Clinicians Need to Know: Evolution of DNA Methylation

Joseph M. Raffaele, M.D.

So it’s great to have you on the show, Ryan. Really looking forward to talking to you about all the really interesting work you’re doing at TruDiagnostics. And we’re gonna get into DNA methylation and relationship to telomere biology, which you are also doing some fantastic, predictive work there. But you’re a young guy, and, you know, you’ve co-founded this great company. I first met you when you were working at Tailor Made. Tell me a little bit about your journey from pharmacy through to molecular diagnostics.

Ryan Smith

Yeah, absolutely. It’s been a wild one that’s for sure. I think that, you know, my background even before that is, sort of, biochemistry, undergrad, a little bit of medical school. And it really got to the clinical portion of medical school. Passed USMLE step one, did pretty well, but got to the clinical stuff and just couldn’t imagine doing it forever. And so, through sort of pure serendipitous luck, I ran into this idea of compounding pharmacy and particularly, you know, using a lot of my background to specialize in peptide and protein synthesis. And so that’s what we really did at Tailor Made Compounding, is sort of bring a lot of peptides to market for the first time for clinical use. And that introduced me to a whole new realm of medicine that I didn’t know existed. 

This idea of preventative, you know, optimization, sort of functional medicine. And you know, the medicine I had been previously, you know, exposed to was really dealing and treating sick patients. And I always felt that that was, you know, difficult because I didn’t really think I was impacting change. I think that I was, you know, staving off disaster, but not positively changing medicine. And so I got to sort of, be exposed to this world for the first time via Tailor Made Compounding. 

And with that was really, really exciting, ’cause there’s a lot of new and innovative products coming out that, and as we learn more about just the baseline science, which, you know, continues to surprise me every day, the idea of this preventive medicine became, you know, even more ingrained into my mind. And so, you know, as we considered how to vet some of these other therapies that we were doing, some of these new and innovative molecules. I was always looking for different biomarkers, which might be able to provide a lot of information and make sort of these investigations, a little bit more feasible. And that’s when I was introduced to this idea of epigenetic methylation in biological aging, through epigenetic methylation. Where you might even be able to do, you know, instantaneous, you know, a 40 year placebo controlled trials through these predictive algorithms, which then can tell you how you’re changing and mitigate risk. 

And that to me, was something that was incredibly exciting and something that I wanted to be involved in. Particularly whenever it started to become very clinically relevant in, 2019, with the advient of some of these interventional trials, like the trim trial, which showed changes in biological age, which then can, you know, theoretically show that we can reduce those risk factors, as we age. And so that’s sort of the trajectory that I’ve taken. And now I really committed to this idea of preventative medicine. And also this idea that age, is one of those markers that we can hopefully treat, as a preventative marker, since it has such a high correlation to all of these chronic diseases.

Joseph M. Raffaele, M.D.

Yeah, I mean, that’s absolutely true. The aging process is the bedrock of all the chronic diseases and to be able to measure it is, sort of a holy grail, within the field. You need that technology to vet things like the peptides, and all the other therapies that are coming down the line. But for our listeners who, you know, I’m sure most everybody’s heard of DNA methylation, but maybe a little bit of a primer on what exactly it is, what we’re looking at and how it works, before we get into sort of how it’s utilized, and what the really interesting stuff you’re doing at TruDiagnostics is.

Ryan Smith

Yeah. So the way that I generally like to describe DNA methylation is as talking about every cell in your body has that same baseline DNA sequence. But obviously, you know, our skin cells are a lot different than, you know, the hard cells we have in our body. And so what influences the sort of the baseline DNA to the change in what actually is occurring. And one of the biggest changes that these epigenetic mechanisms, which include things like acetylation, you know, mRNA’s, and then also lastly, methylation. Which the methylation is typically put on certain parts of our genes to silence transcription of that gene. So to make sure that some things are turned off. And as mammals, we have some unique, methylation sort of signatures. 

Sort of by default, we have much more methylation in our system than other animal species. And one of the things that happen as we age, and these methylation changes can sort of change in very, very predictable fashion. Generally as we age, more of our genome becomes hyper methylated. But in particular, there are certain spots of the DNA that are highly associated with that change as we age. Not just with hypermethylation, but sometimes with hypomethylation, where some of those methylation markers get taken off, and then therefore the gene transcription is more likely to be turned on. 

And so, I think that this is just from a platform standpoint, this is how our body regulates and uses all of the capacities of our DNA. And, methylation is one of the many types of epigenetic changes, which can occur to really turn off or silence that sequencing. And through some of these investigations that have been happening, now that we’re getting better ways to look at methylation, so much like the, you know, the human genome project, whenever it first started, doing methylation work was very expensive and it didn’t yield a whole lot of results. 

But now as the technology has been advancing, we’ve been able to really see a lot more from these markers. And due to the other advancements in technology, such as computer learning and artificial intelligence, we’ve been able to now make sense of it as well. So it’s not just this abstract data, but correlated to certain types of health markers, or even non-health related markers, even more trait markers, or other things which then can give us really good insight into what’s happening with an individual, and then how we might be able to mitigate change. And so this is sort of a new field that’s evolving due to breakthroughs into the actual bench, top testing, as well as the interpretation of this through computer learning. And so a lot of really exciting things, we sort of have this whole new book that we get to interpret and read, and we’re trying to really create that Rosetta Stone to understand exactly how to read it and what that information means.

Joseph M. Raffaele, M.D.

