Scientist Stories: Anna-Katerina Hadjantonakis, Challenging Dogma in Stem Cells

I'd like to thank the organizers for giving me the opportunity to talk about my work. Okay, so I'm just— by way of introduction, I'm a bona fide developmental biologist. My lab's research interests really focus on how cells make fate decisions in tissue context. The tissue we're particularly interested is the endoderm, that is the embryonic progenitor of the respiratory and digestive tracts and associated organs. We use the mouse as a primary model, and we supplement our studies in stem— embryo-derived stem cell models, and I think many of the findings that I'm going to show you today will perhaps in the future need extrapolation to the human. So what I want to do to begin with is just take you back to sort of bring you to some of the earliest fate decisions taking place in mammalian embryos, and this is the blastocyst stage. The blastocyst is a universal stage of mammalian development,. It takes 3 to 4 days to get the blastocyst after fertilization in the mouse, and about a week or so in the human. The blastocyst comprises 3 lineages: the trophoblast, epiblast, and primitive endoderm. And essentially, the specification of the blastocyst lineages really involves segregation of the epiblast, the future sort of lineage that will give rise to the future embryo, from these extraembryonic lineages, the trophectoderm, which, or trophoblast, will give rise to the trophectoderm, which is the fetal portion of the placenta, and the primitive endoderm that will give rise to these extraembryonic yolk sac membranes. For over a decade now, my lab have been very interested in how a progenitor population called the inner cell mass will differentiate into two sister lineages, primitive endoderm and epiblast. And this will really be the focus of parts of my talk. I'm going to talk about sort of the dogma of these lineage segregations, the idea that the epiblast exclusively gives rise to the embryo proper, and I'm going to suggest that in fact perhaps primitive endoderm cells may give rise to at least part of the sort of gut endoderm, and I'm going to show you some new experiments that are unpublished on that. I'm also then going to tell you a little bit about our recent forays into querying lineage plasticity. As cells make these decisions, how concrete are they, and can we take cells out of perhaps the constraints of the embryo and ask those questions? And some sort of cell reprogramming between primitive endoderm and epiblast fates. So a couple of years ago now, it's about 3 or 4 years ago, we started to exploit single-cell transcriptomics to investigate some of the decisions that we're very fascinated by taking place in the embryo. So when looking at preimplantation, we can actually plot force-directed layouts of cells isolated of embryos. These are inner cell mass cells, and we can see these cells differentiated down an epiblast or primitive endoderm trajectory. This work was actually part of a large-scale project where we collected 120 approximately cells or transcriptomes of 120 cells going from pre-implantation all the way through midgestation, representing the entirety of the endoderm, both embryonic and extraembryonic. Sonya Novochin in the lab led this effort, assisted by Beth Kuo, and the domain of the pre-implantation embryo was really sort of spearheaded by Vidya Garg. The analysis was a long-standing collaboration with Dana Pe'er, and initially it was Manu Sethi from her lab that was involved in, in sort of doing the computation for us. So effectively what we've done is we can track endoderm development from preimplantation, its bifurcation into embryonic and extraembryonic tissues. And I want to give you some sort of highlights of what we found and put them into some context. So essentially, after the primitive endoderm and epiblast are specified, they actually differentiate. The primitive endoderm will differentiate further into parietal visceral endoderm branch. These are schematized here in a cartoon of an embryo. Rather surprisingly, when we took these data that covered 3.5 to 5.5 days of development— this is a wild-type embryo— what Manu came back to Sonia and said was, well, I see a branch over here. This actually coincides with an area of high differentiation potential. I think there's a differentiation at this point between epiblast or primitive endoderm, or the other way around. By using genetic strategies, Sonja could show where she could label and then sort of identify descendants of epiblast or visceral endoderm. Using both genetics and chimera, she showed that epiblast cells will occasionally differentiate into an endoderm identity, but never the other way around. And we could actually sort of schematize this like this. Hold that thought because I'm going to get back to that later when probing cell plasticity. But effectively, one of the big sort of surprises of this work was the transcriptional convergence of extraembryonic and embryonic endoderm