49 - 62: How Does the Body Know How Old It Is? Introducing the Epigenetic Clock Hypothesis

To many readers, it will seem like a strange idea that aging proceeds under control of a biological clock. If you think in terms of the body accumulating damage over time, then there is no master clock, no separate record of the state of the body's age - there are only the various parts of the body in various states of disrepair at any given time.

A recurring theme from research in animal models over the past two decades is that aging is not a passive accumulation of cellular damage, but is actively regulated at the level of the whole organism. However, even this regulation need not imply the existence of a biological clock; it may be merely that the activity of the body's repair mechanisms is modulated by an internal calculation based on external cues, so that the damage might accumulate at a variable rate without the body having to follow a scheduled program.

One motivation for thinking about a clock is a kind of Pascal's Wager¹ for the gerontologist: If there is an aging clock, then it suggests a convenient target for medical intervention that will have a highly leveraged effect on aging and disease. Speculation about the existence of an aging clock is interesting because, if such a thing does exist, it might be possible not just to slow the rate of aging, but to act directly on the clock, to set it back.

But Pascal's Wager (concerning the existence of God) appears to modern sensibilities to be a form of wishful thinking and an inversion of causal logic. We need a better reason for exploring the premise of an aging clock than the fact that it would be a boon for gerontologists if it turns out to be true.

Here are four arguments in favor of an aging clock, which we shall explore in some detail presently. (1) In some animal models, interventions are known that do not simply slow the pace of decline, but actually cause reversion to a younger, more robust state. (2) Gene expression is known to change with age, including characteristic profiles that seem to be associated with senescence. (3) Attrition of telomeres seems in some ways to act like an aging clock. (4) Evidence that aging is an adaptive evolutionary program implies the existence of an aging clock.

(1) Many experimental interventions are known, for a given species, that cause the individual to revert to a younger state. Carrion beetles can be starved until they regress to a larval stage. Renewal of feeding causes them to mature again, and starvation can induce a repetition of the cycle through multiple ‘lifetimes' [1]. The coelenterate Turritopsis has been observed to perform a similar feat outside the lab, in a natural setting [2,3]. Lab mice have been rejuvenated with telomerase [4,5] and with blood factors [6,7]. In experiments with flies that are switched from a fully fed diet to caloric restriction (CR) in mid-life [8], it is found that the flies' mortality curve jumps quickly from the fully fed curve to the CR curve, as if they had been on CR from an early age. This suggests that, at least concerning those traits involved in mortality, CR does not merely slow the pace of future decline, but induces a change to a younger metabolic state.

Fahy [9] describes several more intriguing examples. Some of these might be conceived simply as upregulation of repair mechanisms, such that damage is temporarily being repaired faster than it accumulates. But even this conservative interpretation suggests that the body repairs itself more efficiently at younger ages, and this implies that the body ‘knows how old it is' and chooses an age-appropriate efficiency of repair.

(2) Gene expression is now routinely profiled with DNA microarrays. Differences between late-life and early-life gene expression have been catalogued in several different species [10,11,12]. In theory, these differences might be accounted for as the body's response to different levels of damage, rather than an age-dependent program; however, the nature of the changes with age suggests that they may be a cause of aging rather than a response to aging [12,13]. For example, inflammation seems to be upregulated and immune function suppressed (in mice) by genes expressed late in life [12].

Early expositions of the pleiotropic theories for evolution of aging [14] were formulated before anything was known about the cell's and the body's elaborate controls over gene expression. In Williams's early conceptions of the fundamental pleiotropic mechanism, he imagined that if a gene was beneficial early in life, the body was stuck with it, even if its operation became detrimental to fitness later in life. We now know that this naïve version of pleiotropy is inconsistent with fundamentals of biochemistry. In particular, there is far more DNA devoted to transcription controls than there is genetic material that codes directly for proteins; and timing of gene expression is a fundamental element of all biological systems. All of development and maturation is controlled by biological clocks, so it is completely plausible that biological clocks are available for control of aging if we believe it possible that natural selection could have led in that direction.

(3) Short telomeres are associated with higher mortality and shorter life expectancy in many species, and the correlations persist when age is statistically factored out. Since telomeres shorten throughout the life span, and telomerase expression seems to be modulated in a manner consistent with regulation of life span, the idea of telomeres as an aging clock has been attractive to a number of authors [15,16,17,18]. A full section of the present chapter will be devoted to this hypothesis.

