Dr. Junyue Cao is a professor at the Rockefeller University, and his lab develops ultra-high-throughput single-cell technologies and applies them to the biology of aging. In a recent paper published in Science, his team used a technique called EasySci-ATAC to profile chromatin accessibility in about seven million cells from 21 mouse tissues across three ages, producing what is probably the most comprehensive epigenomic atlas of mammalian aging to date.
This atlas paints a detailed picture of how the body’s cellular landscape changes with age. For instance, about a quarter of all cell types shift significantly, many of these shifts are coordinated across organs, and males and females often age quite differently at the cellular level. Understanding those changes is a prerequisite for designing interventions that target aging itself rather than individual diseases. We talked to Dr. Cao about what his data tells us about the nature of aging, where the data points in therapeutic terms, and why he thinks that aging having certain program-like features is actually good news.
Let’s start with how you ended up doing what you’re doing. Do you feel passionate about human lifespan, or is it just a research subject for you?
I’ve been a big fan of aging biology since high school. It’s one of the most critical questions to answer. On one side, there’s aging itself – something people have been interested in for thousands of years. On the other, aging is associated with many chronic diseases, which means that if you target aging, you’re not only targeting one molecular process but potentially slowing down many diseases at once. That’s why I went to Peking University to study biological science.
At the time, I was very naive. I thought the major challenge was that we didn’t have a drug that could reverse aging. So, my first thought was: maybe it’s a drug design problem. That became my undergrad research – designing small chemicals to target specific inflammatory factors. But as I went through my studies, I realized the problem wasn’t that we couldn’t design drugs. The problem was that we didn’t know what the targets were. Even with a great drug design platform, if you can’t find a target that really addresses aging, it doesn’t help.
So, I switched back from chemistry to biology to identify those targets, and I moved to the US. I started as a research assistant working on three signaling pathways: the unfolded protein stress pathway, mTOR signaling, and stress signaling. All three are associated with aging.
I spent three years on that, and it was very helpful for understanding the system, but at the end, I realized that these signaling pathways aren’t directly linked to aging at the organism level. They operate inside individual cells, and different signaling pathways have different activities in different cells. Their changes don’t affect all cells – they may just change a specific group. That makes it hard to directly link them to the aging process.
I felt the need to fill this gap: first identify which cells actually change during aging and then target those cells and their molecular programs. The question became: how can we identify what’s happening in individual cells across the entire system?
That’s why I started my PhD in Jay Shendure’s lab at the University of Washington, which focused on developing highly scalable single-cell technologies. At the time, commercial platforms could only profile several thousand cells. But a mammalian body contains billions of cells – if you only sample a few thousand, you can’t capture all the different cell types and know which ones are really changing with aging. My goal was to expand from several thousand cells to millions.
And in this paper from February this year, you ended up with seven million, right?
Yes, but that was after about ten years. We developed the first technology that could give you one or two million cells in a single experiment – about a hundredfold increase over commercial platforms.
We continued improving the technology throughput so we could scan even more cells. Last year, we published a paper where we sequenced over 20 million single-cell transcriptomes, and in this paper, we profiled the single-cell chromatin landscape of seven million cells, fully devoted to aging. It’s maybe not a traditional path – I started from drug design, which is the most downstream part, and kept moving upstream.
You kept moving upstream, which makes a lot of sense. So, let’s talk about the technique. You used ATAC-seq rather than RNA-seq, which most atlases rely on. Why was that your choice?
We also use RNA-seq a lot in the lab. Last year, we published a large-scale single-cell RNA atlas of aging across many different tissues. But gene expression changes are downstream – they tell you which genes get changed, but not what the upstream regulators are. The advantage of ATAC-seq is that it tells you which regions of the genome are accessible, which parts are open for the cell to read. Basically, you can imagine the genome as the recipe for making dishes, and ATAC-seq tells you which pages of that recipe are accessible to the chef.
And RNA-seq is like a snapshot of what’s being prepared in the kitchen right now.
Exactly. So, using this information, we can not only see what has changed but also infer the transcription factor activity – which TFs are driving these changes. There’s another advantage: the optimized version of single-cell ATAC-seq that we developed can scale to scan the entire organism – over 20 different tissues in a single experiment. That’s relatively hard for RNA-seq, especially because some cells have low RNA content. Senescent cells, apoptotic cells, certain low-content cells are very hard to capture with RNA-seq. But ATAC-seq is based on genomic DNA, so it’s relatively easy to capture them. That’s also why we could use this to scan the entire system.