within the midgestational gut tube. So this actually argued that a structure within the embryo present at midgestation might comprise some component of extraembryonic cells. This actually disagrees with the dogma of the epiblast being the sole source of the embryo. So effectively, what our data showed, just to summarize here, a lot of work, and this is a 3-dimensional reconstruction of about 90 serial sections taken through an embryo, and effectively what it shows is every dot is a cell. I'm going to play it again if I can. And effectively every dot is a cell. Green cells are extraembryonic, i.e., grey cells are embryonic descendants, and essentially what the data showed us—we complement this with FATE mapping and high-resolution imaging—is that the midgestational gut tube that has yet to form sort of overt organs comprises cells of two distinct origins. It also showed us that this tissue is patterned into distinct organ identities that represent the organs of the adult. And we think this is very interesting. It's something we're following up on. How does an amorphous sort of tubular structure acquire pattern at midgestation? How does this pattern then become elaborated into the organs that we recognize in later development? And how is this done both through interactions within the endoderm and crosstalk between the endoderm and neighboring mesoderm? But what I want to get back to, rather than talk about those ongoing studies, is this mosaicism within the gut tube that we could see here through our single-cell transcriptomics. And shown here across pseudospace, predominantly sort of posteriorly, we see visceral endoderm descendants. Surprisingly, when we looked at the transcriptomes of these cells, that irrespective of their developmental origin, they seem to express the same genes, suggesting that they acquire some level of transcriptional equivalence. However, there is a small subset subset of genes, about 3 or 4 of them, that are specific only to visceral endoderm. ROCKS5 is one of them. This is the computational prediction for the expression of ROCKS5. This is a validation in a microdissected gut tube where we're doing a ROCKS5 mRNA in situ that demonstrates that it is expressed or localized within a subset of cells. And this is a targeted reporter that Sonia has generated where we can look in 3D with a nuclear-localized GFP localised or targeted to the ROCK5 blockers, and we can see a distribution of visceral endoderm descendants across the entire gut tube, with the yolk sac endoderm being our positive control. The fact that we can make this type of reporter allows us to address the long-standing, and I think very profound, question of do these extraembryonic descendants persist at later stages? Essentially, do they contribute to our adult or paediatric organs?. And I'm going to show you some data from a work in progress. It's still a work in progress, but I think the results are very interesting and quite profound. So the idea being, does this extraembryonic endoderm contribute to the embryo, and do any cells that we found in the embryo persist to later stages? What we want to do is to use a genetic approach. We've used ROCKS5, and we're targeting a Cre-RNA inducible Cre to the ROCKS5 locus. We actually have 3 different genes that we're using to do this experiment because we need to cross-validate with different genetic models. We pulse early before gastrulation. So Amos Tanai told us about gastrulation yesterday and the formation of definitive endoderm. So we want to pulse before definitive endoderm is specified. We know ROCKS5 is not expressed in the definitive endoderm, and in fact is downregulated even in visceral endoderm cells beyond midgestation. But essentially we're pulsing early, we're chasing later. Recognizing that we're actually probably only going to be sort of a mosaic in the cells that we label, this is not an experiment of 100% efficiency., and we may be looking at a needle in a haystack because essentially cells within the endoderm represent a very minor proportion of cells within the embryo. We've actually sought, rather than to perform serial sections, to perform clearing. I know there's a lot of sort of discussion of different clearing methods. Ali Ertugrul and Hiroki Ueda, and I think also Alan Shedetal, have been really instrumental in pioneering the use of clearing methods. And in our instance, with our samples, iDISCO+ seems to be our preferred method. And what it will allow us to do is to glean insights of endodermal tissues at unprecedented resolution internally in embryos that are actually larger than any embryo I've worked with previously. And as we venture forward into looking at later stages of development, the clearing methods that are actually actively being discussed, and in our case, iDISCO is going to be sort of essential for us to be exploiting. So this has allowed Sonja and Beth to use some of their inducible Crees, and to sort of label the endoderm specifically to actually be able to