Yeah, I mean, when I was first introduced to the field and thinking about it, you know, and certainly back when the genome was first decoded, we thought that that was gonna get us all the information, but you know, there is absolutely good information from that, but really the changes in gene expression, you know, what Michael Fossil likes to say is, you know, what makes a nose that different from a toe, is sort of the more important kind of thing that we’re learning. As a biomarker of aging though, another thing that I think maybe we’ve talked about in the past in some of our conversations is, predicting chronological age is sort of important, but for forensics. 

And when the first biomarkers came out with Horvath, predicting very highly R squared of like .96 or something. and then the Weidner clock where you only need three CPGs to have, you know, very high predictive. But predicting chronological age, unless you’re in forensics where you want to know, you know, what the ages of a piece of tissue from a dead body, is not that important, because if it’s really good, then you just use chronological age. So, one of the changes I think, that’s made me more interested in it, is now, now these data sets, these clocks are being trained on different kinds of datasets, not trying to predict chronological age. And that’s I think, kinda some of the interesting stuff that you’ve been doing, I think what makes them more interesting clocks. Why don’t ya tell me a little about, what kinds of training is being done, and what you’re looking at?

Ryan Smith

Yeah, absolutely. You know, I think that is, you know, well-put. The idea of knowing your chronological age. You can really just ask, right? And, so again, chronological age is still highly correlated to aging, so not completely useless, but with that being said. And I should also mention, changes in predicting this chronological age, also then might intuitively tell us things that changed some of the phenotypes of aging as well. And so, you know, it was really exciting for its development, in sort of showing this high link, but with the new developments are definitely about all training these markers to predict outcomes or phenotypes. And so, that sort of separates what they call a first-generation versus these second generation clocks. Where the first generation were trained to predict chronological age. 

Whereas the second we’re trained to predict some other type of outcome. The two best examples are Morgan Levine’s PhenoAge, which used really 10 blood-based markers to create this idea of a phenotype of sort of health. And then secondarily that grim age by Dr. Horvath, which is trained against time till death, to be able to predict a sort of lifespan. And so those are really, really exciting and really make this a lot more medically relevant. Because if we want to stave off you know, those predictions, then we can look at ways to change those, and give us a little bit more of a health outcome.

Joseph M. Raffaele, M.D.

Yeah, that’s a great explanation of it. Those two clocks are sort of the most important second generation clocks. Maybe also for our audience, a little bit more detail what you mean by when something is trained on the date. I used that phrase, we use the phrase, but maybe our listeners don’t know exactly what that means. When I first started reading about it, it was a little complicated as well.

Ryan Smith

Definitely. You know, I’m not sure that I even have the best words to describe it. But when I say trained, I generally mean, looking for correlations, right? With computer learning systems. So the idea that, you know, we’re gathering large and robust amounts of data. So even with, you know, some of the first or second generation clocks, they were using 450,000 or 850,000 locations on the DNA. Each of those locations gives you a number, essentially a percentage of methylation. However, whenever you come up with the final algorithm, which is used to predict those outcomes, you might use something like 500 or 1000. 

So you’re significantly reducing the things which matter most, which have the most predictive capability. And so, we sort of training the mathematical algorithm then to predict that outcome, which you’re sort of correlating it against. And so, that is sort of what it’s meant by training. And now though, this has taken on sort of a life of its own. As the aging concept has proven, that there are some clear methylation signals, which then can link to other outcomes. 

So, one of my favorites, actually, even probably something very topical at this point, it’s just a few days ago, a week ago, there was an article published with methylation algorithm that could predict schizophrenia, in over 80% of individuals. And so, that was really exciting, especially with some of these mental diseases, where, you know, we’re looking at blood methylation. That’s another really important, I would say concept as well, because every cell has a different epigenetic signature, as I mentioned, right? The skin cells and heart cells are going to behave and have different epigenetic expression. 

So, whenever you train these things, you have to train it on a cell type specific, or at least use the same cell type investigation method to create the algorithm as you’re going to use it in clinical practice. And so, for us to be able to see, you know, diagnosis of schizophrenia with high levels of accuracy and blood-based methylation, it sort of comes up with this idea, that we can really train every outcome that we really wanted to look at against these methylation marks. We might have different degrees of specificity and sophistication. 

However, the idea is that this as a platform, has so many robust markers, and the computer learning is getting, you know, so, I would say well-trained, that you can really predict multiple different outcomes. I think aging is definitely one that is probably the most exciting, because instead of just a correlation, it looks to be maybe a causation at this point. And, and so that’s exciting. I think from a mechanism of action, we still don’t know enough to sort of tell you what that cause is, or that causal mechanism. But I think that, that’s why aging is probably the most exciting. But this epigenetic platform in general can do everything from, you know, predicting a athletic performance to telling you your immune cell subsets, to telling you, you know, your list of exposures that you’ve had across a lifetime, from certain chemicals or diet nutrition or et cetera. So this will be a platform that continues to evolve, but aging looks to definitely have a special place, in this idea of epigenetic methylation.

Joseph M. Raffaele, M.D.

Yeah, so that’s great, I was reminded of sort of the same issue in telomere like measurement, when you’re talked about, which cell type that you’re measuring it in. So, you know, we know that every cell has telomeres and they get shorter with age. But what we’re measuring in a blood test is, you know, the telomere length of the white blood cells. Depending on which assay you use, it could be, you know, PBMC’S or lymphocytes and granular sites. But there is pretty good predictive, you know, algorithms, well not algorithms, prediction of outcomes in Chronic diseases, Cardiovascular disease, Alzheimer’s disease, Osteoporosis, et cetera, from telomere length, even though it’s not the cells of those tissues. 

