Aging is the primary risk factor for many prevalent diseases (Niccoli and Partridge, 2012). Indeed, geroscientists have begun to treat aging itself as a preventable cause of many aging-related diseases (Barzilai et al., 2018, Campisi et al., 2019, Matt et al., 2015). The geroscience hypothesis defines aging as the gradual, progressive, correlated biological decline of the entire body over decades (Kennedy et al., 2014; Vadim, 2016). Crucially, individuals of the same chronological age often vary in their rate of biological aging (Belsky et al., 2015). Despite its importance, there is still no agreed-upon measure of biological aging (Ferrucci et al., 2020). To address this, researchers have begun to use DNA methylation to quantify aging. DNA methylation is a highly age-sensitive epigenetic process wherein methyl groups are selectively added to DNA molecules to affect gene transcription. Attempts to develop measures of aging have often used blood DNA methylation because blood is the most widely profiled source of DNA and blood DNA methylation is a biological substrate that is sensitive to age-related changes across the body (Horvath and Raj, 2018, Levine et al., 2015).

In the past decade, several algorithms have been developed to estimate biological aging using DNA methylation (Rutledge et al., 2022). These algorithms are typically referred to as ‘epigenetic clocks.’ The first generation of epigenetic clocks was trained largely on chronological age (Hannum et al., 2013, Horvath, 2013). Subsequently, a second generation of clocks was trained on cross-sectional measures of current health that predict mortality such as C-reactive protein levels, white blood cell count, and smoking packs-per-year (Levine et al., 2018, Lu et al., 2019). This second generation includes clocks such as PhenoAge and GrimAge. Tests for associations between first- and second-generation epigenetic clocks and brain structure have yielded mixed results. Some studies reported large positive associations between accelerated epigenetic age and reduced total brain volume (Hillary et al., 2021), reduced hippocampal volume (Davis et al., 2017), increased white matter hyperintensities (Hillary et al., 2021), and thinner cortex (Proskovec et al., 2020). However, other studies found only small positive associations between accelerated epigenetic age and lower hippocampal volume (Milicic et al., 2022) and increased white matter hyperintensities (Raina et al., 2017). Still others reported null associations between accelerated epigenetic age and cortical thickness, cortical surface area, and cortical volume (Cheong et al., 2022), or even found that accelerated epigenetic age is associated with preserved white matter microstructure, specifically increased fractional anisotropy and reduced mean diffusivity (Chouliaras et al., 2018).

In contrast to these earlier epigenetic clocks, we recently developed a third-generation DNA methylation-based measure that is unique in estimating a person’s rate of biological aging. The DunedinPACE (Pace of Aging Calculated from the Epigenome) algorithm was developed by first measuring people’s rate of physiological change over time and then identifying the methylation patterns that optimally captured individual differences in their age-related decline (Belsky et al., 2022). Specifically, age-related decline was measured over ages 26, 32, 38, and 45 years in 19 biomarkers of the cardiovascular, metabolic, renal, immune, dental, and pulmonary organ systems among healthy midlife individuals of the same chronological age participating in the Dunedin Study (Elliott et al., 2021b). Methylation patterns at the end of the 20-year observation period were then identified that estimated how fast each participants’ multi-organ decline occurred during the 20 years leading up to the point of measurement (Belsky et al., 2022). Thus, DunedinPACE was designed to capture methylation patterns reflecting individual differences in the rate of age-related multi-organ decline and it has been robustly associated with multimorbidity and mortality (Belsky et al., 2022, Bernabeu et al., 2023, Faul et al., 2023, Kuiper et al., 2023; Lachlan et al., n.d.; McMurran et al., 2023). Importantly, DunedinPACE allows for readily measuring the pace of aging in individuals who lack data to implement longitudinal physiological profiling.

Of note, DunedinPACE was not trained on any measures of central nervous system decline. Thus, it is not clear whether DunedinPACE is associated with brain structure. Recent studies have used longitudinal multi-organ measurements to demonstrate that aging ‘below the neck’ is related to aging of the brain (Elliott et al., 2021a, Elliott et al., 2021b, Tian et al., 2023). Additionally, prior work has found some correspondence between DNA methylation patterns in blood and brain tissue (Horvath et al., 2012), suggesting that DNA methylation in blood is a promising surrogate for relating the aging of body and brain (but see (Shireby et al., 2020). However, it is unknown whether a measurement from a blood sample at a single timepoint also captures the association between brain structure and body aging. DunedinPACE has been associated with cognitive and clinical measures that are thought to index health of the central nervous system. For instance, faster DunedinPACE has been associated with more rapid cognitive decline (Belsky et al., 2022, Reed et al., 2022), mild cognitive impairment, and dementia (Sugden et al., 2022). These findings suggest that DunedinPACE indexes typical decline of cognitive ability during typical aging as well as in neurodegenerative illness. Associations between DunedinPACE and these key cognitive and clinical phenotypes suggest that DunedinPACE may also be associated with brain structure, though this question has not been formally tested.

We examined associations between DunedinPACE and multiple measures of brain structure across three large datasets spanning mid- to late-life: the Dunedin Study (N=770, mean age=45 years), the Framingham Heart Study Offspring Cohort (FHS-OC; N=903, mean age=63.76 years), and the Alzheimer’s Disease Neuroimaging Initiative (ADNI; N observations=1,707, N individuals=649; mean age=75.41 years). Across all three datasets (N observations=3,380; N individuals=2,322), we tested for associations between DunedinPACE and measures of brain structure derived from high-resolution magnetic resonance imaging (MRI; Fig. 1) including: total brain volume (TBV), hippocampal grey matter volume (HC), white matter hypointensity volume (WMHypo), mean cortical thickness (CT), and total cortical surface area (SA). When available in the Dunedin Study and ADNI, we also tested for associations between DunedinPACE and white matter hyperintensity volume (WMHyper). In addition, we leveraged the longitudinal nature of ADNI to test for associations between DunedinPACE and age-related changes in brain structure. For comparison, we also tested associations between first- and second-generation epigenetic clocks and brain structure.