed sex effects in individual topological measures could be confounded by brain size differences, which were not explicitly controlled in the network construction. Future work should disentangle size-related from organizational sex differences.

Methods and Sample

The study combined diffusion imaging from 9 datasets: Developing Human Connectome Project (dHCP), Baby Connectome Project (BCP), Centre for Attention Learning and Memory (CALM), Resilience in Education and Development (RED), Attention and Cognition in Education (ACE), Human Connectome Project Development (HCPd), Human Connectome Project Young Adult (HCPya), Cambridge Centre for Ageing and Neuroscience (camCAN), and Human Connectome Project Ageing (HCPa). The final neurotypical cross-sectional sample included 3,802 participants (1,994 female, 1,808 male).

All networks underwent fiber tracking using GQI with deterministic tractography. Networks were parcellated using age-appropriate AAL90 atlases (neonatal, 1-year-old, 2-year-old, and standard adult versions) and harmonized across datasets and atlases. For density-controlled analysis, all networks were thresholded to exactly 10% density and converted to normalized weighted networks.

Turning points were identified by fitting fifth-degree polynomials through three-dimensional manifold spaces, calculating gradients, and filtering with a gradient window of 5 years and gradient threshold of 0.8. Ages most frequently identified across all projections were defined as major turning points.

Study Limitations

The cross-sectional design of the study limited inferences about individual trajectories or causality. The turning points identified are population-level averages—individual developmental timing may vary considerably. Older adults may have been healthier than population norms, potentially underestimating age-related decline.  The low statistical power in Epoch 5 suggests that findings for this period should be interpreted cautiously. Future longitudinal studies are needed to validate these findings within individuals over time.

Full disclosures can be found in the study.

Source: Nature Communications