Methylation Aging Clock: An Update

Methylation of DNA is the best-known mode of epigenetic regulation (turning genes on and off).  Methylation patterns are stable unless they are actively changed, and can persist over decades, even across generations.  

Four years ago, biostatistician Steve Horvath of UCLA identified a set of 353 methylation sites that are best-correlated with human (chronological) age.  These are sites where genes are turned on and off at particular stages of life.  A computer analysis of a gene sample (from blood or skin or even urine) can determine a person’s age within about two years.

Two reasons the Horvath Clock is important.  First, it is the best measure we have of a person’s biological age, so it provides an objective measure of whether our anti-aging interventions are working.  Say you’re excited about a new drug and you want to know whether it really makes people younger.  Before the Horvath clock, you had to give it to thousands of people and wait a long time to see if fewer of them were dying, compared to people who did not get the drug.  The Horvath clock is a huge shortcut.  You can give the drug to just a few people and measure their Horvath (methylation) age before and after.  With just a few dozen people over a two-year period, you can get a very good idea whether your drug is working.

Second, there is evidence and theory to support the idea that the methylation sites that Horvath identified are not just markers of aging but causes of aging.  That means that if we can figure out how to get inside the cell nucleus and re-configure the methylation patterns on the chromosomes, we should be able to address a root cause of aging. (Before we get too excited: “Gene therapy” has been around 20 years but is still in a developmental stage; “epigenetic therapy” is what we need, and it does not yet exist, but is technically feasible using genetically engineered viruses and CRISPR.)

The write-up below is taken directly from two talks that Horvath gave, 2016 at NIH in Maryland and just last month in Los Angeles.


In 2012-2013, three papers appeared proposing the idea that the deep cause of aging (in humans and many other higher animals) is an epigenetic program [Johnson, Mitteldorf, Rando].  Genes are turned on and off at various stages of life, producing growth, development and aging in seamless sequence.  (A fourth paper by Blagosklonny proposed a similar idea, but focused on the role of a single transcription factor controlling gene expression (mTOR) and shied away from the conclusion that natural selection might have preferred aging affirmatively.  Here’s an earlier presentiment by Blagosklonny.)

It’s a powerful hypothesis that proposes to resolve evolutionary and metabolic questions alike.  It contains a seed of a prescription for anti-aging research—although epigenetics has proved to be so complicated that practical modification of the body’s gene expression schedule may require a lot more groundwork.

Unbeknownst to any of us working on these theoretical papers, Steve Horvath was already working on calibration and measurement of the epigenetic aging clock, and he published his basic result by the end of 2013.

One remarkable property of the Horvath clock is that it is more accurate than chronological age for predicting who will contract aging diseases and who will die.  Even though the clock was derived with an algorithm that matched the output clock age as closely as possible to chronological age, the result proved to contain more information than chronological age.  “In deriving the clock, chronological age was used as a proxy for biological age.”  People whose “methylation age” is greater than their chronological age are likely to suffer health deterioration and to die sooner than people whose methylation age is less than their chronological age.  

Horvath has openly shared his methodology and his computer program.  Based on the Horvath clock, a California company began last year to offer a commercial test for methylation age.  You can send a blood or urine sample to Zymo Research.

 

Candidate aging clocks

Horvath describes how he came up with the idea of a methylation clock by a process of elimination, beginning with four candidate clocks:

  1. Telomere length
  2. Gene expression profile
  3. Proteomic data
  4. DNA Methylation

In detail:

  1. Telomere length – This had been measured easily and cheaply for more than a decade, but its correlation with chronological age (and with mortality) is not strong enough to be useful as a biological clock.

  2. Gene expression profile: Which genes are being transcribed into RNA at a given time?  This can be measured by extracting RNA, and turns out to be highly tissue-specific.  In other words, it varies according to which part of the body you’re looking at.
  3. Proteomic data:  Genes, once transcribed, are translated into proteins.  Some of these proteins stay in the cell while others circulate through the body.  Gene CHIP technology measures levels of different proteins reliably and inexpensively.
  4. DNA Methylation: Easier to measure than (2) or (3). Methylation is only one of many mechanisms controlling gene expression, but it is one of the most persistent.  Horvath found that a subset of DNA methylation sites seems to be characteristic of age no matter where in the body they are measured.

