At Ending Age-Related Diseases 2021, Jacob Kimmel of Calico Labs talked about epigenetic reprogramming, the Yamanaka factors, the dangers involved in restoring living cells to a pluripotent state, and a way forward with reprogramming that might not even involve the Yamanaka factors at all.
I want to share a pretty short story with you all today about some work we’ve been doing at Calico to explore the intersection of developmental gene programs and aging biology. Before jumping into our own work and some of the more recent data, I want to take us back historically to the origins of some of the questions we’re asking. These really go back to the beginnings of developmental biology, when microscopists could first observe Xenopus embryos forming speciated cells within an individual organism.
That headline question is really whether or not ontogeny can be reversed or whether or not it’s a one-way street. I think some of the best evidence and one of the most important experiments, honestly, in the whole of developmental biology, and also aging biology, addressed this question pretty directly way back 70 years ago, this was performed by a man named John Gurdon.
What he did was he took one of these embryos on the left here, one of these Xenopus embryos, and he removed its nucleus before it actually had begun the process of ontogeny. He showed that you could then take the nucleus from an adult, even aged, cell within a frog and transplant it into that embryo, and it would generate a whole young animal.
I want to just take a step back and think about that, that we’re able to take aged nuclei and that the exogenous signals within this enucleated cytoplasm can reprogram that nuclei back to a more youthful state and actually generate a whole young organism. I like to think of this really as an existence proof, that many of the features of aging can be reversed and are plastic under the right sort of exogenous signaling conditions.
This is certainly fascinating, and I think it makes for a lot of great philosophical conversation. I don’t think any of us as fully developed organisms are going to transplant ourselves inside anucleated eggs anytime soon, so it’s not perhaps the most relevant in terms of considering interventions for aging biology. It’s not the most relevant as a research tool either for that same reason.
Many of you are probably familiar, but much more recently, just now, 15 years ago or so, researchers at the Gladstone Institute showed that you could actually induce a very similar reprogramming process, taking a somatic cell in an adult organism and reverting it back to an embryonic-like state by overexpressing a few transcription factors.
These are master regulator genes that control the expression of many other downstream effectors, and these particular four regulators happen to be highly abundant in embryonic cells, This process has come to be known as induced pluripotency, generating IPSCs, or induced pluripotent stem cells. Fitting the magnitude of this discovery, it was rewarded with the Nobel Prize back in 2012.
This is all interesting developmental biology, but at this stage of this particular story, I think it’s apt to ask how this relates back to the biology of aging that we’re here to discuss today. That connection was made somewhat serendipitously by a number of very ambitious and bold researchers.
Over the past decade, the evidence that there is a connection here has just become overwhelming and difficult to ignore. Going back more than 10 years now, some researchers found that if you reprogram somatic adult cells, from aged and young donors, to the iPSC state and then differentiated those cells back out to a somatic cell state, that many of the gene expression differences that distinguished young and old cells are now ameliorated or masked, suggesting that this intermediary process, process of transiting through the pluripotent intermediary, actually repressed some features of aging.
This is, again, a curiosity. It’s quite interesting, but we’re not going to take cells in a mature organism and reprogram them to iPSCs; that would give you teratomas. Truly, the most shocking experiment in this literature comes from Carlos Belmonte’s lab back in 2016, where his group found a mouse where they have these four transcription factors knocked into a particular locus in their genome.
There’s a bit of engineering in play, but for all intents and purposes, when they give these mice doxycycline, a common antibiotic, these genes turn on. When they take the doxycycline away, these genes turn off. They found that by just transiently activating these genes in undulating waves in a periodic cycle, a couple days on, several days off, they could actually significantly extend the lifespan of progerioid animals.
That’s, again, quite shocking. It’s pretty interesting. As a developmental biologist myself, I never would have predicted this result to be true, had I not seen the data. While this is interesting, it’s in these progerioid animals, and as many of us appreciate, progeria is not aging.
The results that I come back to more than any others are actually some of the next results in the same study and many others that have emerged now, suggesting that the same sort of intervention, transiently activating these developmental programs, can actually restore youthful function in aged tissues, in otherwise wild-type mice.
Here, we see benefits to muscle regeneration, improvements in glucose tolerance after ablation of beta cells, and improvements in nerve regeneration in the ocular compartment. This is really just a sub-sampling even of the positive results seen across diverse physiologies in an organism.