(4) I have collected evidence elsewhere [19,20] that aging is a group-selected Darwinian adaptation, selected for its own sake. Others who have advocated this position explicitly include Skulachev [21], Bredesen [22], Goldsmith [23], Longo [24], Clark [25], Travis [26], Martins [27], Libertini [28] and Bowles [29]. Kenyon and Barja have indicated to me privately that sections of their submitted articles concerning programmed aging have been deleted by peer reviewers as a condition of publication. Many more scientists routinely adopt an implicit assumption that aging is programmed. But most evolutionists find this proposition implausible because the individual fitness cost of aging is high and the group-selected benefit is not confined to genetic kin. A brief summary of the reasons to believe that aging is an explicit evolutionary adaptation follows, and the reader is referred to my chapter in The Future of Aging[20] for details:

• Many of the genes that regulate aging (TOR, IGF, DAF/FOXO) have been conserved since the dawn of eukaryotic life [30]. All other such highly conserved genes have been protected by natural selection because they form an essential core to life processes. Natural selection has evidently treated aging as a core life process.

• Hormesis: The fact that life span can be readily extended under genetic control when conditions are most harsh and challenging (e.g., starvation) indicates that the body is ‘holding back' on life span at times when the environment is more favorable [31,32].

• Breeding animals for longevity does not necessarily impair fertility, as is demanded by popular theories based on pleiotropy or trade-offs [33]. In fact, many single-gene mutations are known to extend life (especially in worms), for which no major cost has yet been identified [34].

• One-celled eukaryotes are subject to two modes of programmed death: apoptosis [35] and replicative senescence [25,36,37]. This fact in itself vitiates the classical theoretical contention that it is impossible for programmed death to evolve, requiring as it would an implausible triumph of group selection over individual selection. Both apoptosis and replicative senescence have been conserved, so that they continue to play a role in the aging of higher organisms, including humans [38,39].

• Many semelparous plants and animals exhibit manifestly programmed death [40,41,42], providing further counter-examples to the claim that affirmative selection for death is excluded on theoretical grounds.

Once we accept the premise that aging is programmed, it follows that the body must actively track its age (in a manner flexibly responsive to environmental cues) in order to initiate senescence at a characteristic age. The process must be governed by one or more master clocks.

The aging clock does not measure strict time, but is flexible in response to environmental conditions. The way in which the aging clock responds to the environment is highly suggestive of an adaptive purpose that helps us to understand the evolution of aging generally.

It is not possible to make sense of this picture if we imagine that life histories have been shaped only by natural selection for individual fitness. For the individual, living longer is always better. Theory predicts that all life histories should emulate the lobster or the sequoia tree, growing larger and more fertile and more robust against major causes of mortality with each passing year.

Instead, in iteroparous animals, we see a fixed life span, and the length of life varying inversely with hardship and environmental challenge, especially hunger. When the death rate from starvation is high, the death rate from aging is low, and vice versa. This is the well-known CR effect, but the same is also true of other environmental challenges: physical duress, infections, temperature extremes, and toxins in small amounts all lead paradoxically to longer life spans. The phenomenon is called hormesis [31,32] and its reality has been established over the last two decades, after facing substantial skepticism of the initial accounts [43].

The demographic impact of senescence is thus to level the death rate in good times and bad, by imposing a higher mortality burden when the body is least stressed and lower just when it would appear to be metabolically most difficult to preserve the soma and avoid aging - under conditions of physical stress and starvation. By damping the most extreme variations in death rate, senescence makes possible the persistence and stability of ecosystems. Without aging, we might imagine a Tragedy of the Commons [44], with predator species competing viciously for their prey, and predator/prey population cycling much deeper than is actually observed in nature [45,46].

The variability of the aging clock, and the existence of the clock itself, might be understood in terms of natural selection at the level of ecosystems. Ecosystems built on species that have a programmed life span are less likely to suffer overshoot, instability and collapse than if the species have indeterminate life span, limited only by starvation, predation, and epidemic infections. I have argued elsewhere [45,46] that ecological homeostasis is a major target of natural selection, tempering and counterbalancing the pressure toward higher individual reproductive fitness.

