Can we trust methylation clocks?

A methylation clock is an empirical construct. There is no understanding of physiology or metabolism built into the process. The clock is engineered to do the best job predicting (in the case of the GRIM-Age clock, for example) future mortality and morbidity based on methylation patterns. The whole process is agnostic about biological mechanism.

It is a legitimate question whether a drug or diet that sets back the methylation clock has actually increased life expectancy. Maybe methylation is a downstream consequence of aging, like grey hair or wrinkled skin. We would hardly expect a skin cream or hair dye to increase life expectancy.

For me, personally, this is an easy question. I have devoted much of my professional career since 1996 to opposing the “selfish gene” version of evolution and promoting multilevel selection. I have collected evidence that aging is a systemic phenomenon, centrally controlled, and that epigenetics (including methylation) is the primary way in which aging is enforced on the body. I was poised to believe that methylation clocks measure something real and important even before the first clocks appeared [2013].

Many other scientists looking at anti-aging interventions have been happy to take a practical approach, not invoking theory at all, but accepting the impressive correlations of aging clocks with other measures of biological age as good enough reason to trust that any intervention able to sett back the methylation clocks is probably setting back biological age.

Morgan Levine is a biostatistician par excellence. As a post-doc working with Steve Horvath, she developed what I consider to be the best, most usefull methylation clock..With her own research group at Yale, she has continued to innovate, with a promising approach based on the mathematics of principal component analysis [my write-up last September]

In a recent preprint, Morgan Levine has deeply questioned whether methylation clocks can be trusted in the way that so many of us have trusted them. Although young and just at the beginning of her career, she has done more than anyone except Horvath himself to advance methylation clock technology. For her to question the foundational value of her own work is a gesture of courage and deep intellectual honesty.

Before the post-doctoral work with Horvath that created the PhenoAge clock, Levine studied evolutionary biology of aging with the incomparable authority, Caleb Finch. She has her own ideas about the evolutionary origins of aging, and they are rooted in classical evolutionary theory. She sees the cause of aging as somatic evolution and accumulation of damage. She is deeply influenced by Peter Medawar’s [1952] hypothesis that what happens late in life is outside the influence of natural selection.

And so she raises the deep question: how much of epigenetic change associated with age is a driver of aging, and how much is a response to the body’s increasingly damaged state?

“Though the connection between risk and time may appear probabilistic on the surface, the emerging pathology is rooted in the molecular and cellular remodeling of the organismal system over its lifetime. Such changes likely result from accumulated damage, selection pressures at the level of cells, compensatory mechanisms, and/or the unintended consequences of a biological program. However, alterations to a complex system must abide by a hierarchical structure2, initiating at lower levels of biological organization (e.g. molecules) prior to manifesting at the higher levels in which they are typically observed (e.g. tissue and organ dysregulation and failure, and eventually death)3. Thus, to delay, prevent, or even reverse the maladies currently awaiting us in late life, we must discover how to decipher and remodel the molecular fingerprint of aging.”

Here, Levine is raising the most basic questions about the origin and meaning of methylation changes with age. Levine proceeds from her strongest skill—She is master of a wide array of sophisticated statistical tools.

Part of the classification has to do with Yamanaka programming, which is about stem cell vs differentiated cell. Another part comes from Framingham Heart study patients, and which CpGs change with age in FHS subjects. She distinguishes sites that increase methylation with age from others that decrease methylation with age. She associates some CpGs with cancer.

She maps Shores, Islands, and Open Seas. CpG islands are promoter regions of the genome that have lots of CpGs in close proximity. Shores are regions on the boundary of CpG islands, and Open Seas are regions in which CpGs exist as isolated, disconnected units.

Having divided 4779 CpGs into 12 groups, she can ask, How much of each group is represented in each of the most commonly used methylation clocks?

And which modules are performing best and most consistently across clocks?


Levine entertains the idea of “epigenetic drift” as part of the story, however she recognizes that the changes that underpin the most reliable clocks are not “drift” but clearly directional. She asks, to what extent do methylation changes cause aging and to what extent are they responses to various, incidental results of the aging process?