So the same thing, I guess you’re saying you can do with DNA methylation age. Now I know that Horvath has looked at the DNA methylation signatures in individual tissues, and he has the pan-tissue clock. And then he has, you know, individual clocks, but there’s pretty, pretty good correlation. And so what you’re saying with the schizophrenia is that, it’s fascinating that we can look at blood and see what’s happening in the brain, in terms of complex mental illness, you know, that is pretty fascinating. But in terms of the sort of the degree of accuracy, the clocks that predicted chronological age had, you know, an R squared, which is in the .9 range, which is amazing. These other clocks for predicting things like, lymphocytes subsets, and other things they’re low, right? I mean, what’s the range for those?

Ryan Smith

Yeah, definitely lower, you know, it all depends on the algorithm, you know, in terms of which ones you’re using and looking at those R-squared squared, but there’re definitely lower and they’re getting, you know much, much more improved as the time goes on, you know? As I mentioned, you know, the schizophrenia predictor has a very, very high correlation value. Whereas, you know, some of those immune cell subsets have wide ranges of error, depending on how you’re using it and what data set you’re looking at. And so those are getting better, you know, and the benefits of those particularly for immune cell subsets is that, you know, one of the things that is really helpful in all of these investigations is you need to train it to a phenotype. And oftentimes that means outcome phenotypes.

Joseph M. Raffaele, M.D.

Right.

Ryan Smith

And so, that data that can be hard to get, right, if you need to really have patient samples from, you know, 20 years before they develop the phenotype. And that data is really hard to get, especially as we talk about things like immune cell subsets, where you might have to store PBMC’s and high volumes and stability can be difficult, and you have to collect it in the right way and then get into storage within a few hours. And so some of those things like, you know, immune cell subsets have been, you know, really exciting because our immune system changes according to a lot of different diseases. 

But now by having a surrogate marker that just looks at DNA, we can then apply this to a lot larger data sets from biobanks that go back 40 or 50 years. And so the idea is, it might not be as accurate as traditional immune cell subset testing at this time, but it’s still exciting because we can do it in retrospect, ’cause DNA is a little bit easier to store. There’s no stability concerns. There’s no types of draw concerns. And then we’re able to look at some of those data to look at how the immune cell subset is affecting, you know, disease prevention and risk. 

And so that’s exciting, even though they’re not as accurate at the moment, they can still be, you know, sort of helpful clinically. I always use some good examples from some of our practitioners where we do some of their immune cell subset testing. And just to clarify on what that is. We’re sort of telling you the percentage of your different types of immune cells, and we’re not telling an absolute concentration, just telling you a relative percentage. And so one of the things we also do there-

Joseph M. Raffaele, M.D.

That’s an interesting point. Sorry not to interrupt. That’s an interesting point. So you can’t give an absolute count in which you would think would be difficult. You can give a percentage, so you can give a ratio of CD8’s to CD4’s, you know, helpers as suppressors to helpers. You do it for natural killer cells and B lymphocytes and things like that. I’m just curious, what kinds of things are they looking at? Is this because those cells have certain gene transcription, then you look at the methylation patterns of those?

Ryan Smith

Yeah, absolutely. So the idea is you, in order to create these methods, de-convolution is a very, very exciting thing, not just for, you know, lymphocyte blood based tissue, but for any tissue, right? If you were even looking at a cancer cell for instance and look at, you know, regions that might have this genetic predisposition or polymorphism versus others. And so, so this whole field of de-convolution really at first establishes a reference where you look at an individual cell type. So we might look at natural killer cells and then look and compare that to eosinophils. And we might say between these two investigations, what is different? You know, and what is different to a high degree? And so, whenever we define what is different, those are called differentially methylated regions. 

And the good news is for a lot of these immune cell subsets, these differentially methylated regions are robust enough to be able to tell, you know, how much refining of this region versus of this region. And then being able to combine that, to get an idea of those percentages. And, again, more developments are being made to break those down into more advanced subsets. You know, senescent T-cells is one that is, you know, we’ve talked about before and is very exciting, to be able to finally quantify senescent T cells in a way that is a little bit more reliable than some of the other methods at this point and a little bit less expensive and easier to perform. 

And so that’s exciting. And so we’re looking at those differentially methylated regions, and that’s really honestly what we’re doing for most investigations. We’re defining the areas for a certain type of phenotype or a certain identity, that is different than what we would traditionally expect.And by looking at such, you know, 900,000 spots, those can be relatively robust. We might have 100 to 7,000 spots, which are differentially than the methylated regions. Then we can break down subsets instead of just knowing CD8 cells, we can know all the different types of CD8 T cells or you know, those types of things. And so really in order to do that, you just need those individual cell types sorted and profiled individually, and then having those same identifications done on the larger tissue as a whole, or the larger blood sample. And so that is becoming much more robust and there’re actually are ways to quantify the absolute quantity as well, using some new plasma technology, but we’re not there yet. But the idea is that hopefully it will be there, here very, very soon, which is again, a hopefully another exciting development, which just goes to show you all the broad level impacts that this is a biomarker, methylation’s a biomarker can have on sort of the healthcare system as a whole.

Joseph M. Raffaele, M.D.

