Epigenetic clock

An epigenetic clock is a type of DNA clock based on measuring natural DNA methylation levels to estimate the biological age of a tissue, cell type or organ. A pre-eminent example for an epigenetic clock is Horvath's clock, which is based on 353 epigenetic markers on the human genome.

History

The strong effect of age on DNA methylation levels has been known since the late 1960s. A vast literature describes sets of CpGs whose DNA methylation levels correlate with age, e.g. Two publications describe age estimators based on DNA methylation levels in either saliva or blood.

Horvath's epigenetic clock was developed by Steve Horvath, a professor of human genetics at the David Geffen School of Medicine at UCLA and of biostatistics at the UCLA Fielding School of Public Health. The scientific article was first published on Oct 21, 2013 in Genome Biology. Horvath spent over 4 years collecting publicly available Illumina DNA methylation data and identifying suitable statistical methods. The personal story behind the discovery was featured in Nature. The age estimator was developed using 8,000 samples from 82 Illumina DNA methylation array datasets, encompassing 51 healthy tissues and cell types. The major innovation of Horvath's epigenetic clock lies in its wide applicability: the same set of 353 CpGs and the same prediction algorithm is used irrespective of the DNA source within the organism, i.e. it does not require any adjustments or offsets. This property allows one to compare the ages of different areas of the human body using the same aging clock.

Relationship to a biological process that causes aging

It is not yet known what exactly is measured by DNA methylation age. Horvath hypothesized that DNA methylation age measures the cumulative effect of an epigenetic maintenance system but details are unknown. The fact that DNA methylation age of blood predicts all-cause mortality in later life strongly suggests that it relates to a process that causes aging. However, it is unlikely that the 353 clock CpGs are special or play a direct causal role in the aging process. Rather, the epigenetic clock captures an emergent property of the epigenome.

Motivation for biological clocks

In general, biological aging clocks and biomarkers of aging are expected to find many uses in biological research since age is a fundamental characteristic of most organisms. Accurate measures of biological age (biological aging clocks) could be useful for

• testing the validity of various theories of biological aging,
• diagnosing various age related diseases and for defining cancer subtypes,
• predicting/prognosticating the onset of various diseases,
• serving as surrogate markers for evaluating therapeutic interventions including rejuvenation approaches,
• studying developmental biology and cell differentiation,
• forensic applications, for example to estimate the age of a suspect based on blood left on a crime scene.

Overall, biological clocks are expected to be useful for studying what causes aging and what can be done against it.

Properties of Horvath's clock

The clock is defined as an age estimation method based on 353 epigenetic markers on the DNA. The 353 markers measure DNA methylation of CpG dinucleotides. Estimated age ("predicted age" in mathematical usage), also referred to as DNA methylation age, has the following properties: first, it is close to zero for embryonic and induced pluripotent stem cells; second, it correlates with cell passage number; third, it gives rise to a highly heritable measure of age acceleration; and, fourth, it is applicable to chimpanzee tissues (which are used as human analogs for biological testing purposes). Organismal growth (and concomitant cell division) leads to a high ticking rate of the epigenetic clock that slows down to a constant ticking rate (linear dependence) after adulthood (age 20). The fact that DNA methylation age of blood predicts all-cause mortality in later life even after adjusting for known risk factors suggests that it relates to a process that causes aging. Similarly, markers of physical and mental fitness are associated with the epigenetic clock (lower abilities associated with age acceleration).

Salient features of Horvath's epigenetic clock include its high accuracy and its applicability to a broad spectrum of tissues and cell types. Since it allows one to contrast the ages of different tissues from the same subject, it can be used to identify tissues that show evidence of accelerated age due to disease.