“If DNAm changes were purely reflecting entropic alterations or epigenetic drift, we would expect to see a bias against changes in CpGs that start around 0.5 (corresponding to random chance of methylation at a given site) . However, what we observe is actually a regression away from the mean, in which these heterogeneous populations of cells are systematically losing DNAm with time. This suggests that the green-yellow module’s notable pattern of epigenetic aging is unlikely to stem from noise or aberrant DNAm changes with age. Instead, DNAm changes may reflect cellular selection pressure or clonal expansion in which the cells without DNAm at these CpGs are able to outcompete (proliferate more than) the ones with DNAm . Alternatively, it could reflect a regulated compensatory mechanism that gets initiated with aging, or a continuation of a developmental program that is not turned-off . These scenarios have different implications for our understanding of epigenetic changes. The first would suggest that individual cells are not changing DNAm patterns with age, but rather the changes that are observed in bulk data are happening at the level of cell populations, shifting prevalence of cells with heterogenous states. The second and third scenarios, on the other hand, would suggest within cell DNAm changes, perhaps as a response to extracellular environment or signaling changes with aging (e.g. integrated stress response (ISR) ), or as an extended developmental program that fails to be extinguished—somewhat aligned with the hyper-function theory of aging . In moving forward, single-cell DNAm data may help distinguish individual vs. population changes.”

Here she references “Integrated Stress Response” as a theory of the aging metabolism. She also refers to Mikhail Blagosklonny’s idea that developmental programs have a momentum that spills over into aging phenotypes.

Morgan Levine is a brilliant scientist, facing the harshest possible self-criticism of her work. Her conclusions are tentative and open-ended.

A more definitive, empirical approach

The big, interesting question may not require theoretical analysis. What we want to know is whether we have been justified in using methylation clocks to indicate whether aging has been slowed or reversed. The most direct answer to that question comes from Harold Katcher’s rats, — the most successful example of setting back methylation age in a whole animal. Currently, Katcher has two sets of 8 rats that are the same chronological age; one group has been treated with E5 and has a much lower methylation age. He is waiting to see how long each group lives. So far, 3 of the untreated rats have died, and 1 of the 8 treated rats. Of course, the treated rats look and act much younger, and have physiological characteristics of younger rats. Over the coming months, the survival test will produce an answer to the important question whether a younger methylation age implies a younger biological age, in a form that is independent of theory.

46 thoughts on “Can we trust methylation clocks?

  1. Thanks for the balanced viewpoint Josh. I admire Harold’s work and I’m hoping for the best, but I’m not certain that based on one study comprising a rather small number of subjects that a cause and effect can be concluded with absolute certainty. It would be suggestive certainly. I also believe that one of the treated rats has died.

    • And 3 (not 2) control rats have died.

      A nice piece by Josh covering Levine’s amazing work.
      And now Levine and Horvath are reunited at Altos.
      Exciting times!

      • Are we sure the world can’t spare, say, 50 rats for critical studies like this? Maybe get some actual statistical significance? After all, this study is GOOD for the rats 😉

  2. I would like to know if any of these DNA methylation clocks are looking at the methylation status of the 36 genes that get turned off with aging (by methylation) in all mammals and the 12 genes that get turned on with aging (by demethylation) that were discovered by Horvath as being found as an evolved, conserved, aging program. If these clocks are looking at these methylation sites then I am pretty sure that methylation /demethylation is the cause of aging

    • Yes, Horvath showed that actually it is the increases in CpG methylation that counts (across species). So I would guess we are getting a block on differentiation of stem cells with age and consequently significant changes in cell populations, i.e. forced proliferation of self-renewing, only partially differentiated cells.

      A little ridiculous Levine is questioning this NOW. I raised this point years and years ago on this very blog – and only now are they thinking to explore cell specific DNAm changes. Just goes to show the myopic approach of the lab scientists and the unquestioning acceptance of most others following their work.

    • The question is whether the methylation status of the 36 genes that get turned off with aging, and the 12 genes that get turned on with aging is reversed with young plasma. That question Akshay can answer

      • I am not suggesting it is the cause of aging, necessarily. It is part of the change of dynamics in stem cell differentiation – even fully reversing all methylation based changes from youth to age is not going to magically extend telomeres (in fact some of these changes may help stem cells maintain telomeres, whilst decreasing their differentiation potential in the body) – so the rats (or people) will still die without the replenishment in mitotic cells required.