When we take a step back and try and consider the many types of interventions that have been suggested for age-related decline across tissues, this is one of the few that shows this sort of beneficial pleiotropy having benefits in many diverse lineages of cells and tissues rather than in a single localized context.
These results are amazing, but there are still several questions left outstanding. At this point, I’d say the biology here has been explored at the level of phenomenology, but not necessarily mechanism.
Hopefully, we can just scratch the surface of some of that mechanism today, by no means dig into all of it, but just give you a slight foray in that direction. One of the first questions we had, having considered many of these amazing results, is that we were curious what happened to the somatic cell identities during this partial reprogramming process. There’s actually fairly scant evidence on to what degree, a transient intervention with these reprogramming factors would permanently alter cell identity in terms of the somatic cell identities that these adult cells actually started with.
Do they remember what identity they were and come back to that position, or is something permanently altered after one of these transient pulses? Another question we had was to ask whether there are any necessary or sufficient factors within this Yamanaka factor set that’s been used for these transient reprogramming interventions. Here, with these particular results, I’ve just shown, we’re intervening by overexpressing four transcription factors, each of which turns on many, many target genes.
It’s plausible that only a subset of that program is actually required for the rejuvenative benefits we’re seeing. We wanted to explore that question using a classical genetics approach searching for the necessity or sufficiency of different sub-programs within this cocktail.
Finally, in sort of a shot in the dark, we were curious whether this rejuvenative biology we’re seeing is unique to the Yamanaka factors themselves or whether other transient reprogramming interventions that temporarily differentiate a cell might have similar benefits. Without further ado, I’ll jump into the data. Before that, I want to highlight that this work I’m about to show you really is a collaborative effort across many groups at my current institution and could not by any means be done by myself alone.
Without further ado, then, we’ll jump into the first of these key questions and try and interrogate what happens to the somatic identity of an adult cell when you intervene with one of these transient reprogramming interventions. We wanted to ask what happens to this somatic identity, and we ideally wanted to capture many intermediary states a cell might adopt along this trajectory.
To do that, we set up a system where we could transiently activate these four Yamanaka factors using a lentiviral cassette. Similar to those mice I just discussed earlier, we can give cells doxycycline, and it will turn on these genes; we take the doxycycline away, the genes turn off.
We use single-cell RNA sequencing as our readout here, because the mRNA abundance profile provides a high-fidelity readout of cell state. It also allows us to capture a trajectory of intermediate states that might occur during this process, even in just a single-snapshot time point due to the heterogeneity of kinetics within a population.
We perform these interventions in a couple of different young-aged cell types here: adipogenic cells from the fat and some mesenchymal stem cells from the skeletal muscle. In both cases, we were able to capture these really rich trajectories of reprogramming that I hope you can appreciate here in these nonlinear dimensionality reduction plots.
Right off the bat, we wanted to ask a simple reproducibility question: can we find a restoration of youthful gene expression, as has been reported when these interventions are performed in vivo? Lo and behold, we could at multiple levels; we found that looking at single genes, there were thousands of genes that were restored from something like the normal level in an aged cell back to something closer to the normal level in a young cell.
We compare whole populations of cells and ask, effectively, how different are the young-aged cells in a control condition versus how different are control young cells from treated aged cells, we found that this Yamanaka factor treatment actually makes makes aged cells more similar to young controls.
This is pretty strong statistical result. We were convinced by these sorts of data that there was a there there. We are seeing a restoration of youthful gene expression in this process. We likewise see similar suppression of age-related gene expression programs in other cell types as well. We’re convinced this wasn’t necessarily just a lark or a result that occurs in only one of these contexts.
Now that we have these trajectories, and we’re convinced there is some restoration of youthful expression, we want to dig deeper into that identity question and ask, what’s happening to the somatic cell identity, is it suppressed, is it transient, is that permanent? To do so, we can, again, leverage these single-cell RNA sequencing data and fit a mathematical trajectory through the individual cells we have, representing a pseudo-temporal process, and ask questions about which gene programs are being activated or suppressed at different stages in this trajectory.
We find here a result that might be consistent with your a priori expectation that in most reprogrammed cells in our populations, we’re seeing a suppression with many normal somatic cell identity genes. Here, these are adipogenic genes, like LPL, and activation of some specific pluripotency program genes, like the transcription factor NANOG.
This is at the level of just individual genes, and we’re relying on some biological heuristics. We can also get a more comprehensive result using cell identity classifiers, which predict the identity of a cell using a reference atlas.