Well, that’s a great segue into sort of a more clinical sort of look at the application of this in sort of day-to-day use. As I said, I wasn’t that excited about using DNA methylation in my practice because, you know, when I started ordering it from a company that was looking mostly at, predicting chronological age, I didn’t see a lot of variation. Which just shows how good it is as a forensic test, but then it’s not that useful. So, your company is looking at it somewhat differently, in a couple of ways, which I’ll let you explain in a minute, but you have a true age, which is you can explain exactly what that is it. And then you can break that down into intrinsic aging and extrinsic aging, which brings in this concept of the changes in immune subsets that can alter, you know, your DNA methylation pattern. And then at some point we want to talk about the really exciting thing that you have, which is, you know, the rate of aging that you’ve gotten from the Dunedin and co cohort in New Zealand. As a clinician, tell us how you would use it, why your tests should be used in a practice so we can, you know, give us some take home Monday morning, use of it.

Ryan Smith

Definitely, so it all comes down to, you know, I think that everyone who hears about the testing is sort of impressed with the correlation to age and the correlation to predicting age outcomes. But, the one biggest limitation, and the thing that I think is even still to this point, preventing wide-scale adoption is the question, what do I do about it? Right? You know, how do I change this metric? And you know, what’s the use of testing, if I can’t make a difference, right? If you can’t change something, then does it really matter if you know what it is. And that has been one of the biggest restrictions of this testing. And one of the reasons that still break this down into intrinsic versus extrinsic aging. The idea there is that extrinsic and intrinsic gauging had been looked at in many, many datasets from the very beginning. And so we know more than most other algorithms, what changes these metrics. And so we have ideas of treatments. Unfortunately today, there’ve only been five interventional studies, which look at a baseline measurement of epigenetic age, a treatment, and then an outcome. And those studies have various degrees of robustness as well. 

The first one, as I mentioned, only had nine patients, right? You know, none of them have had you know, I should say, none of the non epidemiological ones have had, you know, over 40 patients. And so these are still early investigations and so they have some limitations to them. And so that’s definitely something that we’re trying to do, is to add context to expectations on interventions. Everything from bariatric surgery to plasma exchange, to stem cell therapy, to even just simple things like diet, and nutrition, exercise, stress management, things that are more accessible to people without even having to have medical intervention. 

And so, we’re definitely trying to build that up and to be more robust. But that’s the reason we break it down is because, there are different risks and there are different treatments associated with intrinsic versus extrinsic aging. So by looking at an individual, you can sort of start to say, I might recommend these interventions versus these. And it allows you to be a lot more personalized. You know, I think as you alluded to, the algorithm that we are, I think by far, the most excited about, is almost what I would consider, a third generation algorithm.

Joseph M. Raffaele, M.D.

Wait a second, can I just interrupt you this,

Ryan Smith

Please.

Joseph M. Raffaele, M.D.

that’s the most exciting thing I want to get into, but, can you explain what the differences between intrinsic and extrinsic aging?

Ryan Smith

Oh, absolutely

Joseph M. Raffaele, M.D.

And, I know that I have patients and I myself have a significant difference between my intrinsic and extrinsic aging and you know, which one’s more important? Or not so much, what does each one tell you? And I think you’re right, that you know, different interventions will move the needle on each of those differently. But how do you go about deciding? I mean, the original clocks were sort of…. talk about that.

Ryan Smith

Yeah, yeah, definitely. So the idea about intrinsic age is that, it is sort of a baseline fundamental process of aging. And what I mean by that is it’s not confounded by some of the effects we might see in the organism as a whole, that also changes with age. One of the biggest effects is probably, you know, a topic I’m sure you discussed on with some of the other people in this series, which is immunosenescence, right? This idea that as we age, our immune system gets worse. And, you know, just being a, you know, still in a pandemic, I’m sure it’s another thing that people can relate to, it’s the reason we unveil these vaccine distributions in elderly populations first, because they’re more at risk, ’cause their immune systems are not as robust. And what that also means is that the cells in our blood, those immune cells that constitute the majority of the things that we’re testing, also change in concentration. And so the idea is that, you know, we might have more natural killer cells, but less effective natural killer cells. 

We might have more senescent T cells, less naive T-cells. And if we’re not accounting for that change, we can get some weird readings. And so the intrinsic age, sort of uses the estimation of immune cell subset percentage, to then control for any change over time. So it is able to sort of factor out those immune system changes. But sometimes we don’t want to do that. Sometimes we do want to get a good idea of how the immune system is changing and how that’s impacting this idea of chronological age prediction. Particularly in the events of you know, death prediction or overall longevity, and also in cancer risk. Which also should be relatively intuitive, because the idea that the immune system can help clear or prevent cancer is one that’s I think, widely accepted. 

And the link between T-cell subsets in aging and longevity have also been well-connected. So the extrinsic age is tied probably a little bit more closely to things that relate to the immune system. And so, as a result for most people who have healthy and robust immune systems. That metric is going to be a little bit lower probably, than it would be for the intrinsic, which we almost always find, is a little bit more accelerated. And so, with that being said, it’s hard to say which one’s more important. I think it also depends on, as an individual what you’re trying to prevent against or what some of your other predispositions might be. 

But the idea is that it gives us a better idea of this full clinical picture to make the appropriate decisions, based on the way that you want to control your own health and address your own preventative medicine. And so, I think the more information is better. And that really helps us break it down into understanding how the immune system is working, versus this fundamental baseline process of aging, which, you know, in, in sort of discussing that, I always like to use the analogy that, you know, Dr. Davidson Sinclair uses in his Lifespan book. Where he talks about the information theory loss. 

