Molecular and phenotypic biomarkers of aging

DNA and chromosomes

Telomeres. Telomeres are ribonucleoprotein complexes at the end of chromosomes and become shorter after each replication, as telomerase, the enzyme responsible for its replication, is not regularly expressed in somatic cells⁴. The length of telomeres in leukocytes has been associated with aging and life span⁵ as well as age-related diseases, such as cardiovascular diseases^6,7, cancer⁸, and neurological disorders⁹.

DNA repair. The link between DNA damage and repair has been implicated in aging by the accumulation of senescent cells¹⁰ or genomic rearrangements¹¹. More recently, this link was directly demonstrated, and controlled induction of DNA double-strand breaks in mouse liver inducing aging pathologies and gene expression was shown¹². Immunohistochemistry of γ-H2A.X is an established quantitative biomarker of aging because H2A.X is a variant of the H2A protein family, and phosphorylated H2A.X, γ-H2A.X, is an initial and essential component of DNA damage foci and therefore a reliable marker of the extent of DNA damage^13–15. Serum markers of DNA damage, including CRAMP, EF-1a, stathmin, N-acetyl-glucosaminidase, and chitinase, have also been established¹⁶. Of note, the dermal fibroblasts from centenarian donors were shown to be less sensitive to H₂O₂-induced DNA damage than fibroblasts from young and old donors¹⁷. Such ex vivo experiments could also be a potential biomarker of aging.

Epigenetic modifications. Age-related changes in DNA methylation patterns, notably as measured by the epigenetic clock, are among the best-studied aging biomarkers^18–20. Analysis of methylation profiles in the blood found that only three CpG sites could predict age with a mean absolute deviation from chronological age of less than 5 years²¹. The association between age and DNA methylation can be extended to age-associated diseases, such as diabetes²². For a full review of the epigenetic regulation of aging, see Sen et al.²³.

RNA and transcriptome

Transcriptome profiles. With rapid progress in single-cell RNA sequencing (RNA-seq) technology, it has begun to be applied to the study of biomarkers of aging. Lu et al. have recently shown that cell-to-cell expression variation, as measured by single-cell RNA-seq of high-dimensional flow cytometry sorted T cells, is associated with aging and disease susceptibility²⁴.

A recent study used whole-blood gene expression profiles from 14,983 individuals to identify 1,497 genes with age-dependent differential expression and then used them to calculate the ‘transcriptomic age’ of an individual, suggesting that transcriptome signatures can be used to measure aging²⁵.

Non-coding RNAs. MicroRNAs (miRNAs) are a class of small (21- to 23-nucleotide) non-coding RNAs that, through base-pairing mechanisms, regulate a broad range of biological processes, including metabolism²⁶ and aging²⁷. Among them, circulating miRNAs can be stable in plasma by residing in exosomes or binding to protein or lipoprotein factors, thus making them easy-to-access biomarkers. miR-34a was the first observed circulating miRNA with an altered expression pattern during mouse aging²⁸. Its expression is found to correlate with age-related hearing loss in mice and humans²⁹. miR-21 was defined as an inflammatory biomarker in a study of 365 miRNAs in the plasma of healthy and old humans³⁰. miR-151a-3p, miR-181a-5p, and miR-1248 are reported to be significantly decreased with age in humans, in which all three miRNAs also show indications of associations with inflammation³¹. miR-126-3p has been found to be positively correlated with age in 136 healthy subjects from 20 to 90 years of age³². Through expression of GFP driven by miRNA promoters, Pincus et al. found that levels of mir-71, mir-246, and mir-239 in early adulthood vary across individuals and are predictive of life span³³. A recent review²⁷ summarized the associations of other types of circulating non-coding small RNAs, such as tRNA and YRNA.

Long non-coding RNAs (lncRNAs) are a heterogeneous class of non-coding RNAs which are defined as transcripts longer than 200 nucleotides and devoid of evident open reading frames³⁴. Two recent reviews summarize the role of lncRNAs in aging^35,36. The diverse functional mechanisms of lncRNA are beyond the scope of this review, and readers may consult a recent review on this topic³⁷; here, we list lncRNAs that function in aging. The lncRNA MIR31HG was identified to be upregulated in oncogene-induced senescence and required for polycomb group–mediated repression of the INK4A locus³⁸. Downregulation of lncRNA AK156230 occurs in replicative senescence and its knockdown in mouse embryonic fibroblasts induces senescence through dysregulation of autophagy and cell cycle pathways, as shown by expression profiles³⁹. Meg3 is upregulated during cardiovascular aging as well as in senescent human umbilical venous endothelial cells⁴⁰. As most of the lncRNAs studies have been anecdotal, high-throughput lncRNA studies, such as CRISPR-Cas9 screen of functional lncRNAs⁴¹, will be a useful future step toward understanding lncRNA functions in the aging process.