Statistical approach

The basic approach is to form a weighted average of the 353 clock CpGs, which is then transformed to DNAm age using a calibration function. The calibration function reveals that the epigenetic clock has a high ticking rate until adulthood, after which it slows to a constant ticking rate. Using the training data sets, Horvath used a penalized regression model (Elastic net regularization) to regress a calibrated version of chronological age on 21,369 CpG probes that were present both on the Illumina 450K and 27K platform and had fewer than 10 missing values. DNAm age is defined as estimated ("predicted") age. The elastic net predictor automatically selected 353 CpGs. 193 of the 353 CpGs correlate positively with age while the remaining 160 CpGs correlate negatively with age. R software and a freely available web based tool can be found at the following webpage.

Accuracy

The median error of estimated age is 3.6 years across a wide spectrum of tissues and cell types . The epigenetic clock performs well in heterogeneous tissues (for example, whole blood, peripheral blood mononuclear cells, cerebellar samples, occipital cortex, buccal epithelium, colon, adipose, kidney, liver, lung, saliva, uterine cervix, epidermis, muscle) as well as in individual cell types such as CD4 T cells, CD14 monocytes, glial cells, neurons, immortalized B cells, mesenchymal stromal cells. However, accuracy depends to some extent on the source of the DNA. In particular, DNAm age is higher than chronological age in female breast tissue that is adjacent to breast cancer tissue. Since normal tissue, which is adjacent to other cancer types, does not exhibit a similar age acceleration effect, this finding suggests that normal female breast tissue ages faster than other parts of the body.

Comparison with other biological clocks

The epigenetic clock leads to a chronological age prediction that has a Pearson correlation coefficient of r=0.96 with chronological age (Figure 2 in ). Thus the age correlation is close to its maximum possible correlation value of 1. Other biological clocks are based on a) telomere length, b) p16INK4a expression levels (also known as INK4a/ARF locus), and c) microsatellite mutations. The correlation between chronological age and telomere length is r=−0.51 in women and r=−0.55 in men. The correlation between chronological age and expression levels of p16INK4a in T cells is r=0.56. p16INK4a expression levels only relate to age in T cells, a type of white blood cells. The microsatellite clock measures not chronological age but age in terms of elapsed cell divisions within a tissue.

Age acceleration according to Horvath's clock

By contrasting DNA methylation age (estimated age) with chronological age, one can define measures of age acceleration. Age acceleration can be defined as the difference between DNA methylation age and chronological age. Alternatively, it can be defined as the residual that results from regressing DNAm age on chronological age. The latter measure is attractive because it does not correlate with chronological age. A positive/negative value of epigenetic age acceleration suggests that the underlying tissue ages faster/slower than expected.

Genetic studies of epigenetic age acceleration

The broad sense heritability (defined via Falconer's formula) of age acceleration of blood from older subjects is around 40% but it appears to be much higher in newborns. Similarly, the age acceleration of brain tissue (prefrontal cortex) was found to be 41% in older subjects. Genome-wide association studies of cerebellar age acceleration have identified several SNPs at a genomewide significance level. Gene and SNP sets found by genome-wide association analysis of epigenetic age acceleration exhibit significant overlap with those of Alzheimer’s disease, age-related macular degeneration, and Parkinson’s disease.

Cancer tissue

Cancer tissues show both positive and negative age acceleration effects. For most tumor types, no significant relationship can be observed between age acceleration and tumor morphology (grade/stage). On average, cancer tissues with mutated TP53 have a lower age acceleration than those without it. Further, cancer tissues with high age acceleration tend to have fewer somatic mutations than those with low age acceleration. Age acceleration is highly related to various genomic aberrations in cancer tissues. Somatic mutations in estrogen receptors or progesterone receptors are associated with accelerated DNAm age in breast cancer. Colorectal cancer samples with a BRAF (V600E) mutation or promoter hypermethylation of the mismatch repair gene MLH1 are associated with an increased age acceleration. Age acceleration in glioblastoma multiforme samples is highly significantly associated with certain mutations in H3F3A. One study suggests that the epigenetic age of blood tissue may be prognostic of lung cancer incidence.

Obesity

The epigenetic clock was used to study the relationship between high body mass index (BMI) and the DNA methylation ages of human blood, liver, muscle and adipose tissue. A significant correlation (r=0.42) between BMI and epigenetic age acceleration could only be observed for liver.