Where, you know, when we were first born, all of our epigenetic expressions, perfect. What should be turned on is turned on. What should be turned off, is turned off. But as we age, some of that regulation becomes a little bit less ideal. And that leads to this progressive loss of information, which might also lead to the progressive loss of function, which is sort of that definition of aging. Which is again, why we think that this might be a causal mechanism for the aging process, rather than just a correlative process.

Joseph M. Raffaele, M.D.

That’s a great explanation. I was thinking when you said that, we just published a study looking at the effect of T8 65 on immunosenescence, as defined by the CD28 negative cells. Those are the suppressor cells, CD8 positive cells that have lost the expression of this important market allows them to proliferate briskly when they encounter their antigen. And that number increases as we get older, but particularly when we have chronic viral infections and particularly CMV. I was thinking that, you know, it will be interesting to see, since you’re able to do that kind of deconvolution, whether or not we would pick up the same signal, just doing extrinsic DNA methylation testing, if we substituted that for the immune subset panel. And that might be an interesting thing to do.

Ryan Smith

Yeah, very interesting that you mentioned that actually, because just even very recently, there was a algorithm that can predict CMV infection. So, you know, I.

Joseph M. Raffaele, M.D.

Good.

Ryan Smith

So, you know, I’m sure that-

Joseph M. Raffaele, M.D.

Just send me that paper.

Ryan Smith

Yeah, I will. So you know, I think the idea is that maybe you can even look at some of those differentially methylated regions for CMV infection, compare it to those immune cell subsets, you know, differentially methylated regions, and maybe even have, you know, some idea, if the same thing that you saw with your most recent publication is also reflected in some of those epigenetic methylation, and maybe even postulate a mechanism, that is previously unknown for the reason TA-65 is having such a positive benefit. And so, we can actually even do that analysis, I think without having to actually get into any specific data sets, but just sort of looking at regions and locations of methylation, to get a good idea if there, might be a reason to explore further.

Joseph M. Raffaele, M.D.

And you just need a serum for that, right. What do you need?

Ryan Smith

Correct. 

Joseph M. Raffaele, M.D.

Yeah

Ryan Smith

Yeah, and so,

Joseph M. Raffaele, M.D.

You might, have to talk about that. And the other reason I bring that up is that, a very powerful predictor of mortality in older patients, is the ratio of CD4 to CD8. Your two, four and six year mortality has looked at in the Octo and Nona trials in Sweden, is, you know, 40, 50% increased, if you are in the, what we call the immune risk phenotype below one CD4 to C8 versus above one. And, you know, you’re right. I mean, if you can predict those subsets, then that would be interesting thing to look at as well. 

I wonder if they have serum from those trials, maybe ask them about that. That would be well.- So yeah, so there’s, you know, great work that can be done with that. But what your test can also do, which is fascinating, is give a rate of aging. And I was incredulous when I first heard this, ‘Cause I’m sort of like, no, I’m not great at calculus, but I think that if you need more than one time point to get a rate of aging. And in fact, you know, explain how you’ve done that, and what benefit it has, and how, you know, you can have two people with the same DNA methylation age, either intrinsic or extrinsic, but have different rates of aging at a particular time point and why that is, and why that could be really useful clinically.

Ryan Smith

Yeah, absolutely. This is the one that I’m most excited about. And I sort of referenced it earlier, you know, almost calling it a third generation clock ’cause it’s not looking at the overall organism, right? It’s looking at a set point in time. And that can be useful for a lot of reasons. One of the big reasons is that up to 40% of some of these other previous clocks, including the second generation clocks like GrimAge and PhenoAge, 40% of that can be decided based on hereditary factors. You know, epigenetics is, even passed down through our DNA, which is both exciting and scary, I think to some degree. 

You know, for instance, you know, Jews who have experienced, you know, periods of famine, that epigenetic signature can actually be even found several generations later. And, you know, a lot of people want to be in control of their own destiny. Some people might have to work harder and epigenetic change is still doable, but this idea that hereditary factors can play a big role, might give some people a baseline context, which is not the most ideal, or for instance, people who are trying to turn around their lifestyle from an aging perspective who are now, you know, had some, you know, some introduction to this and why this is important, and, you know, we’re trying to lose weight and trying to eat healthy. 

The idea is that if we took a snapshot of their overall health, they might’ve already accumulated some of those epigenetic markers, which increase their overall biologic age. But if we were to look at their instantaneous rate, what we would see is, is significant improvement, right? We would probably see that there, you know, the rate of aging has now decreased from what they were previously. And that’s exciting because we’re able then to vet things like interventions, even with, you know, healthy people to say, you know, how does XYZ, you know, affect me versus affect you. And so, that’s really exciting in practical application, but it’s also exciting in terms of how they actually created this metric. ‘Cause as you mentioned, you know, in order to create a pace, you were an average, you know, sort of velocity, you do need to, you know, an idea of, you know, sort of where you’re coming from and where you’re going. And, so this is a really unique study, that probably won’t be replicated because it started really in 1973. 

And so in 1973, in a town called Dunedin, in New Zealand, they started measuring over a thousand patients with many different phenotypic biomarkers at age three. So things like cognitive processes speeds, everything from gum health, from dentistry exams, to you know, functional brain MRIs, to you know, standing and walking and physical function measurements, as well as even skin appearance. And so they, they measured a lot of these different phenotypic biomarkers. And then over the course of now, you know, sort of now that these people who started at age three are now over age 45, they were able to get this idea of trajectories of aging as it related to those consequences. 

So instead of looking at like the second generation clocks, it really large datasets where the samples we’ve been banked in the past, they were able to track the same patients from three years of age to 45 years of age, and quantify this idea of their trajectory to these aging phenotypes. And then creating an overall score of that phenotype and regressing that against DNA methylation. And what they found was sort of astounding. 