determine whether descendants of extraembryonic endoderm will sort of persist and where they localize. So this is a movie of one of their embryos. It's the only data set I'll show because we're still collecting data, but I think it's actually very tantalizing. What you'll see in the first spin is the endoderm. By 11 days, the gut tube is elaborated into the lung buds, the liver, the dorsal ventral pancreas, and then the intestine.. And I think this is really exquisite as imaging goes. It's a light sheet dataset that's been reconstructed. We can visualize the endoderm in its entirety, and then we can layer onto this descendants of extraembryonic endoderm. What you'll see is that they persist. They're specific within the endoderm and gut tube derivatives. They predominate within the intestine, but we can also see cells within the right pharyngeal endoderm and also in the ventral pancreas. So these data are very provocative. We're now collecting data from additional and later stages, going all the way through to prenatal or postnatal stages, to determine where these cells end up, whether they're transcriptionally, as suggested by our sort of transcriptional analyses of midgestation, whether they're transcriptionally equivalent, whether perhaps if we probe their epigenome through a single-cell ATAC-seq, whether they retain a memory of their past history, whether they're functionally identical, with their neighbors, and ultimately whether they behave distinctly under sort of homeostasis versus disease progression. And then prompted by our experiments in the mouse, if it turns out that we can see these cells at later postnatal stages, we will of course be motivated to look in the human. So what I now want to go back to is to really sort of address this question of, you know, these two sort of lineages, the primitive endoderm and the epiblast. They actually arise contemporaneously in the early embryo, and we wanted to ask the question of what is their plasticity, how set are they in their fates. So what I want to do as a developmental biologist who predominantly works in the embryo, I want to actually tell you about an experiment where we are actually— where we are leaving the constraints of the embryo. And what I mean by that is the embryo has a number of constraints. Firstly, it has a temporal constraint, Cells and embryos need to make decisions on time, and they can't procrastinate about a decision, perhaps a decision they don't like to make. Secondly, cells in early embryos are limited by number. We're now talking about tens, hundreds, maybe thousands of cells. If an event is rare, the probability of seeing it is low. So what we choose to do is to look in stem cell models, and luckily there are stem cells that represent each of the lineages of the mammalian blastocyst. That can be derived from blastocyst stage embryos. All of you are familiar with the ES cells or ESCs. Xen cells or extraembryonic endoderm cells represent the primitive endoderm, and TS cells or trophoblast stem cells represent the trophectoderm. These stem cell types have also gleaned rather a lot of interest recently because they can be aggregated together to generate embryo-like structures. So, for example, blastoid-like structures for the human and previously the mouse have now been reported, and they approximate to the sort of blastocyst sort of structure that we see in the embryo. So our interests are of course the primitive endoderm and epiblast lineage, and we wanted to ask questions regarding their plasticity in a stem cell model. So this was the project taken on by Vivek Garg in the lab, and what he wanted to do is to ask what happens if we try and direct ES cells to adopt a Xen or primitive endoderm fate, and also how What happens if we ask them cells, or cells of a primitive endoderm identity, to adopt an epiblast fate? While many before us have looked at cell reprogramming or cell differentiation and trajectories taken, very few have looked at states bookended by two distinct states in both directions. So essentially what we're doing is taking ES to Xen and Xen to ES so that we can compare the trajectories taken. So previous experiments by many investigators, including ourselves, has demonstrated that overexpression of some of the key endoderm transcriptional regulators, including GATA factors, GATA4 or GATA6, is sufficient to take ES cells to a stem-like state. Interestingly, even though many studies have reported reprogramming using OKS or OKSM, so pluripotency-associated transcriptional factor sort of cocktails. This has never been reported for the Xen or endoderm-like state, so we thought we'd actually take a look. Biber has adopted parallel systems to have the experiment as controlled as possible, where he's using Dox inducible transgenes in ES and Xen cells to overexpress reg or master regulator of the sort of an endoderm identity versus OKS. And each of these cassettes, I should say, is tagged with a reporter where we can visualize the transgene We also