Trisomy 21 (Down syndrome)

Down Syndrome (DS) entails an increased risk of many chronic diseases that are typically associated with older age. The clinical manifestations of accelerated aging suggest that trisomy 21 increases the biological age of tissues, but molecular evidence for this hypothesis has been sparse. According to the epigenetic clock, trisomy 21 significantly increases the age of blood and brain tissue (on average by 6.6 years).

Alzheimer's disease related neuropathology

Epigenetic age acceleration of the human prefrontal cortex was found to be correlated with several neuropathological measurements that play a role in Alzheimer's disease including diffuse plaques (r=0.12, p=0.0015), neuritic plaques (r=0.11, p=0.0036), and amyloid load (r=0.091, p=0.016). Further, it was found to be associated with a decline in global cognitive functioning (p=0.009), episodic memory (p=0.009) and working memory (p=0.011) among individuals with Alzheimer's disease. However, no relationship was found between a clinical diagnosis of AD and epigenetic age acceleration of the prefrontal cortex. The epigenetic age of blood relates to cognitive functioning in the elderly. Overall, these results strongly suggest that the epigenetic clock lends itself for measuring the biological age of the brain.

Cerebellum ages slowly

It has been difficult to identify tissues that seem to evade aging due to the lack of biomarkers of tissue age that allow one to contrast compare the ages of different tissues. An application of epigenetic clock to 30 anatomic sites from six centenarians and younger subjects revealed that the cerebellum ages slowly: it is about 15 years younger than expected in a centenarian. This finding might explain why the cerebellum exhibits fewer neuropathological hallmarks of age related dementias compared to other brain regions. In younger subjects (e.g. younger than 70), brain regions and brain cells appear to have roughly the same age. Several SNPs and genes have been identified that relate to the epigenetic age of the cerebellum

Centenarians age slowly

The offspring of semi-supercentenarians (subjects who reached an age of 105–109 years) have a lower epigenetic age than age-matched controls (age difference=5.1 years in peripheral blood mononuclear cells) and centenarians are younger (8.6 years) than expected based on their chronological age.

HIV infection

Infection with the Human Immunodeficiency Virus-1 (HIV) is associated with clinical symptoms of accelerated aging, as evidenced by increased incidence and diversity of age-related illnesses at relatively young ages. But it has been difficult to detect an accelerated aging effect on a molecular level. An epigenetic clock analysis of human DNA from HIV+ subjects and controls detected a significant age acceleration effect in brain (7.4 years) and blood (5.2 years) tissue due to HIV-1 infection.

Parkinson's disease

A large scale study suggests that the blood of Parkinson's disease subjects exhibits (relatively weak) accelerated aging effects.

Developmental disorder: syndrome X

Children with a very rare disorder known as syndrome X maintain the façade of persistent toddler-like features while aging from birth to adulthood. Since the physical development of these children is dramatically delayed, these children appear to be a toddler or at best a preschooler. According to an epigenetic clock analysis, blood tissue from syndrome X cases is not younger than expected.