What is DNA methylation?

Adjacent to many genes is a promoter site, a location on the same chromosome which stores temporary information about whether the gene is turned on or off.  Promoter sites contain the base sequence C-G-C-G-C-G-C repeated.  This is called a CpG island (where the “p” just tells you that the C is linked to G on the same strand, rather than being linked across strands, in which C is paired with G.)

C stands for “Cytosine”, and the Cytosine molecule can be modified by adding an extra methyl group (CH3) to form 5-methyl Cytosine.

The cell has molecular workers that are deployed to go around specifically adding methyl groups in some parts of the DNA or removing them in others.  The bottom line is that methylated Cytosine is a sign that says “don’t transcribe the adjacent gene.”  When the methyl groups are removed, it is a signal that the gene are to be transcribed once more.

Enzymes called methyl transferases are deployed to precise regions of the genome to turn genes on and off.  Methylation can be transient.  There is evidence for circadian cycles of methylation.  Or it can be quite long-lasting.  Methylation patterns can persist for decades, and are copied when cells replicate, so that methylation patterns can be passed to offspring as part of one’s epigenetic legacy.  Inherited methylation sites are the exception however; most of the genome is programmed fresh with age-zero, pluripotent methylation patterns when egg and sperm cells are generated.

 

How the methylation clock works

Using a standard statistical algorithm, Horvath identified 353 CpG sites that were most strongly correlated with chronological age, no matter where in the body he looked.  The same algorithm provided 353 numbers to be multiplied by methylation levels at each site, then added up to produce a number.  The number is not directly a measure of age, but in the last step a table is used (an empirically-derived curve) to associate the number with an age.

This is the raw output of the function before it is transformed into an age.  Notice that methylation changes very rapidly during the first 5 years of life, gradually slowing during the growth phase and straightening out to constant slope after about age 18.

Even though the Horvath clock was designed to be independent of what part of the body DNA was drawn from, some variations appear.  Most noticeable is female breast tissue, which ages faster than the rest of the body, and brain tissue, which ages more slowly.  Blood and bone tissue tend to age a little faster.  (Sperm and egg cells are “age zero” no matter the age of the person from whom the germ cells were drawn.  Placentas from women of all ages are age zero.) Similarly, induced stem cells (using the 4 Yamanaka factors) have zero age.  In contrast, a similar treatment can change one differentiated cell type into another, for example, turning a skin cell into a neuron.  This does not affect epigentic age.

Liver cells tend to be older than the rest of the body in people who are overweight, and younger than the rest of the body in people who are underweight.  Other tissues don’t seem to show this relationship.  For example, fat cells do not have older methylation ages in people who are obese.  And, perhaps surprisingly, weight loss does not reverse the accelerated methylation age of the liver (at least, not within the 9-month time frame of the one study looking at this).


Studies have been done correlating methylation age with various diseases and, of course, mortality.  Corrections are made for every kind of environmental factor, including smoking, obesity, exercise, workplace hazards, etc, called collectively the “extrinsic factors”.  The result is that methylation age rises with extrinsic factors, and independently methylation age is also correlated with intrinsic (genetic) factors that affect lifespan.  Horvath estimates that genetics controls 40% of the variation in methylation age (as it differs from chronological age).

Men are slightly older than women in methylation age.  This is already evident by age 2. Delayed menopause is associated with lower epigenetic age. Cognitive function correlates inversely with methylation age of the brain.

Speaking before Horvath at the same conference, Jim Watson claims there are many supplements and medications that can slow the Horvath clock.  The one he focuses on is metformin, which, he says, has epigenetic effects via an entirely different pathway from lowering blood sugar (the purpose for which it has been prescribed to tens of millions of diabetics).