We find here a very similar qualitative result that the normal mesenchymal identity of these cells is suppressed the further we go into this reprogramming trajectory. The entropy, the sort of uncertainty about what type of cell this may be increases dramatically.
This is pretty interesting from a basic biology perspective, but a rightful question to ask at this stage would be, why is this important? Why does understanding this phenomenology help us in any way?
I’ll argue that seeing this transient suppression of cell identity and this activation of pluripotency programs actually suggests there might be a neoplastic risk here and something we should be concerned about going forward. This is based on some lineage tracing data from Jacob Hanna’s lab at the Weissman, where they performed an experiment where they transiently activated these programs for several days.
They found that after just three days of pulsing, like we’re doing here, you actually do get stably reprogrammed cells, and these cells reliably form very nasty-looking teratomas. This suggests to us that even in this transient activation context, you haven’t totally eliminated the possible risks that might be inherent in it in a temporary treatment like this.
We next want to ask whether the suppression of cell identity was going to be permanent. Once those cells have been projected along that trajectory, are they going to stay there, or do we think they’re going to revert back? Here we can turn to RNA velocity analysis, which uses a trick to infer splicing ratios of new RNAs versus maybe older mRNAs that you’ve profiled, and you can then try to infer not only where a cell is in the mRNA abundance space, but where it’s going.
Here, we can project these results in what I like to think are fairly pretty maps suggesting where cells in a given location might be headed with an arrow. These are nice and interesting results to look at, and we can interrogate which genes underlie those arrows and we find that these these velocity results are suggestive that cells are inverting the MET that occurs in reprogramming and they’re now activating an EMT. It’s suggesting cells are going back to their baseline state.
Quantitatively, we can also use phase simulations, a technique from dynamical systems to test what happens to one of these reprogram cells if we simulate its trajectory within this velocity field. In both cases, we find again, they’re significantly more likely to return to their baseline state than to stay within that reprogrammed condition. We think this suggests that this suppression of cell identity is going to be transient but nonetheless still risky in terms of neoplasia and something we should consider going forward.
Given those results, we next wanted to move on to ask these necessity and sufficiency questions to try and interrogate which aspects of this particular gene expression program are actually necessary for the rejuvenative benefits we’re seeing. In order to do that, we first wanted to develop some tools so we could run these experiments at scale and ideally expand this sort of platform in the future.
The tools we developed were set up in such a way that we could run an experiment where we deliver some viruses. Each of these viruses can carry a different transcription factor, in this case, a different component of the Yamanaka factor program. Then we can induce the expression and turn them off the same way with doxycycline.
The key here is that we’re able to read out which particular transcription factors each cell received using these single-cell RNA sequencing assays. Now when we collect each cell, we not only get its mRNA profile, but we collect which viruses were actually perturbing that specific cell.
In order for this to work, we actually have a few competing desires for our molecular biology. We need these genes to be temporally controllable, so we can turn them on and off, we need a constituent of selectable markers so we can sort the cells that get infected, and we need to have some way to detect them with RNAseq.
Each of these challenges was solved by a colleague of mine, Antoine Rao, who helped develop this molecular biology tool, where we have this responsive cassette that we can use to induce factors while at the same time having barcodes and constituative reporters that we can read out by single-cell RNA sequencing.
What this allows us to do is in silico, again, when we collect this data, we can demultiplex and actually see the results of many possible experiments, where here we’re testing different possible combinations that all occurred within the same well in a really nicely controlled way.
We wanted to then apply the system now that we’ve built it, and begin to get back to this question I started out on this section with, are any of these factors absolutely necessary, or some dispensable? To do so, we ran an experiment where we profiled all the combinations of the Yamanaka factors and a couple cell types.
Here, I’m just showing you the mesenchymal stem cells we talked about earlier. We can go ahead and do some sanity checks with ourselves and see that we are indeed recovering real biology with this demultiplexing scheme. Because the more transcription factors we’re detecting, the more suppressed the somatic cell identity program is, and the more activated a pluripotent gene program might be.
We next wanted to ask how distinct some of these perturbations were. If we find that two combinations give very different results, that suggests that we’d be able to tell them apart using something like a classification scheme. Whereas if two combinations actually yield a fairly similar outcome, then perhaps we wouldn’t be able to tell them apart, even with this sophisticated classifier.