have knocked in reporters identifying the end states that we can visualize them as the states progress. I should say that this project is a collaboration with Effie Apostolou, who's our neighbor at Weill Cornell across the street. Effie is an expert in cellular reprogramming and the transcriptional regulation of cell identity. And in fact, in her former life as a postdoc with Konrad Hockenlinger, she actually developed this transgene. So just to talk to begin with regards for conversion for primitive endoderm to epiblast, this is the scheme of the experiment. This is the genotype of the cells. Essentially, you'll see over a period of time we can track the transgene, we can track an endogenous reporter of pluripotency, Oct-4, and you can see we can generate pluripotent-like cells. But what is really interesting and something worthwhile noting is this takes 3 to 4 weeks and is less than 1% efficient. However, just to set the record straight for anyone who's curious, what we notice is that our ZeniPS cells actually have lost all Zen markers and have adopted gene expression profile of ES cells. And in the most stringent test, which is the embryo chimera, our ZeniPS cells will exclusively contribute to the epiblast of the embryo proper in chimeras, and in fact generate completely embryo-derived or ES cell or iPS cell-derived embryos, demonstrating a complete change in their developmental potential. What we've actually gone on to do, and I'm summarizing here because it's a lot of experiments and I want to get to some of the single-cell data, is that we've identified a growth trajectory with usual suspect markers, that this trajectory goes through a co-expression state, and we can identify bottlenecks. And it seems as though the downregulation of the endoderm network seems to be a prerequisite and bottleneck for the acquisition of pluripotency. We wanted to compare this with forcing the expression of GATA4 in ES cells using a comparable system, and again to look at intermediates. This is the expression or the experiment shown here. Here we have a GATA reporter for the endoderm lineage. We have our GATA4 again labeled with Cherry. We can visualize this. I'll play it again. This is a time series, and what you'll notice is over 4 days you can see expression of the GATA reporter and expression of the other endogenous GATA. Within 6 9 hours of induction with doxycycline, we can see activation of the endoderm network. If we sort the cells acquiring an endoderm identity, we see a persistence of pluripotency-associated factors such as SSEA1, and we can draw a trajectory a bit like this. Again, a co-expression state en route to an endoderm identity. But what is really interesting here is that this process ES takes 3 to 4 days and is over 99% efficient, which we thought was very surprising. So this is essentially summarizing our findings thus far. Effectively, if we go from Zen to ES, we're very slow and very inefficient. However, taking the trajectory in the other direction is highly efficient and fast. So what we decided to do is to take this experiment and start to analyze single-cell transcriptomes. We also have now analyzed some, or beginning to analyze, or currently analyzing some single-cell ATAC data. So we took various time points across each of these two trajectories and sort of loaded them onto a 10x Genomics Chromium chip and performed single-cell RNA-seq analyses. We could then compare these trajectories. We could also compare them to the data from the embryo to understand whether trajectories taken in vitro mirror those that we observe in vivo, which I think is really our gold standard. So what we've done is collected 10,000 cells from each of these time points. We've used Palantir, a tool developed in the PEAR lab, sort of by Manu Sethi, a former postdoc who's now got his own group at the Fred Hutch, to essentially pseudo-time order and place on a force-directed layout each of these cohorts of cells so we could actually aggregate them. We're analyzing 50,000 cells total for each of these two experiments.. And what Manu can do when processing through the Palantir pipeline is look for differentiation potential and different terminal states. Interestingly, if we look for differentiation potential in the ES2 Xen, these cells have a sort of a lack of potential because there's a single terminal state, and this trajectory is very efficient and very fast. By contrast, if we looked at Xen to iPS, we noticed two terminal states which essentially act as sinks through which cells cannot get out of en route to the final stage of iPS. And these are essentially regions of low differentiation potential where cells essentially get stuck. Essentially, we can schematize this like this, that cells, if they get to T1 or T2, they have trouble getting back out.. And then we can ask questions of probability of getting to T1 from the start state, and this is high, getting from T2 to T1, and ultimately out of T2 to T3, which is essentially, we think, a final