If a program for aging were designed by a human engineer, it would be based on a central (flexible) time-keeping mechanism; but this is not necessarily the system that natural selection has bequeathed us. In particular, there is long-term group selection in favor of aging, but strong short-term individual selection against it. In order to shield affirmative aging mechanisms from dismantlement at the hand of individual selection, it is likely that evolution has embedded them below the surface, and deployed redundant time-keeping [14,46]. A single master clock would be easily hijacked by individual selection. We might expect to find several interdependent and redundant clocks for aging.

Under the paradigm of programmed aging, senescence may be a continuation of development, and we might suspect that whatever controls the timing of development has been extended to regulate the timing of aging. But neither a developmental clock nor an aging clock has yet been discovered. The closest thing we know of is cellular senescence, based on telomere length [47]. But there are some organisms that age, in which telomerase is freely expressed, telomeres remain long, and thus the telomere clock is not operating. Examples include some rodents [48], bats [49] and pigs [50]. These considerations make it more likely that there is another clock and that it is somehow ‘hidden in the works' of metabolism, so that it would not have been obvious to investigators thus far.

My hypothesis, proposed below, is that gene expression itself forms a kind of aging clock. Time is maintained within the signal networks of metabolism, and a running record of the organism's age is imprinted in the methylation state of the genome, as well as transcription factors and other regulators of gene expression. Gene expression products are part of a signal cascade that affects all aspects of metabolism, but that also feeds back (e.g. through methyl transferases) to increment the clock.

I will first briefly survey known biological clocks, including aging clocks, and discuss prospects for medical interventions that might manipulate them. These include thymic involution, the suprachiasmatic nucleus, the hypothalamus, and replicative senescence. The latter, based on telomere attrition, has been the subject of intense research in the past decade, with promising developments ongoing. Finally, I argue based on evidence and logic that the methylation state of the genome, within stem cells in particular, may be a promising place to look for a stored record of organismic age that informs the body's growth, development and senescence.

Telomeres in stem cells (and hence in their somatic progeny) suffer attrition over a human lifetime because they lose a few hundred base pairs with each cell division. The primary means by which telomere length might be restored is through the enzyme telomerase, which includes an enzyme which crawls along a chromosome end, and also an RNA template for copying the repetitive sequence. Telomerase is expressed copiously during early stages of an embryo's development, but very little telomerase is expressed after the individual passes beyond embryonic development. Hence, telomeres are permitted to shorten with age, even though the gene for telomerase remains (unexpressed) in the nucleus of every cell in the body.

The hypothesis that cellular senescence represents a primary aging clock was promoted by West, culminating in a popular book published in 2003 [17]. That same year saw Cawthon's actuarial study [39], associating leukocyte telomere length with mortality in humans. In the years since then, evidence has accumulated for the importance of telomere length in aging of birds and mammals including humans, and several herbal extracts that are claimed to address telomeric aging have reached the market. Meanwhile, several companies are researching more potent telomerase activators in the belief that this is a path to substantial extension of the human life span.

The Cawthon results forced many researchers to consider for the first time the possibility that people could be dying for lack of telomerase. Association between telomere length and life expectancy was confirmed in 3 studies of animals in the wild [51,52,53]. The question remained open whether longer telomeres were a marker or a cause of life expectancy. This question has been addressed with animal studies. Telomeres have been extended by adding ectopic copies of the telomerase gene, by genetically programming the expression of telomerase via a tamoxifen switch, and by oral administration of a plant-derived compound that promotes telomerase expression. Life span extension has been detected in worms, mice and rats.

Joeng et al. [54] created a strain of Caenorhabditis elegansworms with longer telomeres using not telomerase, but a telomere-binding protein called HRP-1. Life span was extended by 19% by this intervention. The result was unexpected because telomeres do not erode over the life span of C. elegans. In fact, the adult worms are postmitotic: there are no stem cells, no replenishment of tissues during a single worm's lifetime. It should not be possible for telomeres to function as an aging clock. Life extension of the HRP-1 worms was dependent on the presence of DAF-16, an upstream modifier of aging that is thought to be a master regulator of dauer formation in response to environmental hardships.