          • Hello All

            It turns out that the 36 genes that get shut down with aging are controlled by TET enzymes which keep them demethylated when acting properly in the presence of AKG and antioxidants (Vit C) as we age AKG levels drop dramtically then these 36 genes gradually quit making transcription factors that differentiate your cells and your cells forget what kind of cells they are. I have been taking large doses of AKG and or Ca-AKG for the last 9 months and the hairs on my arms and lower stomach are growing in dark again…eyesight has gotten better, and it has reversed inflammation, pain and stiffness in my shoulders that Ive had for 7 + years now 100%…I read some where the demethylation of the 36 genes happens upon cell division so it takes awhile for this to kick in. This also can explain why the Ponce De Leon Company experiment where they reversed dna methylation age in subjects taking 1 gram of slow release Ca AKG per day by 8 years in 7 months. I beleive the 36 gene system is the older aging system and evolved first befoe the evolution of sexual reproduction = somatic aging. And I also think the 12 genes that get turned off are controlled by the changes in reproduction related hormones like melatonin progesterone DHEA pregnenolone FSH LH. hCG..why? amongst the 12 genes include some are involved with Alzheimer’s disease. It has been shown that LH and progesterone and melatonin are involved with triggering and or protecting from Alzheimer’s.

          • @ Jeff, when cells divide half of the DNA requires the addition of methylation by DNMTs and then the demethylation of excess methylation by TETs – this last step seems to be the lagging factor, hence why cell division seems to cause the accumulation of methylation on CpG islands. I suspect something similar happens during DNA repair.

      • @Jeff, interessting observations. Have you noticed any improvement in skin quality? It’s said that AKG improves collagen production, but turnover rate is very slow, so I would expect improvement to be slow but steady?

        • Hi there

          yes I am notcing some skin changes the first one was the two horizontal furrow marks next to my eyebrows when I squint…I think peple get botox to stop this well maybe 5 months into my experiment I touched them and I felt little tiny crystals under the skin right where they were…I pushed on them with my thumbnail and heard them crunchng into little pieces and were quickly gone. My furrow depth was much reduced and the skin was more elastic and snapped back in place faster then later it seemed the surface wrinkle on top of the skin was kind of peeling off. Also I think the skin on my upper arms is changing as it had started to get a bit loose and crepey. Im getting little kind of age spots popping up on the skin which can then be scraped off..Hope it is not wishful thinking.

        • Oh and one more thing…..

          the skin above my upper eyelid used to be looser and thicker so I could pull it down over my whole eye quite easily…suggesting I could probably use an eyelid job….after about 9 months I have noticed that this skin has shrunk and is almost gone I dare say I can only pull it down half way down my eye to my eyelashes now…

        • CaAKG… I see thickening of the skin, narrowing of the small wrinkles. Slow but steady. Bigger wrinkles less change. Anyone combined it with peptides, like BPC- 157 or epitalon? What are results?

    • Do these 36 genes that get shut down with aging , reverse their state when introduced to young plasma is the main question. If these 36 genes reverse their state upon interacting with young plasma, then they are dependent on the signals within plasma and the plasma decides their status, rather then these genes deciding the state of the plasma and rate of aging of the organism.

  3. Hi, Josh, Thanks for the amazing piece (as always). Apparently, one treated rat is dead and more have cancer.

    One thing that caught my eyes based on the data of this second study was that the biometric parameters after E5 were far worse than those on the first study.

    Apparently, Katcher thinks it may be a gender issued (only females were included in the longevity curve study) and is thinking about increasing frequency of treatment to recapitutale the effects on males.

    Also, is there a way to know before the study if the rats have an incipient cancer? Don´t think so and that´s one of the reasons we would appreciate a higher N (money allowing).

  4. According to informational space diagnostics of the ICD, aging is embedded in the control programs of the Human Soul and begins at the age of 16-20. The rate of aging depends on the structure of the water. With aging, the number of all body cells changes. In the reverse process of aging, rejuvenation is the restoration of the number of body cells. Human experiments show that the rejuvenation process is about 10 times faster than the aging process.

  5. Fascinating stuff. We still haven’t done much work on the original epigenetic modification, though: telomerase activation. Several telomerase enhancement and telomere extension experiments showed ~40% extension of lifespan on rats (which is weird, because rodents don’t shut off telomerase).

    But no one ever followed up with a primate experiment. Bill Andrews was trying in 2008 at Sierra Sciences, but his backers lost their money in the subprime crash.

    It’s not just telomerase gene vectors, either. Powerful telomerase activation chemicals have been discovered, e.g. the madecassic acid from Centella asiatica. Why no research interest in this field?,strong%20natural%20telomerase%20activator%20with

    • Couldn’t agree more Bill. Seems like much was expected from TA65 and the baby was tossed with the bath water. Mark and I have discussed the interesting possibility of combining rapamycin with gotu kola to achieve both mTOR inhibition along with telomerase activation. Might be a potent combination.