deterministic stage of this reprogramming. What I think is very interesting, I mentioned that OPT4 seems to come on around the time of this co-expression stage, and this is effectively shown here from our single-cell data. OPT4 is activated activated as cells exit T1 and start to enter the T2 state. However, other pluripotency-associated markers are only activated late as cells exit to T2 to enter T3, and only after primitive endoderm-associated factors are downregulated. So this is giving us some detail of the dynamics of sort of transcription factor expression, and we have details of the genes associated with each of these two so-called terminal states. What is interesting is we can merge the trajectories from either zen to ES or ES to zen, and what this reveals is that the states traversed seem similar. This then leads us to conclude that the differential conversion rates and efficiencies between these two directions reflects the relative plasticity of cells, and in fact the direction the cells are going in. As embryologists, we're, you know, we're taking this question out of the context of the embryo, but we want to put it back and ask, are any of these trajectories reminiscent of anything that we see in the context of the embryo? And the answer is yes. So we can take these trajectories, we can bin them along pseudotime, and we can see that if we go from primitive endoderm to epiblast in the xen to iPS paradigm, these cells will actually creep up the primitive endoderm arm, and then they'll toggle over to the epiblast arm and go down the epiblast arm. What we found was very interesting is that cells will never enter an ICM-like bipotential state, and this is, we thought, interesting because this state has been very elusive to capture in culture. And this would argue perhaps that there is some form of bistability in the system, that the cells will recapitulate in vitro the trajectory that we see in vivo, and then they will jump from one arm to the other. This is the same for when we go to the ES to Xen, they go back up and then they toggle back to the primitive endoderm side of things, which we found to be again very interesting because the cells in the dish seem to be recapitulating trajectories that are reminiscent of the embryo, and they are mapping back to states that we recognize in the embryo. So I just, in my final slide over here, I wanted just to reconcile some of our data. In the force-directed layout of these cells isolated from wild-type blastocyst-stage embryos, it was always interesting to us that the epiblast is closer to the inner cell mass than the primitive endoderm, perhaps arguing that primitive endoderm, although it forms contemporaneously with epiblast, perhaps represents a more differentiated cell type, but it's perhaps more difficult to reprogram. We also have found to have this very curious finding when we would add data from 5.5 embryos that argued that cells will traverse one to other sort of lineage, and we could only see them in the experiments that we performed in embryos by fate mapping or through chimeras as transverse, going from epiblast to primitive endoderm. We could never see primitive endoderm to go to epiblast. And of course, as Amos Tanai sort of discussed in his talk yesterday, 24 hours later, during gastrulation, epiblast will differentiate into an endoderm identity. However, sort of primitive or visceral endoderm has a limited developmental potential until, of course, it gets into the gut tube, which we think is very interesting. So perhaps these sort of in vivo observations can somehow reconcile the observations we've made using this bidirectional reprogramming paradigm whereby ES cells will more readily differentiate into a primitive endoderm identity than these primitive endoderm cells into an epiblast-like state. We now have data on the transcriptomes, on the key sort of regulators associated with bottlenecks, and we're currently working with Jana Pair and her colleagues to mine single-cell ATAC data to try and understand why sort of the one direction is ultimately more sort of efficient than the other. Some of the data that we have, and I'm not going to go into too much detail, is that we know that the endoderm transcription factor network inhibits the establishment of pluripotency. When we think about endoderm identity, we think of 3 key transcription factor networks: the GATA, the SOXs, and the FOXs. And we and the lab of Claire Chazot and our own have previously shown an essential requirement for GATA6 GATA6 in the acquisition of a primitive endoderm identity. Interestingly, when we knock down GATA6 along with GATA4, which is also expressed in the primitive endoderm but later during reprogramming to iPS cells, we can knock these genes out individually or combinatorially. Our data suggests that it is GATA4 but not GATA6 that is essential for stalling the establishment of pluripotency. And thereby somehow extinguishment of the GATA sort of network seems to be, or GATA expression seems