Tomás-Loba [55] first demonstrated life extension in mice using a strain that was engineered with extra copies of the telomerase (TERT) gene. Because it was widely believed that telomerase expression could cause cancer, they used mice that were cancer resistant via modified p53. These mice lived 40% longer than controls, and markers of senescence such as inflammation, glucose tolerance and neurological measures appeared on a delayed schedule. This result was unexpected because wild-type mice express telomerase copiously, and their telomeres are long enough to last through several lifetimes without obvious effects on health and longevity [56,57,58].

There have also been some negative indications, casting doubt on the hypothesis that telomeres are an aging clock. Telomere length is widely variable in newborns [59], with a standard deviation about 7% of the mean. Telomere length predicts mortality and life expectancy less well with advancing age, with correlation disappearing for ages >80 [60,61]. A few studies [62,63] have reported a tumorigenic effect of telomerase, suggesting that the evolutionary purpose for rationing telomerase is the opposite of an aging clock.

There are two known mechanisms by which telomeric aging might lead to senescence of the organism and greater risk of mortality. First, as stem cells suffer from telomere attrition, they slow down in replacing the skin and blood and immune cells that are constantly turning over. Second, cells with short telomeres enter a senescent phase where they emit toxic signals, including proinflammatory cytokines [64,65]. Hence, it is not surprising to see cellular senescence emerge as a primary driver of senescent phenotypes.

The role of telomere length as a cellular aging clock recapitulates a similar function in some protoctista. In paramecia, for example, telomerase is not expressed during mitosis, but only during conjugation. Hence, paramecia may reproduce clonally through a few hundred generations before their telomeres become shortened and they enter a senescent state, losing viability. They are compelled to conjugate, blending their genomes sexually with a partner cell, or they cease to be able to reproduce. Hence, the rationing of telomerase serves to enforce an imperative to share genes, and helps to assure that the local population remains diverse, and thus robust. Telomeric senescence serves to damp a winner-take-all form of individual selection, and enhanced diversity is insurance against excessive specialization, which might offer temporary advantage [25].

Circadian biological clocks have been widely studied and are partially understood. There is evidence for an annual clock that contributes, along with environmental cues, to patterns of migration and hibernation; but mechanisms have not been identified. And the timing of growth, reproductive maturity, and senescence remain more mysterious yet.

The mammalian circadian rhythm is known to be regulated from the suprachiasmatic nucleus, a small structure in the middle of the brain [66]. A chemical mechanism (based on peroxiredoxins) has recently been discovered that might underlie all circadian clocks [67]. The cycle is responsive to light and dark, and the intrinsic period is close enough to 24 h that circadian timing becomes entrained with diurnal cycles.

It has been proposed that seasonal cycles in mammals are mediated through a response to light in the pineal gland. The mechanism is thought to be independent of the circadian clock [68]. Again, there is an intrinsic cycle time that is modulated by temperature and duration of daylight.

In female mammals, onset of puberty is controlled by a single chemical signal: gonadotropin-releasing hormone (GnRH). But the timing of this trigger is controlled in turn by a complex calculation, based in neural as well as hormonal mechanisms. ‘The anatomical development of the GnRH secretory system occurs relatively early in life, and the synthetic capacity is present well before puberty in that GnRH mRNA expression reaches adult levels' [69]. Timing responds to olfactory cues, stress, fat reserve, activity, season of the year, and other stimuli. The workings of this clock remain mysterious.

Aging responds to these same cues, and perhaps others. There is reason to believe that the aging clock mechanism is at least as complex as the developmental clock. Though aging is programmed, it may not be programmed in a simple way. This accounts for the challenge that aging has posed for research and medical intervention. The fact that aging progresses under genetic control suggests a promising approach to anti-aging interventions, and yet the complexity of the timing mechanism has slowed the pace of progress.

Although relationships between the circadian clock and the aging clock have been documented, these are not such as to suggest that the aging clock depends directly on a count of circadian cycles. For example, dysregulation of the circadian clock in either direction leads to accelerated aging in flies [70,71].