 

Here’s a curious clue:  There is a tiny number of children who never develop or grow, and continue to look like babies through age 20 and perhaps beyond.  These children have normal methylation age.  Whatever it is that blocks their growth, it is not the methylation changes in their DNA.  Does this mean that there are other epigenetic controls, more powerful than methylation, that control growth and development?  Or does it mean that children with this syndrome have normal epigenetic development, but something downstream from gene expression is blocking their growth?  Conversely, Hutchinson-Gilford progeria is caused by a defect in the LMNA gene which causes children to age and die before they even grow up.  Hutchinson-Gilford children have normal methylation ages by the Horvath clock.

Radiation, like smoking and exposure to environmental oxidation, tends to age the body faster.  This is independent of methylation age—which is unaffected by radiation.  Neither smoking nor radiation exposure affect epigenetic age.  HIV also accelerates aging, and HIV does affect methylation age.

Methylation age and telomere age are both correlated with chronological age, and they both predict mortality and morbidity independent of chronological age.  But the two measures are not correlated with each other.  In other words, the information contained in the methylation clock and in measures of telomere length complement one another to offer a better predictor of future aging decline than either of them separately.

Diet has a weak effect on methylation age.  Very high carbohydrate, very low protein diets are noticeably terrible.  Beyond this, there seem to be two sweet spots: one for the Ornish-style protein-restricted diet and one for the Zone/Atkins style diet.  Weak evidence to be sure, but suggestive that they both work.

“The epigenetic clock is broken in cancer tissue.” [ref]

 

Building on the original clock

The original clock was optimized to track chronological age, and yet it fortuitously provided more information than chronological age.  In a second iteration, Horvath set out explicitly to track biological age.  He used historic blood samples from the 1990s, and paired them with hospital records and death certificates to search for methylation sites that correlate best with aging-related health outcomes.  The result was the phenotypic clock, DNAm phenoAge.  This uses 513 methylation sites to predict

  • all-cause mortality
  • cardiovascular mortality
  • lung disease
  • cancer
  • diabetes
  • (loss of) physical strength
  • (loss of) cognitive ability

On the drawing board:  An epigenetic clock specialized to work well with skin and blood cells, (which are the most accessible).  (Enough skin cells can be scraped painlessly from the inside of your mouth (buccal epithelial cells) to do a DNAm test.)

 

Connection to Parabiosis and Plasma Transfusions

Several groups have begun to experiment with transfusions of blood plasma from a young donor as a possible path to rejuvenation.  Horvath reports an encouraging finding:  Sometimes older people contract a form of leukemia that requires a blood and marrow transfusion (including the stem cells that give rise to new blood) from a donor.  The finding is that after this treatment, the blood of the patient continues to show the methylation age of the donor, not the patient.  

 

Epigenetic Aging and Telomere Aging Bound to a See-Saw Relationship

(This was the most exciting new result for me personally, because it relates to an idea I have held dear for more than a decade.)

Methylation age is older or younger than chronological age in different people, generally by about +2 years.  40% of the variation is due to genetics.  Some common genetic variants can make the clock run faster or slower.  The most prominent genetic variants link telomere aging to methylation aging.  The faster your epigenetic clock runs, the longer your telomeres.  The slower your epigenetic clock runs, the shorter your telomeres. [preprint]

There’s a word for this in the genetic theory of aging.  It’s called Antagonistic Pleiotropy.  Back in 1957, George Williams theorized that the genes causing aging ought to have simultaneous beneficial and detrimental effects.  That would explain why natural selection has permitted aging to occur, despite the fact that it cuts off fitness.  Williams said: Nature had no choice but to accept the genes that cause aging because there was no other way to get the benefits of these same genes (which he surmised ought to enhance fertility).

My theory of Antagonistic Pleiotropy is that it is not a situation of “forced choice”; rather, aging is important for the health of the community, and mother nature has been faced with the dilemma: how to keep aging in place despite efficient natural selection against it on the individual level.  Aging is so important to the community that evolution has been motivated to find ways to keep it in place, despite the short-term temptation for natural selection to favor those with longer lives (thus greater opportunities to leave offspring).  In my hypothesis, evolution invented pleiotropy to address this problem. The telomerase-epigenetic clock connection is an example.  There is no physically necessary connection between telomerase and epigenetic aging, but the two have evolved a see-saw link so that it is more difficult to mutate aging away.