Let’s move to the core findings. First, you identified around 1,800 cell subtypes, and some of them are more vulnerable to aging.
This is potentially the unique feature of the study. Each of these 1,800 cell states has its own unique molecular features. It’s just like a human society – different individuals have different jobs, different proportions of the population work on different things. The cells in your body are a cell society. We visualized the aging process not as the disruption of a specific cell type or organ, but as the disruption of the entire cell society.
It’s like you can zoom in or zoom out, and then you have this bird’s eye view.
Exactly. On one side, we can see which cells are more vulnerable during aging. It’s not one or two cell types – it turns out that about a quarter of the global cell population, a quarter of the “jobs,” are highly vulnerable to aging. Their abundances change significantly. The rest are generally more stable. So, it’s not that every cell gets changed, but it’s not just a few either. And now we know which ones are more vulnerable, so we can think about ways to target them.
The second interesting feature is the very strong coordination across different organs. For the same cell lineage – endothelial cells, fibroblasts, immune cells – we consistently see the emergence or depletion of the same cellular state during aging, even though they’re from different organs. This includes not just immune cells that can easily transport across organs, but also endothelial cells and fibroblasts that don’t really move around.
This cross-organ coordination seems pretty heavily driven by immune cells and by the response to various cytokines. How does it relate to the concept of inflammaging?
It could be related to the inflammaging process, or it could be related to other factors – we’re not sure yet. We also see similar transcription factor activation across the same cell lineages during aging in different organs. Maybe the same group of cells in different organs are vulnerable to similar damage, resulting in a similar response, or they could be more vulnerable to the same external inflammatory signals.
But I think this generally means that instead of targeting each cell type separately, in theory, they may be driven by the same upstream signals, either internal or external, that we can target to rescue them all. What exactly those signals are – that’s what we’re currently working on in the lab.
There’s a lot of immune cell dynamics in your paper. I personally think immune aging is underestimated as a factor in aging in general. What can your study tell us about its role?
Immune cell dynamics is definitely one of the focuses. The good thing with this data is that we can capture the global map of immune cells across many different organs – from the very early progenitor cells in the thymus to many other T cell states in different organs. That’s not easily captured in studies that only look at one or a few organs.
One surprising thing: we generally assumed that since immune cells circulate through the body, they should show the same dynamics everywhere. We do see a lot of shared dynamics across organs, but most immune cell aging-associated expansions or depletions are still limited to a few organs – they’re not truly universal.
We also identified some less-characterized immune cell state changes. For example, we found the well-known depletion of naive T cells, but also a group of less-characterized T cell subtypes that show depletion in the intestine and other tissues. Because we profiled all immune cells across organs, pooling them together gives higher power to detect these less-characterized populations.
Another surprising finding is the very strong sex specificity. Some specific T cells and B cells show quite different dynamics in males and females. They mostly change in the same direction, but the scale is quite different. And it’s not that one sex always has higher expansion of inflammatory immune cells – each sex has its own preferred immune cell dynamics.
For example, we found a type 17 T cell expansion that’s preferentially expanded in males. When we checked the literature, this has also been reported in humans, where researchers attributed it to male-specific behaviors like smoking. But in our mice – who obviously don’t smoke – we still see the same dynamics.
We also saw aging-associated B cells, two different groups of them. One was well-characterized before, but the other wasn’t, and we confirmed it across multiple organs. Both show female-specific expansion, and their gene markers relate to autoimmune disease, which potentially correlates with female vulnerability to autoimmune conditions.
The sex dimorphism is indeed one of the most interesting findings. It sits well with what we know about sex differences in aging and, importantly, how geroprotectors work differently in males and females. There might be some serious translational consequences here.
Exactly. One thing I should mention is that about one-third of the aging-associated cell states we identified show significant differences between males and females. One potentially translational example: in the kidney, we see a specific reactive inflammatory epithelial cell state that emerges during aging, and it expands in a very female-specific way. This relates to the fact that female animals are more vulnerable to kidney dysfunction at end of life. And when you check the gene features of this cell state, they’re also associated with human disease.
This is a study of the aging process, but when we check the aging-associated cell states and their gene features, they also link to human chronic diseases. More analysis is needed to confirm these links, but this study can be used to infer the targeted cell types for chronic disease. I think it will give people a list of critical cell types to focus on for the future.
You show a clear switch in transcription factor motifs with age – inflammatory factors opening up and stemness factors closing down. How do you interpret this?