The Neuroendocrine Theory of Aging was proposed by Vladimir Dilman in 1954 [72]. Homeostatic control of hormone secretions is supported by the hypothalamus, and different hormonal levels are maintained as is appropriate for different stages of growth and development. Dilman's hypothesis was that the trajectory of changes in the hypothalamus has a kind of momentum (‘hyperadaptosis') that carries forward after maturity and results in ‘dysregulation' that characterizes the aging phenotype. The Neuroendocrine Theory is an early precedent for the Epigenetic Theory described below. ‘The life span, as one of the cyclic body functions regulated by “biological clocks”, would undergo a continuum of sequential stages driven by nervous and endocrine signals' [73]. But Dilman did not frame this theory within the context of an adaptive program shaped by natural selection, and therefore the concept of a biological clock sits uncomfortably within its narrative.

The Immunologic Theory was proposed by Roy Walford in 1962 [74,75]. The proliferation of immune cells in the blood constitutes a kind of clock, which becomes dysfunctional as the number of cells multiply. The body's cells are mutating as the number of different immune memory cells is multiplying. Chance coincidences result in the immune system attacking self with increasing frequency over time. Walford noted how many diseases of old age have a relationship to autoimmunity, but never connected this to programmed aging. He saw thymic involution as an independent cause of immune failure, and perhaps another aging clock.

In the fall of 2012, an article [13] appeared by Adiv Johnson and a diverse team of scientists from the US and Europe pulling together evidence that the methylation state of the genome is related to the body's age, and proposing methylation as an appropriate target for anti-aging research. I would extend their proposal to argue that, if we believe there is an aging clock, the methylation state of the genome (especially in stem cells, because of their persistence) is logically the first place to look for its ‘clock dial'. Seeking a system of global signals that affects the metabolic state of the entire body, we would look as far upstream as possible. Upstream takes us to gene expression. Further up, there may be signals that affect gene expression globally, but these, too, are products of genes, and hence they can be regarded as part of a self-modifying program for gene expression. If there is not in evidence another separate clock which feeds down to affect gene expression, then it is logical to assume that this self-modifying program functions as a clock in its own right.

We know that gene expression changes with age, and that this has the potential to affect all aspects of the metabolism and the aging phenotype. If there is an aging clock, then its output must be transduced so as to affect gene transcription. Merging the clock into the transcription state of the genome would be the most economical implementation of a clock mechanism, obviating the need for a separate record of the age state of the body. Gene transcription is affected by transient signaling, and also by more persistent epigenetic markers. The most important of these persistent markers is the genome's methylation pattern. The ‘methylome' contains information that is both programmable and persistent. Cytosine (the ‘C' in ACGT) is one of the four nucleic acid residues that form chromosomal DNA. Within the DNA molecule, cytosine can accept a methyl group to form 5-methylcytosine, and this suppresses transcription locally where methylation has occurred in gene promoters [76]. Methylation patterns tend to be copied along with DNA replication, and they can even last through several generations as a form of epigenetic inheritance [77].

An epigenetic clock has the potential to regulate growth, development and sexual maturity, as well as aging. If no other clock has been discovered that controls the timing of both development and aging, then our default hypothesis ought to be that the epigenetic state of the genome is its own clock.

The methylation state of the genome is also self-modifying in the sense that transcription of methyl transferases and related enzymes creates the mechanism for feeding back upon the methylation state. This feedback implies the basis for a clock mechanism. Genes that are transcribed today create the metabolic environment that cascades into signals that reconfigure the methylation state and program the genes that will be transcribed tomorrow.

The above constitutes a general, theoretical argument for epigenetic state as an aging clock. There are also specific experimental results that point in this direction:

• Gene expression profiles change substantially with age. There is reason to believe that an individual with youthful gene expression is functionally a youthful individual [13].

• Methylation has about the right degree of persistence. We know that methylation contains epigenetic information that is passed on in a soft way when DNA is replicated.

• In general, methylation decreases with age (though there are characteristic regions that become hypermethylated). Hypomethylation has been associated with ‘frailty' and markers of biological age [78].

• Fruit flies with an extra copy of the methyl transferase dnmt2 in their genome lived 58% longer than control flies. Conversely, flies engineered to be +/- for dnmt2 had lifespans 25% shorter [79].