This also relates to my coverage last fall of the telomerase-cancer connection.  At the time, I was scratching my head, why should genetic variants that lengthen telomeres be associated with higher rates of some cancers?  Here is a clue: The same genetic variants that lengthen telomeres also accelerate the epigenetic aging program.  The specific example of a cancer that is most closely tied to higher telomerase levels is melanoma, which is a cancer that is less sensitive to age than other cancers.  People tend to get melanoma earlier in life than other skin cancers. Therefore, I predict that other pleiotropic links will be found between these genetic variants that promote longer telomeres and other mechanisms linked specifically to melanoma.


The Bottom Line

All these data in a field so new is a tribute to Horvath’s industriousness and to the promise and fruitfulness of a new methodology.

The data so far suggest that methylation programming is a big part of the driver of aging, but not the whole story.  Smoking affects life expectancy, but it doesn’t affect methylation age.  Weight loss benefits life expectancy, but it is invisible to methylation age.  Most curious are those children who fail to develop, or age prematurely, even though their methylation age is progressing on schedule.

What does it mean that radiation ages the body without advancing the methylation clock?  Perhaps that accumulation of damage is part of the phenotype of aging, though I remain hopeful that the body remains capable of undoing that damage even late in life, if it is re-programmed to want to do so.  What does it mean that AIDS advances the aging clock?  Perhaps that the immune system is a central signaling mechanism in the aging process.

So, it’s “methylation plus”.  Plus what?  Not just methylation plus damage”; though we can certainly shorten our lifespan with radiation or smoking, we can’t increase our lifespan by avoiding toxins.  “Methylation plus other epigenetic programs”—this would be my first guess.  “Methylation plus mitochondrial state” would be a close second. Methylation is all in the nucleus, and the cytoplasm of the cell seems to store independent information, and can even re-program the state of the nucleus, as suggested by parabiosis experiments. There is also evidence for“Methylation plus telomere shortening”. 

How to Test Interactions among Life Extension Treatments

The Most Promising Way Forward for Anti-Aging Science Today

We now have many effective interventions (mostly of small effect) for longevity and preventive care.  Most readers of this blog take more than one supplement each day; some of us (I confess) take very many.  We make an unthinking assumption that “more is better”, or rather that if A is beneficial and B is beneficial then if we take A and B we can get the benefits of both.  We don’t think to question or deconstruct this reasoning.  It is rooted in a reductionism that works pretty well in the physical sciences, much less well in biology.  

We know that the benefits of all these interventions don’t just add up like numbers in a spreadsheet, but we continue to act as though this were our reality.

The truth is that we know almost nothing about the cross-talk among different health interventions.   The reasons so few experiments have been done are plain enough, but the situation has become untenable.  There is an urgent need to understand the interactions among treatments. We might begin with those that are individually most promising, but expect surprises.  The combinations that offer the greatest longevity benefits may turn out to be pieced together from components that individually have little or no effect.


I might have said ‘the most promising way forward for medical research today, because I believe that anti-aging science is the most productive area of medical research.  If you are reading this page, you probably know this already, but we all take comfort in confirmation of what we already know. So:

Prevention is more cost-effective than treatment.  The root cause of most disease in the developed world is aging.  This point has been made decisively [for example], most eloquently by Aubrey de Grey.  (The root cause of most disease in the third world is poverty, and ending poverty is also an essential imperative, but it is not a subject for medical research.) 

The problem of interactions has been neglected for a number of reasons:  

  • Unconscious linear thinking
  • The dizzying number of combinations that need to be studied
  • The want of a guiding paradigm that would provide context for individual studies
  • Scientific inertia:  researchers are more likely to study (and funders are more likely to support) research programs that are established and proven

But the problem is potentially of great import.  We expect a great deal of redundancy among the mechanisms of action of various interventions we know about.  Taking two or three or four drugs that address the same biological pathway is likely to be a costly waste.  More rarely, longevity drugs may interact in ways that actually interfere and reduce overall effectiveness.  

But we have good reason to hope that in rare cases there are combinations that are more than the sum of their parts.  These fortuitous combinations synergize to offer greater benefits than they provide separately.  Finding a few such combinations would be a jackpot that justifies many, many expected null results.