One of the things that I love to talk about this study on is that, even things like mental processing speeds at age three were predictive of health outcomes at age 45. Which goes to sort of, you know, I would say, which makes us maybe believe that aging rates are even set in adolescents, even in infancy. Our aging is determined and can be mitigated and managed. This whole idea that, early lifestyle factors can influence your rate of aging. And then that influences your risk for these disease predispositions, sort of flips maybe our concept of aging on its head a little bit, where we sort of talked about people when their, you know, in their forties, you know, starting the aging process. 

I think there’s evidence now to suggest that it’s not that way, that aging process can even start even earlier than that. And so that’s exciting too, but the idea that you can quantify this and predict health outcomes by an instantaneous rate becomes more exciting, because it is linked to so many different health phenotypes. And also, I think, as we’ve talked about before, oftentimes you have this juxtaposition of quality of life versus lifespan. And I think that this is a really interesting way to tie all of these together, because the rates of aging were actually predictive of things like, you know, sarcopenia in balance testing, grip strength, cognitive IQ, and memory processing speeds, and even facial appearance in terms of how old you look. And so all of those things we mentioned don’t necessarily, you know, tell you how long you’re going to live, but they can improve your quality of life. And that’s also exciting. So, I don’t, think that they’re mutually exclusive. I think that you can consider, you know, treating aging and then having improvements in both the quality of life, as well as, your overall health span. 

And so I think that that’s exciting as well. And so that algorithm is very unique with the data set, that’s going to be very hard to replicate. And it’s already been started to look at, in some type of interventional clinical trials. The one that is most notable is the Calorie Study, where they did 20% caloric restriction and showed significant decreases in the rate of aging over the course of two years through the cork restriction, meaning caloric section might be a good way to improve aging rates across many different populations.

Joseph M. Raffaele, M.D.

So, the rate of aging, is looking at, I guess, different CPGs, than the other clocks are looking at, because it’s picking up something that’s happening right now in terms of changes in gene expression, because you’ve stopped smoking or because you started exercising. And that’s how it works, that you’re able to do that because this is unlike the vast majority of clocks that have been trained, this is trained on longitudinal data.

Ryan Smith

Yeah, absolutely. And I’m sorry, I didn’t, answer that question first, but that’s exactly right. One thing reportedly as stresses, these are different algorithms. So they develop on different data sets, they developed on different populations. And so, they’re also trained against different outcomes. And so you would expect that there would be some overlap and there is, there is some overlap with these individual CPGs, but there’s not much. However, you still see that, for instance those people in our datasets who improve their rates of aging over time, also tend to improve their overall biological age over time as well. And so, they definitely look to be at least, you know, correlative, even with very few overlaps in the actual things we’re measuring. And so generally what we often try and say is encourage people to really look at, the rate of aging and because it’s a little bit more sensitive as well. 

You know, one of the things, one of the metrics that we often look at to determine if a test is reliable or it’s a good clinical biomarker, is the interest sample variability, right? If you test the same sample twice, you know, how accurate is that that reading? And that’s been definitely one of the biggest limitations, again with epigenetic testing as well. Even the Dr. Horvath original 2013 algorithm, had a mean absolute error of around 1.9 years. Which means that unfortunately, if you were to have an investigation with that metric in a period of 1.9 years, it becomes very much more difficult to see a statistically significant result or a change and learn what is actually changing that metric. 

It’s different with this DNA and Pollium, ’cause it’s much more accurate from an inter sample variability standpoint. You know, one of the interesting things is Morgan Levine from Yale, who created PhenoAge with Dr. Horvath at UCLA. She just published an update to all of the traditional algorithms, called the Principle Component Analysis Algorithm, which significantly improved the reliability and decreased the mean absolute error of those algorithms, which is a huge step forward. But it still, even with those improvements, the ICC, or sort of this measurement of intercept , is still best in that you need an algorithm. And so, by sort of focusing on that rate of aging, you can maybe imply that you would then see improvements in the other overall biological ages, and it’s more sensitive for more frequent investigations. And so as a result, it is maybe a little bit more exciting and probably more useful clinically than some of those other aging algorithms.

Joseph M. Raffaele, M.D.

So I have had a number of patients who’ve gotten the reports back and I’ve got mine back. And when you report out the rate of aging, one, is sort of average it’s, aging one year for one chronological year. And you go from .6, I think to 1.4, when you say it’s a pretty tight in terms of its variability. Would you say that if I went from .8 to .7, that, that would be something that is meaningful? What is the actual plus or minus on it? We’re talking about fractions here.

Ryan Smith

Definitely. Yeah so, that is a significant change, absolutely. If you’re changing by around, you know, .1 point, that would absolutely be significant. And really, if we see any change above really .05 or .06, that is when we consider it a significant change, which unfortunately that might seem, you know, like in the overall biological aging scheme, a minutia, but, it means it is still significant. And so, seeing change like that is still very, very, you know, helpful. And we can say that we have seen that with some of our interventional trials. And so, you know, from a P value calculation point, that would be significant.

Joseph M. Raffaele, M.D.

Yeah, so in the calorie trial, were able to correlate that where you saw that… it was a two-year trial, you said? So..

Ryan Smith

Ah, correct.

Joseph M. Raffaele, M.D.

You did rate of aging at, how many time points?