      • I wonder how Bill tests for telomerase activity.
        Multiple studies have shown resveratrol activates telomerase significantly but not directly, one suggested it activated it through activating sirtuin 4 and another through activating splicing factors.

        The interesting thing was that telomerase activation from resveratrol was similar to telomerase positive cells in young animals but much less in older animals. This is likely do in my opinion to the aging related loss of NAD+ that is needed for sirtuin function. I hypothesize that an NAD+ booster combined with Resveratrol will likely result in telomerase activation even in old and might very well result in lifespan extension.

      • Hi Paul

        As an MD, have you had any empirical evidence of telomere’s increasing with Use of Hyperbaric Oxygen therapy (HBOT)?


        “In this study, for the first time in humans, it was found that repeated daily HBOT sessions can increase PBMC telomere length by more than 20% in an aging population, with B cells having the most striking change. In addition, HBOT decreased the number of senescent cells by 10-37%, with T helper senescent cells being the most effected.”

        A session costs about $70 to 80 dollars for this purpose after a stint of daily therapy, treatments maintenance therapy would be needed several times per month, according to some research.

        • Hi Heather, It is arguable whether this is a good thing. Oxygen therapy kills lots of old cells, causing a rapid proliferation of replacements from the underlying progenitor pool. Sure you’ll be healthier, but with a reduced reserve. So similar to your comment below, it depends what you want – health now or a longer life.

          • Hi Mark:

            Thank you for the response and good point. Yes, long term studies would be required to make a decision on it’s total benefit. Early studies can be misleading.

            Maybe if a person is 90 it would be worth trying now.

    • Hi Bill, what makes you think it was the madecassic acid that did the magic? I always assumed it was the asiaticoside?

      In answer to your implied question, rats still experience telomere shortening across their tissues, it is the net effect of shortening vs. telomerase driven lengthening that counts in lifespan. Blasco has shown that quite clearly. In humans, even without telomerase we still lose telomeres only slowly.

      @Darian, Bill measures telomerase activation by TERT RNA, i.e. transcription of the gene, and by that measure resveratrol does not work. As a HDAC inhibitor it will have very unpredictable effects depending on cell type and their age. So wouldn’t rely on it as a telomerase activator.

  6. Medawar’s hypothesis (that what happens late in life is outside the influence of natural selection) has a subtle flaw in it. It makes sense for humans, where we have intervened, outside of genetics, and created “unnaturally” long life spans.

    But we also see aging in wild animals. And we know it is possible for creatures to evolve longer lifespans.

    That means “old age” for a creature is whatever evolution decides it is.

    So how can we say there is reduced selection pressure in “old age” when the definition of “old age” for a given creature is a product of selection pressure itself?

    This theory is circular; it sounds reasonable when you think of humans only. But when you consider creatures in general, it’s effectively just saying a creature doesn’t evolve to live longer than it has evolved to live – a non-statement.

  7. I remember reading about the OTHER study, the one with the whole plasma being infused, and seeing that one of the treated rats died. At first I was dismayed believing it was the e5 study. Then i realized it was that other study, and NONE of the treated rats died yet.
    Did one of the treated rats in the current e5 study really die? I never got an email update from that NTZ site saying that.

    • Yes, according to a comment from Askay, one of the treated rats died. Still, in the photographs, the rat looked younger.

      So, guess it depends on what a person’s desired outcome is: …. Feeling and looking more youthful or living a very long life.

      • … I don’t agree with that take.

        That one of the rats has died doesn’t mean that E5 does not achieve a significant lifespan extension: it’s worth noting that the one dead rat so far has lived to an extreme age for a rat. Also: even if E5 (or other present or future plasma derived products) are very good at fighting most aspects of aging, it doesnt instantaneously make it a panacea for every disease: Worth noting, in a recent long-term parabiosis study the authors found that while the treatment did result in median and maximum lifespan extension, there was a mismatch between life expectancy and epigenomic and transcriptomic rejuvenation and lifespan extension, mostly at the expense of (late!) cancer mortality. This doesn’t mean that the treatment doesn’t work; it does mean you cannot forget about dealing with other aspects of health and disease. Note that while the risk of cancer increases with age, you find cancer cases at all ages; it’d be unexpected for a blood factor treatment to instantly nullify said risk.

        Also, if anything, the fact that E5 is not perfect out of the box makes the results more robust in my eyes. I’d hardly expect an absolute success out of a first attempt, and I expect there will be further refinements to come from the field of plasma therapy (ie: what if the therapeutic plasma factors were used alongside the Conboy’s plasma exchange technique?)