This is potentially one of the most critical findings of the paper. Previously, when we thought about aging, we imagined it could be random damage or stochastic shifts in the epigenetic landscape – something hard to rescue. But from this data, we can zoom into the changed cell types and see that there’s a highly programmed process at work. These aging-associated chromatin changes aren’t only seen in one cell – they’re consistently observed across many cells in the same animal, in the same organ, across different organs, across different individuals, and in both sexes.
This means there’s a lot of programmed process that could potentially be targeted. That’s why we did two levels of analysis. First, the internal molecular program: by identifying the TF regulators, we found the stemness-related transcription factors closing down and the inflammatory factors opening up, along with many other transcription factors that haven’t previously been linked to aging. Importantly, we found not just individual transcription factors but also transcription factor interactions – it’s not just the activity of a single factor that gets strongly altered, but their combinations.
Second, we identified the external signals. We integrated our data with a cytokine response dataset to nominate the top cytokines that could be driving the cellular and molecular state changes – factors like TNF-alpha and interferon signaling.
You said something really important about this being a programmed process. I wonder what this means for the major theories of aging – hyperfunction, damage accumulation, and so on.
I think all our results suggest that aging is not just a stochastic process. It’s a highly programmed process. There are specific groups of cells that change, and they reproducibly change across many different individuals. These cells are not randomly chosen – they have unique molecular programs that determine whether they expand or deplete. At the cell level, it’s a programmed process.
Related to that, in a paper we published last year using single-cell RNA-seq, we saw the same pattern and found that these changes are highly constrained to specific time windows. In a very early phase, one group of cells changes; at another time, another group changes. They’re highly coordinated across organs and across time.
When we check these cell groups across multiple animals at a given stage, they all show the same depletion or expansion. It’s very reproducible. And these cells that are distributed across different organs – even though they may seem unrelated – they show very sharp changes in the same time window. They don’t deplete gradually across the lifespan in a linear way. One group shows sharp depletion between, say, three and six months. Another group depletes sharply between six and twelve months. Another shows strong expansion after middle age. They don’t change linearly – they change in waves.
I think our data do not argue against classic theories such as damage accumulation or hyperfunction, but they add another layer. There is certainly stochastic molecular damage during aging. However, what we see is that the organismal response to that damage is not random. Specific cell populations expand or deplete reproducibly across animals, across organs, and within defined time windows. This fits very well with the idea that aging involves regulated biological programs becoming maladaptive. For example, immune activation, tissue repair, inflammatory signaling, and remodeling programs may be useful earlier in life, but later they can persist or become overactive — which is related to the hyperfunction view of aging.
So, when I say aging is “programmed,” I do not mean that the body is intentionally programmed to die. I mean that the cellular remodeling of aging has program-like features: it is reproducible, temporally organized, and cell-type-specific. Damage may be stochastic, but the downstream cellular response is highly structured. In that sense, our work suggests a synthesis: aging may begin with accumulated molecular stress and damage, but its effects are executed through specific gene-regulatory programs, cell-cell interactions, and population-level changes. That is important because it means aging is not just something that happens passively to every cell. It is a coordinated remodeling process, and some of its key regulatory nodes may be targetable.
That actually sits well with some recent research in humans, where we also see those waves of aging rather than a clean curve.
Yes, it clearly correlates with human changes. This supports the idea that it’s a highly programmed and coordinated process, similar to early developmental processes.
Personally, I’m not sure if it’s good news or bad news that aging resembles a program. On one hand, there’s something to act upon, rather than a multitude of cell-specific stochastic changes. On the other hand, this program might be deeply rooted and hard to override.
I think it’s a great relief. If it were just random changes in every cell in your body, it would be very hard to reprogram. But because it’s a programmed process, we can reprogram it. As long as we identify the cell types and understand the program, we can target them.
And the coordination across different cell states is another relief – it means they may be driven by the same upstream signals, so we may not need to target each of them separately. If we can identify the upstream signaling, we may be able to rescue multiple cell types at the same time. This gives me hope that we’re working in the right direction.
Of course, this requires a lot of follow-up studies. We end up with maybe around a hundred different cell types and several hundred cell states that strongly change with aging, but we don’t know which ones are real drivers and which are just passengers. We need to identify the functionally important ones. That’s why we’re currently developing highly scalable perturbation platforms – so we can perturb many different cells and molecular processes to identify the real drivers.