Ryan Smith

So unfortunately I’m not sure. So the majority of that data was done and I wish I could take credit for this entire project, but, we’ve been lucky enough to license this from a joint collaboration from duke, Dr. Dan Belsky from Columbia, and then the University of Otago, which is based in Dunedin New Zealand. And so, we unfortunately had not had access to that calorie data only Dr. Belsky has. And so I’m not quite sure about some of the data sheets going into there. But I do know that the outcome of that was showing a significant reduction over the course of that two year timeline.

Joseph M. Raffaele, M.D.

Yeah, so, what would be perfect is if, you got the rate of aging of .9 and then a year later, your biological age was, you know, not a year older, but .9 older, that would be, perfect.

Ryan Smith

Yeah.

Joseph M. Raffaele, M.D.

But you know, just for the listeners, you know, biomarkers of aging, even the Horvath clock where it’s 1.9 years for any biomarker of aging for that kind of variation is actually very good. Other clocks, other biomarkers that are sort of an R-squared of .4, .5, which is still a very good one, like arterial stiffness or pulmonary function. It’s up closer to .7. I mean, you’re talking about more like three to five years, which is frustrating when you’re looking at trying to see the effect of effective therapies, because, you don’t know for a couple of years and unless, you know, what the trajectory is. 

I had patients that come back to me and said, I did my baseline telomere length with another doctor. And then I started something and, you know, six months later it was a year or better. And I’m like, yeah? It doesn’t mean anything. There’s at least three years of variation in that. And you need more time points for that sort of thing, which is you know, frustrating. 

But with that rate of aging, being a metric, really telling people, you know, whether or not they’re intermittent fasting or they’re, you know, three to five day fast or their NAD or whatever combinations that they’re, that’s what I like about it for clinical practice, is we’re doing stuff for our patients that is on the cutting edge. But if we have a metric that is quite well validated, like yours and in aggregate, whatever we’re doing is moving it in the right direction. I think we can have good confidence and say that we’re doing a good thing for the patient, particularly for checking other biomarkers and they’re going in good directions as well. And pretty good for the newer things like, you know, The Senolytic therapies that some people are trying, that would be very interesting to see how those, are affect the extrinsic aging. 

But for, you know, NAD therapies, hyperbaric oxygen, all these sorts of things would be very interesting to see what’s happening with the rate of aging. And I’m sure we will be seeing that, with your technology. Let’s pivot for a second. So we have a little bit more time to talk about your prediction of telomere length, and where that’s going, since I’ve been measuring telomere lengths in my practice for 14 years now. And it has the same issue of some variability, but when you get a pretty good idea over time. What’s, the predictive ability of that? And how did you go about doing that?

Ryan Smith

So, we base most of our work off of a Dr. Horvath algorithm. From the looks at sort of, several different publicly available data sets, that have both methylation data and telomere length data. And then uses sort of the same type of, you know, regression modeling to then pick out the CPGs, which are predictive of telomere length. And so, this is something we’ve been able to do and replicate with some of our own data sets, in telomere length, As is it’s a relatively common metric with a lot of the people that are doing our tests. And so, what we’ve sort of found particularly even with Dr. Horvath found, is that there is a relatively high correlation between telomere, or I should say, estimated telomere length via methylation and age. So, in Dr. Horvath’s, particularly there was around a .7 R squared value for correlation from telomere length estimation to age, which shows it to probably be a little bit more highly correlated, with certain types of outcomes, right? Or I should say, more correlated with age. And then in addition to that, compared to traditional telomere length testing in those subsets. 

And so that is also one of the caveats, you know, there are a lot of different types, and methods of telomere testing. There are many types of subset breakdowns, things like critically short telomere length or individual immune cell subtypes telomere lengths, which weren’t conducted in these studies. And so, comparing, you know, the best in telomere testing to the best in methylation testing, or methylation estimation via telomere testing, is still a hard thing to do. It’s hard to compare those apples to apples, but what they had shown essentially, is it that it was more predictive of some outcomes, than traditional telomere length testing. 

And that was particularly congestive heart failure, cardiovascular disease, time till death, and then also sort of a correlation with smoking history. And so, that gave us reason to believe that, we could also then train these methylation marks to predict telomere length. You know, I think that there’s definitely more work that needs to be done. Particularly looking at things like critically short telomere length, or particular immune cell subset versions of telomere length. But the idea is that hopefully we’ll be able to train this even with a methylation prediction algorithm and do one test which can mean a lot of information, it’d be highly accurate versus just a, you know, having to do multiple different tests to get the same amount of information.

Joseph M. Raffaele, M.D.

Yeah, I mean, that’s absolutely, you know, really fascinating. I take it that was the correlations with Dr. Horvath’s work, was mostly QPCR?

Ryan Smith

Correct.

Joseph M. Raffaele, M.D.

And so that’s, yeah, that’s one of the, I think, you know, less accurate methods of doing it. It’s good for large observational studies in large batches, but hopefully over time, we’ll be able to accumulate some data, cause all the telomere lengths measurements we’re doing with the Physio Agent and TruDiagnostics will be able to get some flow fish data, I hope. The critically short ones is, really interesting. And I don’t think anybody really offers that clinically now. I mean, not critically short, Lifeline offers 20%, but so, you know, we’ll have to see whether that comes down the pike at some point. Well, you know, it sounds like, you know, you’re working in this field and you know, one of the best algorithms is probably that the rate of aging. Where do you see the field of epigenetic analysis going at this point? Where do you guys want..