Metabolism

That dietary restriction is the most conserved means to extend life span and health span from yeast to mammals⁴² points to a pivotal role of metabolism in aging regulation and to the potential for metabolic factors to be biomarkers.

Nutrient sensing. The insulin/insulin-like growth factor 1 (IGF-1) signaling (IIS) pathway, which participates in glucose sensing, is the earliest discovered and the most well-known pathway to antagonize longevity. Paradoxically, IGF-1 declines in wild-type mice or mouse models of premature aging whereas attenuating IIS activity extends life span⁴³. Such observations led to the potential inclusion of IIS pathway members, such as growth hormone and IGF-1, as biomarkers of aging^44,45.

The mechanistic target of rapamycin (mTOR) protein senses high amino acid concentrations. Inhibition of mTOR can extend life span⁴⁶. Unlike the IIS pathway, mTOR activity increases with age in the ovarian surface epithelium of aged human and mouse ovaries, which contributes to pathological changes⁴⁷. Phosphorylated S6 ribosomal protein (p-S6RP, or pS6) is a downstream target and also a known marker of active mTOR signaling^47,48, which is a potential biomarker of aging as indicated in the research of aged ovaries⁴⁷.

In contrast to IIS and mTOR function, 5′-adenosine monophosphate (AMP)–activated protein kinase (AMPK) and sirtuins sense nutrient scarcity instead of abundance. AMPK detects high AMP levels whereas sirtuins are sensors of high NAD⁺ levels, and both mark low-energy states. The upregulation of AMPK activity by metformin, a drug for type II diabetes, could mimic some of the benefits of caloric restriction, and metformin extends life span in male mice⁴⁹. AMPK is upregulated with age in skeletal muscles⁵⁰.

Sirtuins have the ability to directly link cellular metabolic signaling (reflected by NAD⁺) to protein post-translational modifications through a chemical reaction (deacetylation of lysine). During aging, NAD⁺ is reduced⁵¹ and sirtuins are downregulated^52,53. An analysis of primary human dermal fibroblasts found that SIRT1 and SIRT6 are downregulated through passaging⁵⁴. Similarly, levels of SIRT1, SIRT3, and SIRT6 detected by Western blotting showed significant decrease in ovaries of aged mice⁵⁵. In human peripheral blood mononuclear cells, SIRT2 also decreases with age⁵⁶.

Protein metabolism. Protein carbamylation is one of the non-enzymatic post-translational modifications which occur throughout the whole life span of an organism, leading to tissue accumulation of carbamylated proteins⁵⁷. It is considered a hallmark of molecular aging and is related to aging-related diseases, such as cardiovascular disease⁵⁸.

Advanced glycation end products (AGEs) are a heterogeneous group of bioactive molecules that are formed by non-enzymatic glycation of proteins, lipids, and nucleic acids⁵⁹. Accumulation of AGEs in aging tissues leads to inflammation⁶⁰, apoptosis⁶¹, obesity⁶², and other age-related disorders⁶³. AGEs can be detected via high-performance liquid chromatography, gas chromatography-mass spectrometry, and immunochemical techniques⁶⁴. N-glycans are a class of glycoproteins with sugar chains bonded to the amide nitrogen of asparagine. The spectrum of N-linked glycans (the N-glycome) can now be investigated because of the development of high-throughput methods. The accumulation of N-linked glycation at Asn297 of the Fc portion of IgG (IgG-G0) can contribute to low-grade pro-inflammatory status in aging⁶⁵.

Lipid metabolism. Triglycerides are found to increase monotonously with age and thus could be a biomarker of aging⁶⁶. Studies of serum samples by shotgun lipidomics found that phospho/sphingolipids are putative markers, and biological modulators, of healthy aging⁶⁷. However, the design of these studies is questionable in that they have a group of elderly individuals as a ‘not healthy aging control’ and compare them with the ‘successful aging’ centenarian group^67,68, but the two groups are obviously of very different ages. Therefore, it is not clear whether it was the age difference or the success of healthy aging that contributed to the differences in lipidomics.