The huge number of possibilities to be covered

If we begin with 30 individual interventions, there are 435 pairs of interactions and 4060 combinations of three and 27,405 combinations of four.  If we think traditionally, each one of these combinations is a research program in itself, requiring at least several person-years of professional effort plus overhead.  This is the daunting reality that confronts anyone who is intent on beginning to address the problem of interactions.  27,405 experiments of any kind is a labor of Hercules, even for a well-funded, fully roboticized biomedical lab.

There is a hierarchy of experimental models for studying anti-aging interventions:

  • Human cell cultures are the cheapest and fastest, but we learn the least
  • Complementing human cells are yeast cells, which actually have a life expectancy and some biology that overlaps our own
  • Studies of thousands of C. elegans worms can be done efficiently with robotic controls and worm counters.
  • Fruitflies are a great deal “more like us” than worms and they can be raised in large numbers, live just a few weeks.
  • Lab rats and mice are expensive, but they are mammals with biology that is much like our own.  Experiments in rodent longevity last 2 to 3 years.
  • Human trials require extensive safety measures and typically take decades to see subtle changes in health and mortality statistics; but this is the most direct indication of what we want to know.

So, how might we begin?

We have no idea what we will find.  Maybe there will be a few spectacular combinations.  Maybe the interactions will turn out to be small, mostly negative, and boringly expected.  (My guess is that both of these will turn out to be true.)  We should not try to define the second stage of the program until we have results from the first.  

The first step is to choose the most promising interventions to combine.  A great number of drugs and supplements are known that extend lifespan in rodents and/or lower mortality in human epidemiology.  Magalhaes and Kaeberlein have put together a large database of animal studies that seems to be off-line at present.  Another version is live at this address.  Here is a list I proposed in this column two years ago:

  • Rapamycin
  • Aspirin
  • Metformin
  • Melatonin
  • Deprenyl
  • ALK5 inhibitors
  • Epitalon/Epithalamin
  • MitoQ/SkQ
  • Beta Lapachone (Pao d’Arco)
  • Spermidine
  • Berberine
  • Dinh lang (Policias fruticosum)
  • Pterostilbene/Resveratrol
  • Gynostemma pentaphyllum (jiao-gu-lan)
  • N-Acetyl Cysteine (NAC) / Glutathione and precursors
  • Ashwagandha
  • Turmeric/curcumin
  • C60
  • Oxytocin (not oral)
  • J147 (a promising new Alzheimer’s drug)
  • NR, NMN and NAD precursors

We might add

  • Polyphenols from tea
  • Flavinoids from blueberries
  • Angiotensin inhibitors
  • NLRP3 antibodies
  • Acetyl L-Carnitine
  • Piracetam
  • DHEA
  • Statin drugs
  • Cardarine / GW501516 / PPAR agonists
  • Dasatinib / Quercetin
  • FOXO4-DRI
  • Astaxanthin
  • Momordica charantia (bitter melon)
  • Gotu kola / Bacopa
  • Reishi mushroom
  • Astragalus extracts
  • Pine bark extract
  • Ginseng
  • Acarbose
  • BHT

Interventions not in pill form include

  • Exercise
  • Caloric restriction
  • Intermittent fasting (various schedules)
  • Plasma transfusions from younger individuals
  • Platelet-rich plasma
  • Transplanted young thymus
  • Transplanted young suprachiasmatic nucleus

How to prioritize and explore the huge number of combinations?  Here are four ways we might begin to sort through the possibilities:

  1. Use theory: Look for biochemical mechanisms that seem complementary
  2. Traditional Chinese Medicine, Ayurvedic medicine and other ancient traditions suggest combinations of herbs that long experience says function together.
  3. Broad screens for especially potent combinations
  4. Statistical mining of an on-line registry of what people are taking presently

Let’s look at these one at a time.

1Biochemical Theory

We know a few biochemical pathways that are linked to longevity.  They all overlap and talk to each other.  Nevertheless, we expect that treatments that address the same pathway are likely to be redundant, whereas treatments that address distinctive pathways have a better chance of synergizing.  For example, insulin resistance is a robust hallmark of aging.  The insulin pathway is most plastic and most accessible to intervention.  Fasting and caloric restriction address the insulin pathway, as do metformin berberine, jiaogulan and bitter melon.  Exercise has many benefits, some of which work through the insulin pathway.