  8. Very nice coverage. IRS can probably explain part of the changes. Was the Katcher’s study published/bioRxived after May 2020? I guess all animals died by now? Or…?

    Wonder whether in his E5 study vision of these rats changed/improved/stayed the same. How is DNA methylation changed in eye and liver? Would be pretty great to know how to demethylate ELOVL2 in liver at least….

  9. Hi, Josh, a couple of questions.

    1) In your view, is it concerning for the prospects of E5 that one of the treated rats in Harold Katcher’s E5 and longevity experiment has died?

    2) Like you, I am increasingly convinced by the merits of the programmed theory of aging. However, one question I have is how to square it with the finding by Sinclair et al. that a significant increase in double-stranded DNA breaks (DSB’s) by experimental rats leads to what looks like an accelerated aging phenotype.

    What could be the solution here? Hypothesizing that whatever clock is underlying the pro-aging program accelerates in the face of increased DSB’s? Obviously, a more general question would be how the program “knows” when to start aging the organism?

    • We now have updated records that almost as many control rats as test rats have died. I do not yet have detailed data to analyze. Yes, it concerns me. Does it mean that the Horvath clock can be set back without extending life expectancy? That would deeply upset my thinking.

      An alternative possibility is that there is a difference between male/female response to E5. Another possibility is that the rats in this experiment were treateed too late for E5 to be maximally effective.

      We need a lot more research in E5, and we’re not going to see it as long as the formula remains proprietary.

      • Of course the Horvath (or other methylation) clocks can be set back without reversing aging in a meaningful way.

        Take this scenario: you cause cell turnover rate to double. All other things being equal cells would have a younger methylation profile. But life of the organism would not be extended. It would probably be shortened.

        I am not suggesting this is what E5 is doing. It is just a plausible hypothetical for a way to quickly reverse the methylation changes that occur with age (particularly in cells that turnover regularly).

        • This is why we need actual life extension studies before we can draw any conclusions. And even then we’re depending on mice, not humans. It’s far from perfect, but in the longevity field where RCT’s on people are lacking, we’re going to have to deal with the imperfect .

      • Far from ideal, Josh. But data on the male study was far better. And many biomarkers were pushed in the right direction. Methilation was just one measure. And even a rapa longevity curve could look similar to E5 for the time being.

        We will need a decent (n/both genders) lifespan study, OFC.

  10. Regardless of whether methylation patterns have a strong causal effect on aging or are more like tree rings (merely providing a record of what’s happening), I am skeptical that methylation clocks will have much validity once interventions are applied that change them. The interventions could move the methylation patterns into configurations unlikely ever to be reached (even close) “naturally”, without interventions; and so, the clock number reported would just be the result of the mathematical model attempting an extreme out-of-distribution extrapolation.

    Perhaps a better approach, as far as reversing aging, would be to build a model to output a likelihood score for a given target pattern for x years in the past when also given the current pattern. The input to the model would thus be: (1) current methylation pattern, (2) a time x, and (3) a target methylation pattern for x years into the past.

    As an intervention is applied, if the methylation pattern starts to drift into regions of very low likelihood (when treated as the target pattern for x years into the past), other interventions could be tried to move it back into the regions of much higher likelihood.

  11. It will take a lot of fine-tuning/ trial and error attempts to get it right, that was expected.
    It would have been interesting to disclose the most probable causes of death for the reprogrammed mice…
    If they were tumour related, one could apply a supplement regimen including harmless anticancer (immune system boosting?) compounds from d3+cofactors to sulforaphane, melatonin or artemisia annua extract. And maybe also exclude Klf4 yamanaka factor from the cocktail ( oncogen)

    What is more, a simultaneous treatment to lengthen telomeres could make a huge difference ( eg lithium, astragalus extracts, hyperbaric oxygen)

    I don’t think a single intervention would be enough on its own.
    But scientists always apply single type interventions/ single compounds every time they conduct an experiment and thus reject various compounds and therapies, eg quercetin or fisetin, a few weeks ago ( Miller)

  12. Hi Josh,

    Naked mole rats do not age demographically:

    But they do age epigenetically:
    “Together, the data indicate that NMRs, like other mammals, epigenetically age even in the absence of demographic aging of this species.”

    It very, very strongly suggests that there is no causal link between the aging process and whatever methylation clocks measure.

Leave a Reply

Your email address will not be published. Required fields are marked *