Oxidative stress and mitochondria

Biomarkers of oxidative stress have long been regarded as a class of aging biomarkers. The products of oxidative damage to proteins include o-tyrosine, 3-chlorotyrosine, and 3-nitrotyrosine. 8-iso prostaglandin F_2α is a biomarker for phospholipid damage. 8-hydroxy-2′-deoxyguanosine and 8-hydroxyguanosine are produced by the oxidative damage of nucleic acids⁶⁹. The concentration of these biomarkers in body fluids can be detected via high-performance liquid chromatography and mass spectrometry. Shen et al. engineered a circularly permuted yellow fluorescent protein (cpYFP) expressed in Caenorhabditis elegans mitochondrial matrix as a sensor of oxidative stress and metabolic changes; the authors found that adult day 3 mitochondrial cpYFP flash frequency is a good predictor of C. elegans life span under different genetic, environmental, and stochastic conditions⁷⁰.

Although free radicals, the source of oxidative stress, are mainly produced in mitochondria, dysfunctional mitochondria can contribute to aging independently of reactive oxygen species. To measure mitochondria function, blood- and-muscle based respirometric profiling strategies are available, and the association of this potential reporter with bioenergetic capacity of other tissues⁷¹ or phenotypes, such as gait speed⁷², has been investigated. Extracellular mitochondria components can function as damage-associated molecular pattern molecules (DAMPs) (see also “Inflammation and intercellular communication”) and these induce neuroinflammation when injected in mouse hippocampus⁷³.

Cell senescence

In mitotic tissues, the gradual accumulation of senescent cells is thought be one of the causal factors of aging^74–76. Thus, the biomarkers of cell senescence can also be used as markers. Such biomarkers have been summarized in recent reviews^77,78. The most widely used marker is senescence-associated β-galactosidase (SAβ-gal)⁷⁹ and p16^INK4A80,81. SAβ-gal reflects increased lysosomal mass⁸² but can yield false positives because of its low specificity⁸³. SAβ-gal is a cell damage marker, and p16^INK4A is required to induce, and is indicative of, permanent cell cycle arrest⁸¹.

Other senescent cell markers include activated and persistent DNA-damage response (see “DNA repair”), telomere shortening and dysfunction (see “Telomere”), and senescence-associated secretory phenotype (SASP) (see “Inflammation and intercellular communication”).

Inflammation and intercellular communication

SASP is a consequence of cell senescence and may occur in cells that, though undergoing cell cycle arrest, are still metabolically active and secrete proteins. SASP functions in an autocrine/paracrine manner^84,85. The major components of SASP factors are soluble signaling factors, including interleukins, chemokines, and growth factors. Proteins that are associated with the SASP, such as interleukin-6, tumor necrosis factor-alpha, monocyte chemoattractant protein-1, matrix metalloproteinases, and IGF binding proteins, increase in multiple tissues with chronological aging and occur in conjunction with sterile inflammation⁸⁶. Comprehensive catalogs of SASP also include secreted proteases and secreted insoluble proteins/extracellular matrix components and are summarized by Coppé et al.⁸⁷ and the Reactome database (http://www.reactome.org/content/detail/R-HSA-2559582).

The DAMPs, such as heat shock proteins, histones, high-mobility group box 1, and S100, compose a class of molecules released after injury or cellular death⁸⁸ and mediate immune response. The association between DAMPs and other hallmarks of aging has been reviewed by Huang et al.⁸⁹.

Integration of aging biomarkers

Biomarkers of aging can be used to predict the physiological age, which reflects their state of health, via statistics and machine learning algorithms. A single class of biomarkers, which is intrinsically a matrix of features, can be used in the prediction. DNA methylation was used to predict age with an error of about 3.6 years using 8,000 samples⁹⁴. 3D facial images have also been used to predict age with a mean deviation of 6 years⁹².

Integration of multiple biomarkers can be even more powerful. The Dunedin Study⁹¹ has focused on middle-aged people and used different measurements (telomere lengths, epigenetic clocks, and clinical biomarker composites) and compared their performance in predicting health status, as measured by physical functionality, cognitive decline, and subjective signs of aging. The three types of measurements in this study do not correlate with each other, suggesting that there is no single index of biological age. Therefore, another approach is to use statistic distance, D_M^95,96, to assess the degree of deviation of an individual’s biomarker profile from the reference population. D_M is the Mahalanobis distance⁹⁷ of multi-variants (in the simplest case, when all the variants are uncorrelated, this distance is the sum of the absolute values of z-scores), and is proven to be insensitive to biomarker choice across 44 available markers and to be generalizable with multiple marker variants. Recently, a modular ensemble of 21 deep neural networks was used to predict age by using measurements from basic blood tests by training over 60,000 samples, which revealed the five most important blood markers for predicting human chronological age: albumin, glucose, alkaline phosphatase, urea, and erythrocytes⁹⁸.