We might continue classifying interventions that address other pathways.  Here are some longevity pathways of which I am aware:

  • Insulin
  • mTOR
  • Inflammation
  • Immune senescence / thymic involution
  • Epigenetic reprogramming / transcription factors
  • Mitochondrial senescence
  • Autophagy
  • Anabolism / Catabolism imbalance
  • Telomere attrition
  • P53 / Apoptosis

Someone who knows more biochemistry than I do might be willing to classify the interventions I list (and others) according to these 10 pathways (and others).  Here is a template in Google Sheets, which I establish as an open Wiki. http://tinyurl.com/longevity-pathways

2. Eastern and Indigenous Medical Traditions

Many useful modern medicines are derived from ancient folk wisdom.  But this work has proceeded with a deductive logic, isolating active chemicals from whole plants (as aspirin from willow bark, cycloastragenol from astragalus, and curcumin from turmeric).  Many folk traditions, especially Traditional Chinese Medicine, are based on not just whole herbs but combinations of herbs that have been found over the ages to work together.  Ideas may be taken from these traditions to prioritize combinations for testing.  For example, the best known Chinese longevity formula is Shou-wu-chi (;), which is compounded of (list from Wikipedia):

Other ingredients may also include: Hedychium coronarium (white ginger), Jambosa caryophy Llus Ndz, Citron (Citrus medica), and Conioselinum univittatum Turcz

The Ayurvedic tradition is less contains fewer formulas, but combinations that are said to contribute to longevity include these (which I found, just for illustration, at Banyan Botanicals)

  • Haritaki (Terminalia chebula)
  • Guduchi (Tinospora cordifolia)
  • Amalaki (Embelica officinalis)
  • Kumari (Aloe barbadensis, or aloe vera)
  • Guggulu (Commiphora mukul)
  • Brahmi or gotu Kola (Centella asiatia) or closely-related Bacopa
  • Ashwagandha (Withania somnifera)

3. Broad screens for particularly effective combinations

Two years ago in this space, I proposed a screening protocol in which all combinations of 3 interventions from a universe of 15 would be tried on just 3 mice each.  I showed with a computational model that if these included at least one lucky combination that increased longevity by more than 50%, then, despite the small number of mice, it would be identified with at least 80% confidence.  Combinations of three from a universe of fifteen is a kind of sweet spot for this particular experimental design, and much less is learned if the numbers are scaled back.  This means it is not feasible to test the concept on a small scale.  The full proposal requires 1365 mice in cages of three, followed for at least two years.  Cost estimate is about $2 million in the US or Europe, perhaps as low as $500,000 to do the same experiment in China.  I would be eager to work with any lab that has the expertise and the facilities to implement this protocol.  The experimental design and simulated analysis was recently published in English in a Russian journal.

4. Data-mining of an online registry where people record what supplements they are taking and commit to reporting their health history

It would be a great public service if someone were to establish a web-based registry where individuals could share information about what supplements they are taking and what results they are getting.  Over years, this could turn into a data miner’s heaven for information about individual drugs and lifestyles and their interactions.  The subject is too big for a controlled experiment, but enlisting the public would be a great and greatly-rewarding project.

I know there are web sites such as Longecity that are excellent resources for anecdotal accounts of others’ experiences.  But the data is not in a format that lends to statistical summaries.  If you know of an existing online database of this sort, please reach out and share the web address with me.  

I have preliminary plans to create such a web site in conjunction with a forthcoming book project.

The Challenge

There may already be a viable plan for major life extension hiding in plain sight.  There is no extant research program to explore the relationships and interactions among life extension measures.  Eventually, some large, well-funded agency (perhaps NEA or the Buck Institute) will take on this project in a systematic way.  But the large organizations are conservative, and are unlikely to begin until the ice is broken.  Thus, even the first shards of information in this area are likely to be valuable indications of a new research direction.

If you have a research lab, or if you know are connected to someone who might be interested in this project, or if you have a funding source, please let us work together.