In this open access paper, evidence is presented for neuroimaging to be the basis for a biomarker of aging that is as good as the best of present candidate DNA methylation biomarkers. This is most interesting, though I suspect that there might be a higher chance that it will prove unhelpful as a way to assess the quality and effectiveness of potential rejuvenation therapies. That process of assessment is the reason why there is at present a considerable interest in the development of biomarkers that reflect chronological or biological age. Today the only practical approach to assessing candidate rejuvenation therapies is to provide the treatment and then wait and see: this is prohibitively expensive in humans, and too expensive for most research groups even in mice. That expense - and the time required to run life span studies - is holding the field back. If potential approaches to rejuvenation could be assessed quickly, that new capability would considerably speed up the pace of progress.
Why do I think that there is a great risk that neuroimaging might fail to be helpful in assessing rejuvenation therapies? Because most of the present proposed candidate treatments will change cellular biochemistry, remove problem cells, or repair forms of cellular damage, but will not repair the secondary outcomes of aging in the brain, such as white matter hyperintensities and other outcomes of the structural failure of blood vessels. These therapies would be expected to alter DNA methylation patterns, however, which are most likely more a reaction to cellular dysfunction and low-level molecular damage than a reaction to larger-scale structural changes. Still, this is an opinion offered in absence of evidence; we shall see how things turn out.
The search for robust, reliable and valid biomarkers of the ageing process is a key goal for gerontological science. Such tools should enable the quantification of individual differences in underlying biological ageing. This could have great utility for mapping personalised ageing trajectories, for predicting risk of future age-related deterioration and disease and for evaluating potential treatments aimed at improving healthspan or even slowing ageing itself. Given the multi-faceted nature of biological ageing, numerous potential candidate biomarkers have been proposed. These can be anthropometric, physiological or blood-based; indexing immune function, epigenetic signatures, gene expression profiles, physical capacity or body composition. To improve on individual predictors of biological age, panels combining multiple markers have also been proposed. While many of these approaches are highly promising, the results have yet to be translated into clinical practice.
The criteria most commonly used for assessing the appropriateness of ageing biomarkers is how strongly they correlate with chronological age in healthy people. In addition, thanks to the increasing use of machine learning, the accuracy with which chronological age can be predicted using multivariate biological data is also a useful indicator of potential biomarker value. Aligned with this, an independent line of research has emerged from the field of neuroscience. Using neuroimaging data, principally magnetic resonance imaging (MRI) brain scans, chronological age can be predicted accurately in a machine-learning framework. This neuroimaging-derived brain-age model is based on data from over 2000 healthy adults and shows excellent test-retest reliability. This presents the intriguing possibility that in-vivo measurements of brain volume could be used as an alternative ageing biomarker.
It is well-known that ageing affects the brain, both in terms of outward behavioural changes and cognitive decline, alongside alterations to the brain's biophysical structure and cellular and molecular functioning. Using measures of brain volume derived from T1-weighted structural MRI, assumed to reflect grey and white matter atrophy, high levels of age prediction accuracy have been consistently achieved. For example, our work found a mean/median absolute error of age prediction of 4.2/3.4 years, with a correlation between age and brain-predicted age of r = 0.96. This is comparable to or better than leading biological age prediction models, for example using DNA methylation status (r = 0.96, median absolute error = 3.6 years) or a panel of blood chemistry markers (r = 0.91, mean absolute error = 5.6 years).
Given the published data on neuroimaging-derived brain-age, it is worth considering its qualification against a set of consensus ageing biomarker criteria. Paraphrasing from the American Federation for Aging Research recommendations, an ageing biomarker must: 1) Predict the rate of ageing (i.e., estimate where a person is in their total life span); 2) Measure a basic process that underlies ageing, not the effects of disease; 3) Be able to be tested repeatedly without causing harm; 4) Work in humans and laboratory animals. Based on the above evidence regarding prediction of survival, neuroimaging-derived brain-age meets criteria #1. Given the accuracy of age prediction and the fact that brain atrophy occurs in the context of non-pathological ageing, this satisfies criteria #2. As a non-invasive imaging technique, T1-weighted MRI meets criteria #3. Finally, the accuracy of this technique in non-human primates has been recently reported, suggesting that it appropriately meets criteria #4.
While perhaps the major caveat regarding the use of neuroimaging in this context is the cost and potential logistics, projects like the UK Biobank imaging study show that collecting neuroimaging data on an extremely large scale are becoming increasingly feasible. It is timely for a marriage of neuroscience and biogerontology, and approaches that combine the most complementary information on the ageing human body will have the greatest utility in developing effective ageing biomarkers.