Cellular senescence versus epigenetic aging

A confounding aspect of biological aging is the nature and role of senescent cells. It is unclear whether the three major types of cellular senescence, namely replicative senescence, oncogene-induced senescence and DNA damage-induced senescence are descriptions of the same phenomenon instigated by different sources, or if each of these is distinct, and how they are associated with epigenetic aging. Induction of replicative senescence (RS) and oncogene-induced senescence (OIS) were found to be accompanied by epigenetic aging of primary cells but senescence induced by DNA damage was not, even though RS and OIS activate the cellular DNA damage response pathway. These results highlight the independence of cellular senescence from epigenetic aging. Consistent with this, telomerase-immortalised cells continued to age (according to the epigenetic clock) without having been treated with any senescence inducers or DNA-damaging agents, re-affirming the independence of the process of epigenetic ageing from telomeres, cellular senescence, and the DNA damage response pathway. Although the uncoupling of senescence from cellular aging appears at first sight to be inconsistent with the fact that senescent cells contribute to the physical manifestation of organism ageing, as demonstrated by Baker et al., where removal of senescent cells slowed down aging. However, the epigenetic clock analysis of senescence suggests that cellular senescence is a state that cells are forced into as a result of external pressures such as DNA damage, ectopic oncogene expression and exhaustive proliferation of cells to replenish those eliminated by external/environmental factors. These senescent cells, in sufficient numbers, will probably cause the deterioration of tissues, which is interpreted as organism ageing. However, at the cellular level, aging, as measured by the epigenetic clock, is distinct from senescence. It is an intrinsic mechanism that exists from the birth of the cell and continues. This implies that if cells are not shunted into senescence by the external pressures described above, they would still continue to age. This is consistent with the fact that mice with naturally long telomeres still age and eventually die even though their telomere lengths are far longer than the critical limit, and they age prematurely when their telomeres are forcibly shortened, due to replicative senescence. Therefore, cellular senescence is a route by which cells exit prematurely from the natural course of cellular aging.

Other age estimators based on DNA methylation levels

Several other age estimators have been described in the literature.

1) Weidner et al. (2014) describe an age estimator for DNA from blood that uses only three CpG sites of genes hardly affected by aging (cg25809905 in integrin, alpha 2b (ITGA2B); cg02228185 in aspartoacylase (ASPA) and cg17861230 in phosphodiesterase 4C, cAMP specific (PDE4C)). The age estimator by Weidener et al. (2014) applies only to blood. Even in blood this sparse estimator is far less accurate than Horvath's epigenetic clock (Horvath 2014) when applied to data generated by the Illumina 27K or 450K platforms. But the sparse estimator was developed for pyrosequencing data and is highly cost effective.

2) Hannum et al. (2013) report several age estimators: one for each tissue type. Each of these estimators requires covariate information (e.g. gender, body mass index, batch). The authors mention that each tissue led to a clear linear offset (intercept and slope). Therefore, the authors had to adjust the blood based age estimator for each tissue type using a linear model. When the Hannum estimator is applied to other tissues, it leads to a high error (due to poor calibration) as can be seen from Figure 4A in Hannum et al. (2013). Hannum et al. adjusted their blood based age estimator (by adjusting the slope and the intercept term) in order to apply it to other tissue types. Since this adjustment step removes differences between tissue, the blood based estimator from Hannum et al. cannot be used to compare the ages of different tissues/organs. In contrast, a salient characteristic of the epigenetic clock is that one does not have to carry out such a calibration step: it always uses the same CpGs and the same coefficient values. Therefore, Horvath's epigenetic clock can be used to compare the ages of different tissues/cells/organs from the same individual. While the age estimators from Hannum et al. cannot be used to compare the ages of different normal tissues, they can be used to compare the age of a cancerous tissue with that of a corresponding normal (non-cancerous) tissue. Hannum et al. reported pronounced age acceleration effects in all cancers. In contrast, Horvath's epigenetic clock reveals that some cancer types (e.g. triple negative breast cancers or uterine corpus endometrial carcinoma) exhibit negative age acceleration, i.e. cancer tissue can be much younger than expected. An important difference relates to additional covariates. Hannum's age estimators make use of covariates such as gender, body mass index, diabetes status, ethnicity, and batch. Since new data involve different batches, one cannot apply it directly to new data. However, the authors present coefficient values for their CpGs in Supplementary Tables which can be used to define an aggregate measure that tends to be strongly correlated with chronological age but may be poorly calibrated (i.e. lead to high errors).

Comparison of the 3 age predictors described in A) Horvath (2013),[39] B) Hannum (2013),[40] and C) Weidener (2014),[41] respectively. The x-axis depicts the chronological age in years whereas the y-axis shows the predicted age. The solid black line corresponds to y=x. These results were generated in an independent blood methylation data set that was not used in the construction of these predictors (data generated in Nov 2014).