Brain aging comes with changes in the structure and function of the cortex, but these changes unfold differently in different people. Neuroimaging has already described normative patterns of cortical thickness and area, yet how these changes are organized at the systems level and linked to the molecular mechanisms of aging remained unclear. Nearby in the archive sit papers on brain age maps, cellular aging by methylation, sleep as brain cleanup, white matter wiring, and division that preserves memory. All of them look at aging from different angles, but none connects morphometric networks directly to longevity gene expression.
The team took structural MRIs from 952 healthy participants across three open databases (NKI, OASIS-1/2, IEEE-Openhb). For each scan, the cortex was divided into 148 regions using the Desikan atlas, and eight morphometric features were extracted: surface area, gray matter volume, mean thickness, thickness variability, mean and Gaussian curvature, folding index, and curvature index. Morphometric similarity networks (MSNs) were built by computing pairwise correlations of features between regions. Then the SuStaIn method (Subtype and Stage Inference) was applied to identify subtypes and stages within three networks: the default mode network (DMN), the central executive network (CEN), and the visual network. Longevity genes were drawn from the GenAge database (251 genes overlapping with the Allen Human Brain Atlas), and their expression was matched to the networks' nodal strengths via partial correlations, controlling for cortical thickness and area.
SuStaIn revealed two subtypes only in networks based on volumetric features (4vf), but not in curvature-based (4cf) or mixed (5f) networks. Subtype 1 (normative aging) was linked to genes of metabolism, insulin signaling, and immune regulation: IRS1, PPARGC1A, GSK3B, E2F1. Subtype 2 (compensatory) was tied to genes of stress response, apoptosis, DNA repair, and protein degradation: DDIT3, UBB, STUB1, GSK3B. Four genes (E2F1, MXD1, GCLM, GSK3B) overlapped. Subtype 2 had a higher mean age (70.0 vs 65.5 years), but the difference was not significant. The gene-to-MSN link was stronger in the left hemisphere: 24% of explained variance versus 10% in the right. When compared with a cognitive impairment group (CDR 1-2, n=38), subtype 1 turned out to resemble it more closely than subtype 2 did.
This work is the first to connect morphometric networks with longevity gene expression in a large sample. Unlike neighboring studies that focus on individual aspects of aging (age maps, methylation, sleep, white matter), it brings together structural networks, molecular data, and normative modeling. That makes it possible to see the heterogeneity of aging not as noise but as two biologically distinct paths. The subtypes differ not only in their genes but also in the spatial distribution of connections, pointing to different mechanisms: metabolic/immune versus stress/repair.
The topic could move toward longitudinal studies to test whether people shift from one subtype to another over time. It would also be promising to include functional MRI and other omics (proteomics, metabolomics) for a fuller picture. The analysis could be extended to other networks, and more precise gene mapping methods, such as spatial transcriptomics, could be used.
The results will influence the neurobiology of aging, gerontology, and possibly preclinical diagnosis of neurodegeneration. Understanding the molecular pathways of the subtypes may help in developing personalized interventions.
Verify the reproducibility of the subtypes in independent samples. Study how the subtypes relate to cognitive trajectories in a longitudinal design. Investigate causal links between gene expression and morphometry using Mendelian randomization methods.
The work touches on a fundamental problem in neuroscience: how the heterogeneity of brain aging relates to molecular mechanisms, and how to distinguish normal aging from pathological aging. It also raises the question of compensatory processes that may protect against cognitive decline.
The mean age difference between subtypes was 4.5 years, but it was not statistically significant. At the same time, age explained about 93% of the accuracy in classifying subtypes, which highlights the complex relationship between chronological and biological aging.
The idea of different paths of brain aging echoes Greg Egan's novel Permutation City, where people can choose different trajectories of existence. Here, nature seems to offer two scenarios: one leads to decline, the other to adaptation.
Not published on arXiv. Here are its page, the PDF and the formalization: you can check it yourself, not only read our account.