to be key in the reprogramming. So we're very keen to look at the GATA4 locus in terms of why sort of it's not being shut off. In closing, I just want to mention, sort of, or to schematize everything in this cartoon that was drawn for me by Vidya, whereby we can perhaps schematize everything on a Waddington landscape. We can place our epi cells higher up than primitive endoderm, as reflected by here. And in thinking about the reprogramming of primitive endoderm to epiblast, whenever I look at this, one thing comes to mind, and it's essentially this title from this song by David Bowie, which is "Up the Hill Backwards." And I would argue that primitive endoderm cells, when being reprogrammed to epiblast, are actually going up the hill backwards. This comes from an LP called Scary Monsters. It was actually the first LP that I bought as a child. It was released well before Viva was born, and I would argue that rather than scary monsters, we're looking at hopeful monsters. I think I'm going to end there. I've mentioned everyone's work who's been involved in the project: Viva, Beth, and Sonia. We're really fortunate to have outstanding institutional core facilities that support our work and really world-class collaborators. Effie is really a leader in the context of cellular reprogramming and cell fates identities. Dana has been our go-to person both within our lab and also the single-cell research initiative that she established at Sloan Kettering in terms of querying single-cell transcriptomic and in fact single-cell omic data, as well as some forays that I didn't talk about where we're trying to merge 3-dimensional imaging with some single cell omics. I think I'm going to stop there.

Thank you for listening. Kat, thank you so much. That was great. And we're happy to take any questions. It's being at Memorial Sloan Kettering, it's not lost on you, I'm sure, that the relevance of this for driver gene networks in— in oncogenesis. What's your sense of using this platform and this type of knowledge to infer driver gene networks, and what approach would you take to validation of, you know, particularly combinatorial actors and effectors of maintenance of stem cell state, failure to differentiate, and stop dividing?

What are you thinking on that along those lines? I think the person to be asking this or fielding this question to is actually Donna Pear. But in terms of thinking about the relevance of our findings in the embryo to drive the gene networks in oncogenesis, I think just to highlight one study. So we're actually developing quite an extensive effort to look at the emergence of organs out of the gut tube. We're focusing, for example, on the emergence of the lung territory. We're very interested in the lung because through studies, or, you know, through both our developmental interest but also through recent studies between Dana's lab and that of Joan Massagueta's institution, where they've looked at disease progression from adenocarcinoma all the way through to metastatic lung cancer. And effectively what they're looking at, and they're interrogating through various analyses, but also single-cell transcriptomics are networks that are driving sort of oncogenesis all the way through metastasis. And what we're seeing is a recapitulation of sorts of developmental programs. So I think, you know, we're looking at the normal course of events, we're perturbing them in mouse mutants, we're looking in embryos, and I think I see Dana very much as the sort of go-between between ourselves rooted in the embryo and studies from our colleagues who are interested in metastatic potential.

I think Dana will have a chance to talk about that at the end of the day today. One more time for one more question from Anne Grapman-Button. Kat, In the reprogramming, your cells jump to a point on the PRE path. How precise is the point they hop to?

And what is the in vivo equivalent of this, this point in phase? Okay, so I simplified things a little bit. So it's, so that's a very interesting question. Anne is of course the person who would ask it. They actually hop to the point where right after primitive endoderm specification, the onset of GATA4 expression— and may know that as initially cells of the inner cell mass will express one GATA factor but also factors associated with pluripotency, this being GATA6. As primitive endoderm is specified, they will then downregulate the pluripotency network and maintain GATA6. Once they do so, they will then activate GATA4. The cells seem to be hopping to the point where they activate GATA4 GATA4 and then traversing. This is why we think that the mapping back to the embryo, but also our downregulation experiments implicating GATA4 but not GATA6, seem to be somehow sort of hitting on the same point. So, it's really when a primitive endoderm identity is uniquely established and downregulation of pluripotency-associated markers has been extinguished, which is the, the top point of that trajectory. Great, Kat, thank you so much.