Ryan Smith

Yeah, I Definitely. Well, I would think I would feel remiss, if I didn’t go ahead and point out one of the probably, if not the largest area already for epigenetic methylation, which is in taking plasma from the blood and looking at stage zero or stage one, or stage two cancer detection of tissue specific origin. Recently, a test came out just over the last few weeks, from Grail called Galleri, which is able to read over 27 different types of cancer from just a couple, you know, milliliters of plasma. And that’s really exciting because again, obviously cancer is one of the biggest, you know, risks burdens of everyone in these developing worlds. And to be able to detect it early, is a huge clinical benefit. And so that is a huge area of epigenetic evaluation, which is really exciting, because what they can do with that cell-free DNA, that’s found in the plasma, is track it back to the tissue of origin. 

And so you don’t have just you know, a marker that says you have cancer, you have a marker that says you might have cancer in this tissue. And that can be very, very helpful, even just knowing where to look. And so we’re not doing much cancer work, and the money going into that is pretty incredible. And really everyone should have a lot of hope for how we treat even cancer preventatively, I think, but in terms of areas that we’re investigating, I think for us, we are looking to build out some of those investigations into immune cell subsets, and then also different types of diseases, you know, things like cognitive diseases that are associated with aging, like Alzheimer’s, or Parkinson’s, or even just dementia or cognitive impairment that happens with aging. 

We’re very excited to learn more about those and have some exciting things going on to look at those. We’re interested in looking at cardiovascular disease metrics or even early diabetes markers, that might encourage, you know, some type of lifestyle or medication change before we develop diabetes. And so for us, I think that the idea is to gather as much data as possible, to find out where we see the highest correlation between methylation and disease. And then hopefully investigate those with even further, with a little bit more vigger, and to find out how we can predict different outcomes. We’re also very excited about things like the exposome, right? Looking at a list of methylation values and being able to determine what type of exposures you’ve had across your life, to be able to then control for some of those exposures. But then also see if those exposures might impact the outcomes that we see for these people. And so, and then lastly, even things like pharmacoepigenetics, everyone is very familiar probably with how your DNA might anticipate, how you metabolize a drug, or if you’re have side effects with the drug. 

The same thing happens actually with epigenetics, where we might be able to even predict, you know, if you’re going to have a side effects with Metformin, or if you’re going to have A1C responses with Metformin, which has actually already been published, that algorithm’s already been published by a university in Sweden. And so building and expanding on those ideas of just how to better handle clinical practice and getting as much data as possible then to, really build out the suite of this testing. There’s so much information that can be found in the epigenome that we think that this test can, really reduce costs on a lot of lab testing, but increased value. And I think that that’s where we see this being, hopefully going in the future, in really every disease direction you can imagine, I would imagine, would be impacted by this at one point.

Joseph M. Raffaele, M.D.

Yeah, and certainly it’s allows you to personalize medicine, you know, tailor your approach. I’m fascinated by the concept that, you could predict reaction to a medication, epigenetically, which maybe implies that if you make some changes, you can change your epigenetics and get rid of that adverse reaction toward, or lack of effectiveness. And I’m sure those have trials haven’t been done yet, but that’s really exciting as well. I mean, it’s an amazing technology that you guys are at the forefront of and offering clinically now. Is there any thing you’d like to tell our viewers? I know you can tell, you know, how to get to the TruDiagnostic test, anything else you want to say?

Ryan Smith

Yup. You know, I would say that this is a new and developing technology and I want to give everyone also the skill set to adequately differentiate, if they’re using a good platform, right? And I think that we’ve already mentioned some of the things there. I think that particularly you want to be able to use published algorithms, right? Published algorithms are very, very important because that’s how we actually relate the change in those methylation marks, to diseases outcomes, right? Via those studies. And so I would highly recommend, you know, doing some work to vet, if your algorithm has been published, I would also vet the tissue of collection, right? You know, we’ve mentioned, we only do blood. Saliva is becoming a little bit more feasible, due to some additional studies, but again, pay attention to the collection method. 

And then also again, pay attention to the amount of data that you’re getting, right? You know, the amount of new developments in this field are growing exponentially and being able to have a data set, which you can then look at for the future and beyond, I think is also important. If you’re just looking at, you know, 1000 to 2000 locations, then you might not have the same information, sort of across time, or even being able to look back 10 years and say, hey, I have these markers, which I’ve been since changed. And so, those are the three things that I would just, you know, hope every consumer, or every practitioner would incorporate into their critical evaluation of any of these technologies as they continue to expand.

Joseph M. Raffaele, M.D.

Yeah, I’m absolutely in agreement with all those and those they make good sense. I think it is though ready for prime time, as a clinical tool, particularly if you’re applying therapies that, you know, aren’t looking at specific diseases, can’t be measured. You know, for instance, we can’t measure NAD levels very easily right now, but if you’re raising it, if you’re taking an NAD supplement, which a lot of people are already or taking Metformin, even though you don’t have a hemoglobin A1C above, you know, even 5.6, you know, some people are taking it wonder even lower, to know whether or not you’re getting a benefit, particularly from this highly reproducible rate of aging, is a way for clinicians in my field and hopefully, you know, all primary care and other physicians, at some point, will be able to use tools like this to give therapies, A, if they’re needed, B, and know if they’re working in, and even as you said, predict beforehand, who’s going to have an adverse or beneficial response to it. I think the future is really exciting and I can see why every time I talked to you, you seem so excited about what you’re doing. I know why now. So thank you for coming on and giving a great introduction to this and getting into some of the details of this stuff. And looking forward to talking to you again.

Ryan Smith

Yeah. Thanks for having me on, I wish you the best of luck with the other speakers and look forward to watching.