• Similar experiments with mice yield more nuanced results. Early, unreproduced studies reported that methylation was actually higher in dnmt1+/- mice than in +/+ controls [80,81]. More recently, neural deficiencies and low bone densities, increasing with age, have been reported associated with engineered dnmt1 deficiencies [82].

• In monozygotic twins, methylation patterns are similar when young, but diverge over time [83]. This suggests a stochastic component that may account for diverse dysregulations associated with aging.

Age is determined almost certainly by the detailed pattern of methylation and other epigenetic markers, not simply the crude quantity of methylation. And yet there is evidence that senescing cells are characterized by progressive demethylation, so that chromosomes in younger cells tend to be more methylated than older cells [84,85]. The possibility that demethylation may be an aging clock was first proposed by Bowles [29], based on the fact that ‘aging is accompanied by DNA demethylation [86,87,88]. In fact, the animal genome loses practically all 5-methylcytosines during the life, the rate of the loss being inversely proportional to maximal lifespan of the species [88]. The same occurs in cell cultures, again the rate being inversely proportional to the cell lifespan (Hayflick limit) [88,89,90].'

Skulachev also notes a connection between oxidation, which has often been recognized as a stochastic marker for aging, and methylation. In his schema, oxidation is a more fundamental aging clock (rather than demethylation leading to oxidation, as I argue here). ‘[O]xidation by ROS of the guanine DNA residues to 8-hydroxyguanine strongly inhibits methylation of adjacent cytosines [91]. Antioxidants, on the other hand, cause DNA hypermethylation [92]. According to Panning and Jaenisch [93], DNA hypomethylation activates Xist gene expression in X chromosome, which correlates with a dramatic stimulation of apoptosis. All these observations may be summarized by the following chain of age-related events: ROS → DNA de-methylation → apoptosis → aging [90].'

Drugs targeted to sirtuins [94] have found some early success in extending lifespan by indiscriminate silencing of gene expression. Sirtuins' mechanism is mediated via histone deacetylation rather than methylation, but the ease with which simple silencing of genes could extend lifespan is suggestive. Also, protein-restricted and methionine-restricted diets retard aging [95], presumably by dialing down expression of many genes indiscriminately. Methionine is the ‘start codon', essential to the initiation of all gene transcription.

This hypothesis - that the body's age is stored within the cell nucleus as a methylation pattern - suggests a program of research, and an anti-aging strategy. Interventions based on methylation will require both a detailed automated reading of the methylation state of the genome, and a means of transcribing a youthful profile into chromosomes in vitro.

The former is already fairly well developed. Heyn et al. [85] report transcription of the methylome using microarrays. The latter may be far more challenging. Methyl transferases are able to methylate targeted genes, but details of the biochemistry that guides the transferases to their target is not yet understood [13].

More accessible might be interventions to increase methylation broadly. This is a life extension strategy that has been made to work in flies [79] but not yet in mammals. S-adenosyl methionine (SAMe) is the basic methyl donor of all eukaryotes. Simple supplementation with SAMe has been found to relieve arthritis and depression symptoms [96], and SAMe has been shown to protect methylation levels in radiation-challenged mice [97], but SAMe has not been found to extend lifespan in rodents. Johnson et al. [13] catalogue some nutrients that have been associated with reduced methylation, but none with enhanced methylation. They stress that we are yet at an early stage of knowledge concerning the relationship between methylation and aging, and it is not yet proven even that alterations of the methylation state are a cause and not simply a product of aging. Nevertheless, they propose methylation as a promising avenue for foundational and clinical research, and I concur. Since this article was written, CRISPR technology (Clustered, Regularly Interspaced, Short Palindromic Repeats) has advanced rapidly. Just a few years ago, the technique was developed to target a place in the genome for gene editing. It is now expected that CRISPR will also provide a handle for dictating epigenetics [98], turning genes on [99] and off [100] at will. If this comes to pass, CRISPR may provide a dramatic shortcut, cutting through the complex biochemistry of gene expression and permitting us to target genes for promotion or repression. There are already several known genes with presumptive anti-aging benefits (e.g. GDF11, Klotho, possibly oxytocin, melatonin), and others that are overexpressed late in life with detrimental consequences (e.g. NFκB, TGF-β, possibly LH and FSH).

This chapter is adapted from an article of the same title published in Biochemistry (Mosc) [2013;78:1048-1053].