We turned to a cell identity classifier tool we had developed called scNym, and we actually classified these perturbations as if they were cell types. What this revealed to us was that many of the higher-order combinations of Yamanaka factors are actually incredibly similar to one another. So similar, in fact, that even with a sophisticated deep learning model, we’re having trouble telling these profiles apart.
This suggests to us, just on its face, that we might actually see similar levels of youthful gene expression restored from some of these higher-order combinations and similar levels of identity suppression. We wanted to go ahead and quantify that in a more direct way, so we scored the activity of the age-related gene program here on the Y axis and the activity of the somatic cell identity program here on the X axis.
What we found is that most combinations of the Yamanaka factors that have at least two factors significantly suppress age-related gene expression and likewise do suppress the cell identity program that we thought was a little bit risky. As a surprising result, to me, this is one of the few scatter plots where I’ve been happy to see pretty little relationship.
It doesn’t seem there’s any significant correlation between the degree of identity suppression you’re seeing and the degree of youthful gene expression that’s being restored, suggesting that there might be an avenue forward to try and engineer slightly less risky strategies that still nonetheless restore youthful gene expression in a way that’s useful to us.
Now that we have this result suggesting that all four Yamanaka factors are not required, in fact, any subset of three seems to be fine, suggesting that no one factor is even really necessary. It made us curious whether or not the Yamanaka factors were required at all or whether the biology of play here was maybe more related to transient dedifferentiation and redifferentiation like we observed in some of those RNA velocity analyses we showed earlier.
To test this, we wanted to ask whether a multipotent reprogramming scheme that dedifferentiates a cell not all the way back toward its embryonic state, but toward an intermediary progenitor that occurs during development, might likewise have similar beneficial effects.
Here we turn to one of my favorite model systems, or at least my favorite model systems to read about. The experiment we ran was inspired by a salamander that actually has the ability to regenerate full limbs when they’re damaged. In this particular salamander, if you amputate a limb, the muscle fibers actually dedifferentiate. They go from these beautiful interstitial structures to actually separate out into individual cells, it’s quite amazing.
It was shown more than 20 years ago now that this same biology is actually still wired in two mouse cells if you overexpress the same transcription factor that gets overexpressed in this particular salamander, it’s called MSX1. You can likewise differentiate mouse myogenic cells back to a more primitive state.
We were curious whether this sort of multi-potent reprogramming scheme would likewise restore youthful gene expression. To set this up, we performed an experiment where we took aged myogenic cells, we treated them with an inducible form of this transcription factor. Eactly the same as we’ve done for the Yamanaka factors, we pulsed it for a few days, and then turned it off for several days and read out the effect by single-cell RNA sequencing.
In the cells, we captured a rich myogenic trajectory, these cells differentiate in their culture condition. It’s well known that aged cells differentiate less well than their young counterparts. Here, remarkably, we found using a pseudo time analysis that when treated, these transgene positive, or TG-positive cells here actually increase their ability to differentiate more like youthful cells, and they suppress the activity of an aging gene expression program extracted from historical data. This suggests to us that the Yamanaka factors may not be the only reprogramming strategy that can yield youthful expression benefits in the way that we’ve been discussing.
With that, I want to take a step back and just summarize what I think we’ve learned here. There are still far more questions than there are answers, but I do think we can take away a few pieces of biology from our analyses.
The first is that we do indeed see that partial reprogramming restores youthful gene expression but that somatic cell identity is transiently suppressed, and this potentially poses a neoplastic risk that we should consider going forward.
Secondarily, we found through pool screening techniques, that combinatorial complexity, the number of union factors you get in this Yamanaka factor pool, is the dominant determinant of the magnitude of an outcome you see in terms of restoring youthful gene expression and that no one Yamanaka factor is actually required for the beneficial results we’ve been discussing.
Finally, we’ve seen some experiments using multipotent reprogramming strategies. There may be alternative reprogramming regimes whereby we’re not targeting a pluripotent end state. We’re targeting some state in between, we’re targeting some particular cell state that occurs in development that potentially allows us to perform one of these reprogramming interventions with a lower potential for neoplastic risk.
With that, I want to thank all of you for your attention, I want to again highlight this was the result of a rich collaboration with many individuals listed here. I invite any of you all out in the audience to check out our results. All of our data are publicly available at the web address listed here.
My lab is also hiring to work on biology just like that I presented now. Please go ahead and check out our job board at Calico Labs careers, and I’d love to hear from you.