The good news is that the DataBETA project has found a home.  After several months of seeking a university partner, I am thrilled to be working with Moshe Szyf’s lab at McGill School of Medicine.  DataBETA is a broad survey of things people do to try to extend life expectancy, combined with evaluation of these strategies (and their interactions!) using the latest epigenetic clocks.  Szyf was a true pioneer of epigenetic science, back in an era when epigenetics was not yet on any of our radar screens. No one has more experience extracting information from methylation data.

DataBETA is just the kind of study that is newly possible, now that methylation clocks have come of age. Studies of anti-aging interventions had been impractical in the past, because as long as the study depends on people dying of old age, it is going to take decades and cost $ tens of millions. Using methylation clocks to evaluate biological age shortcuts that process, potentially slashing the time by a factor of 10 and the cost by a factor of 100.  But it depends critically on the assumption that the methylation clocks remain true predictors of disease and death when unnatural interventions are imposed. Is methylation an indicator, a passive marker of age? Or do changing methylation patterns cause aging?

Two types of methylation changes with age

Everyone agrees that methylation changes with age are the most accurate measure we have, by far, of a person’s chronological age—and beyond this, the GrimAge clock and PhenoAge clock are actually better indications of a person’s life expectancy and future morbidity than his chronological age.

Everyone agrees that methylation is a program under the body’s control. Epigenetic signals control gene expression, and gene expression is central to every aspect of the body’s metabolism, every stage of life history. Sure, there is a loss of focus in methylation patterns with age, sometimes called “epigenetic drift”.  But there is also clearly directed change, and it is on the directed changes that methylation clocks are based.

But there are two interpretations of what this means. (1) There is the theory that aging is fundamentally an epigenetic program. Senescence and death proceed on an evolutionarily-determined time schedule, just as growth and development unfold via epigenetic programming at an earlier stage in life. Several prominent articles were written even before the first Horvath clock proposing this ideas [ref, ref], and I have been a proponent of this view from early on [ref]. If you think this way, then methylation changes are a root cause of aging, and restoring the body to a younger epigenetic state is likely to make the body younger.

(2) The other view, based on an evolutionary paradigm of purely individual selection, denies that programmed self-destruciton is a biological possibility. Since there is a program in late-life epigenetic changes, it must be a response and not a cause of aging. Aging is damage to the body at the molecular and cellular level. In response to this threat, the body is ramping up its repair and defense mechanisms, and this accounts for consistency of the methylation clock. In this view, setting back the methylation pattern to a younger state would be counter-productive. To do so is to shut off the body’s repair mechanisms and to shorten life expectancy.

So, if you believe (1) then setting back the body’s methylation clock leads to longer life, but if you believe (2) then setting back the body’s methylation clock leads to shorter life.

I think there is good reason to support the first interpretation (1). Epigenetics is fundamentally about gene expression. If you drill down to specific changes in gene expression with age, you find that glutathione, CoQ10=ubiquinone, SOD and other antioxidant defenses are actually dialed down in late life when we need them more. You find that inflammatory cytokines like NFκB are ramped up, worsening the chronic inflammation that is our prominent enemy with age.  You find that protective hormones like pregnenolone are shut off, while damaging hormones like LH and FSH are sky high in women when, past menopause, they have no use for them. There is a method in this madness, and the method appears to be self-destruction.

Until this year, I have been very comfortable with this argument, and comfortable promoting the DataBETA study, which is founded in the premise that setting back the methylation clock is our best indicator of enhanced life expectancy. The thing that made me start to question was the story of Lu and Horvath’s GrimAge clock, which I blogged about back in March. 

The GrimAge clock is the best predictor of mortality and morbidity currently available, and it was built not directly on a purely statistical analysis of direct associations with m&m, but based on indirect associations with such things as inflammatory markers and smoking history. (This is a really interesting story, and I suggest you go back and read the March entry if you have not already. The story has been told in this way nowhere else.)

(Please be patient, I’m getting to the point.) Years of smoking leave an imprint on the body’s methylation patterns, and this imprint (but not the smoking history itself) is part of the GrimAge clock. I asked myself, How does smoking shorten life expectancy? I have always assumed that smoking damages the lungs, damages the arteries, damages the body’s chemistry. Smoking shortens lifespan not through instructions imprinted in the epigenetic program, but quite directly through damaging the body’s tissues. Therefore, the epigenetic shadow of smoker-years that contributes to the GrimAge clock is not likely to be programmed aging of type (1), but rather programmed protection, type (2).

For me, this realization marked a crisis. I have begun to worry that setting back the methylation clock does not always contribute positively to life expectancy. The canonical example is that if we erased the body’s protective response to the damage incurred by smoking, we would not expect the smoker to live longer.

The bottom line

I now believe there are two types of methylation changes with age. I remain convinced that type (1) predominates, and that setting these markers to a younger state is a healthy thing to do, and that it offers genuine rejuvenation. But there are also some type (2) changes with age—how common they are, I do not know—and we want to be careful not to set these back to a younger, less protected state. 

The methylation clocks promise a new era in medical research on aging, an era in which we can know what works without waiting decades to detect mortality differences between test and control groups. But it is only type (1) methylation changes that can be used in this way. So it is an urgent research priority to distinguish between these two types of directed changes.

This is a difficult problem, because the obvious research method would be to follow many people with many different methylation patterns for many decades—exactly the slow and costly process that the methylation clocks were going to help us avoid. My first hunch is that we might find a shortcut experimenting with cell cultures. Using CRISPR, we can induce methylation changes one-at-a-time in cell lines and then assess changes in the transcriptome, and with known metabolic chemistry, make an educated guess whether these changes are likely to be beneficial or the opposite. As stated, this probably will not work because methylation on CpGs tends to work not via individual sites but on islands that are typically ~1,000 base pairs in length. Perhaps changes in the transcriptome can be detected when we intervene to methylate or demethylate an entire CpG island.

Perhaps there is a better way. I invite suggestions from people who know more biology than I know for experimental ways to distinguish type (1) from type (2) methylation changes with age.