Prevention of aging-related diseases (ARDs) is paramount in the current era of population aging. Delaying the aging process can potentially extend healthspan and reduce ARDs burden. It has long been observed that the pace of aging varies from person to person. This highlights a key concept that due to underlying biological mechanisms, biological age at an individual level can be separated from chronological age.1–3 Biological age, defined by clinical and molecular biomarkers, indeed predicts overall mortality and ARDs, sometimes even better than chronological age.4 Similarly, modifiable risk factors predict mortality reasonably well over time, even though independent genetic factors predict mortality only modestly.5 Investigating biological age can help identify individuals at higher risk of disease and death, before clinical manifestation of disease. Biomarkers of biological age have great potential in evaluating healthy-aging intervention programs, selecting suitable candidates for clinical trials, as well as indicating levels of personal health, and predicting risk of aging-related diseases. These applications may provide insight on how to extend both lifespan and healthspan and cope with the burden of ARDs worldwide.
The concept of biological age has been constructed and validated in large human cohort studies for a panel of physiological biomarkers,6–8 differentially methylated sites in DNA (DNAm age),9,10 circulatory metabolites (metabo-age), the levels of messenger-RNAs 11 and microRNAs (miRNAs) in both whole blood12 and plasma samples (Wu J.W. et al., submitted). Physiological biomarkers tend to be stronger predictors of mortality and aging-related morbidity outcomes than molecular biological age measures,4 indicating the complexity of human aging processes. Most of these measures rely on the association between composite biomarkers and chronological age and might be imperfect because of the practice of minimizing the deviation from chronological age through regression and not estimating this entity of interest empirically.4 The main knowledge gap is to what extent the deviation of biological age measure from chronological age predicts risks of mortality and major ARDs onsets, especially dementia, in advanced age populations. We sought to ask the following question: is it possible to build a physiological marker-based biological age algorithm in the absence of a brain marker that predicts elevated risk of dementia for a given chronological age?
In a previous study13, we have validated the approach of Klemera and Doubal14 and showed that over a median follow-up of 11 years, biological age at baseline was superior to chronological age and traditional biomarkers, in predicting mortality, morbidity and onset of specific diseases such as stroke and cancer. Furthermore, compared to chronological age alone or combined with the individual biomarkers, adding a brain biomarker for neurological degeneration (plasma NfL, total-tau, amyloid beta-40 and − 42) further improved the association of biological age with dementia, including Alzheimer disease (AD). On the other hand, the Klemera and Doubal-based biological age did not predict mortality or morbidity when adjusting for chronological age, and had stronger associations with chronological age than with risk of dementia. It became evident that a completely new model for biological age, beyond chronological age, was needed to improve both the association with prediction of mortality and dementia, including AD.
In this study, we aimed to improve biological age algorithms through a population-risk-based framework. Recent work by Levine et al.,4 has introduced a novel measure of ‘phenotypic age’, developed and cross-validated on NHANES III (n = 9,926). A Gompertz proportional hazards regression14 was applied to account for the hazard of mortality when selecting clinical biomarkers in the training dataset. This approach has the advantage of capturing the incremental risk of death and morbidities due to accelerated biological aging, whereas traditional biological age algorithms largely aimed to model chronological age. However, Levine’s “phenotypic age” algorithm was unable to establish a significant link between accelerated aging and elevated risk for AD. We hypothesized that this caveat was not due to the population-risk-framework itself, but because the “phenotypic age” was trained in the NHANES data, which has a relatively younger age distribution, than is common for neurological outcomes. Furthermore, we investigated whether by including markers of neurodegeneration, we could improve the prediction of neurological outcomes in advanced-aged cohorts.
Our approach to improve the biological age models comprised of two steps. First, we validated Levine’s “phenotypic age” algorithm in the Rotterdam Study (n = 1,930). Second, we developed and cross-validated new biological age algorithms in the Rotterdam Study using the same Gompertz proportional hazards regression framework plus neurodegenerative markers (NfL, total-tau, amyloid beta − 40 and − 42). MicroRNAs (miRNAs) age signature was found to be not only predictive of the actual age, but useful as a biomarker of all-cause mortality in both whole blood12 and plasma samples (Wu J.W. et al., submitted). We assessed the deviation of biological age from chronological age through the plasma miRNA expression signature.