But if aging is at least partly a coordinated response to stochastic damage, how can we stop or reverse it without losing those protections? Is this something like a Catch-22?
I don’t think it’s really a Catch-22, because the protective and pathological sides of the program are separated in time and in cell state, not in the signal itself. An acute inflammatory response, for instance, is essential – it clears damage, recruits repair, and then resolves. The problem in aging is not that the response exists; it’s that it fails to resolve, and specific cell populations get stuck in an activated state. We don’t want to cancel the response. We want to restore its dynamics – the off switch as much as the on switch.
The data give us several handles that don’t sacrifice protection. The maladaptive states we see (aging-associated B cells, reactive inflammatory epithelial cells in the kidney, certain activated T cell populations) are largely absent in young animals. They’re the chronic byproduct of the response, not the acute response itself, so we can target those cells specifically rather than block the upstream pathway globally. Many other changes are depletions rather than expansions, where you don’t have to suppress anything, but to replenish! And because the program unfolds in distinct temporal waves, intervening early can prevent the chronic, locked-in version from forming in the first place, which is much easier than reversing it later.
So, the program is not a single switch we either keep or destroy. It is a multi-cell, multi-time-window remodeling process, and we now have a map of where and when it goes wrong. That tells us where to intervene and where to leave things alone.
One particular thing I wanted to ask about: retrotransposons, which are a very hot topic in geroscience. You saw retrotransposon elements becoming more accessible with age while conserved developmental enhancers become less accessible. And this transposon derepression is probably related to inflammation through pathways like cGAS-STING. What are your thoughts?
This finding supports the current direction of the field – that transposable elements could be targeted to see whether we can alter some signaling in the aging process. This activation of transposable elements correlates with increased inflammatory signaling in aging cells.
Our dataset could also be used to ask whether this activation is universal across all cells or whether specific cell types are more vulnerable and show more activation, linking the molecular changes in transposable element activation to vulnerable cell types and phenotypes. We haven’t done a lot of follow-up on that part yet, but it could be a very exciting direction.
But retrotransposon derepression doesn’t quite look like a useful program, rather like an unwanted side effect.
It potentially depends on how you define “program.” If it were a fully random process, we’d expect chromatin to open and close randomly, and we wouldn’t see a strong enrichment of openness specifically in these transposable element regions. But here we see that aging-associated open regions are enriched in transposable elements, which indicates a strong bias for these regions to become activated in aging. We can regard that as a random stochastic process, but we can also regard it as a partially programmed process.
Let’s talk about translational implications. Is there something in this dataset to go after therapeutically?
It’s not directly linked to clinical application; it’s more about understanding basic aging mechanisms using animal models. Many aging-associated changes are shared between mouse and human, including many aging hallmarks like stem cell depletion and signaling dysregulation. So, the dataset could be used to identify aging-associated cell states that relate to chronic disease in humans.
Also, this could serve as a foundation for evaluating anti-aging interventions. Previously, we knew some interventions work and some don’t, but now, using this platform and technology, we can directly tell you which specific group of aging-associated cell states is targeted by a given intervention. If two interventions target distinct groups of aging-associated cell states, it means we can combine them for a more powerful combination therapy.
One very important thing is that we see cell population changes happening in a very temporal manner. Some cells are already depleted before middle age, which means that if you want to tackle aging, you may need to start early, before the depletion is finished. This is also consistent with current data showing that interventions have a very strong time-dependent effect – if you start at a very late stage, the effect is much weaker than if you start earlier. This data can also tell people the potential intervention time windows.
Although now we have cellular reprogramming that might help recreate depleted cell populations.
Exactly. And now we can tell which cells you need to replenish.
Because of this monumental work, how have your own views on aging changed? What priors have you had to update?
One critical update is the realization that aging is not a single pathway or a single cell type problem. It’s a society. It’s due to the disruption of the cell society. This is really helpful for thinking about the aging problem, because it’s a system process. For any intervention we test in the lab, we always want to see how it affects the entire society instead of just looking at specific organs.
This gives you a global view of the system so that you don’t miss something. When you study the disruption of the society, you have to consider every “job” – every cell type – because each could be important for maintaining homeostasis.
I think it also demonstrates the importance of developing highly scalable perturbation technologies so we can systematically test all these different aging-associated molecular programs to see which ones are really important. This is actually what we’re mostly focused on now: combining highly scalable profiling with high-throughput perturbation to identify the real drivers.
View the article at lifespan.io
