Scientist Stories: Jeannie Lee, RNA and the New Genetics

Good morning, ladies and gentlemen, prior Gairdner Laureates and our current crop of Gairdner Laureates. We're gonna have 3 lectures from leaders in the field. I just want to say one or two things. I'm Philip Marsden. I work with Dr. John Dirks and the Gairdner Foundation, and my co-chair, Dr. Paul Cubbs, and I play a role in the selection committee for the Gardner Prize. Each year we get about 100 to 150 active nominations and about 30 to 50 new nominations per year. The goal of the Gardner Medical Review Panel and the Medical Advisory Board is to sift through those awardees. Today's sessions have started off with some brilliant talks, and the next 3 sessions, each of 30 minutes in duration, Move us on in the field with people who were there right from the start, and our first speaker is Dr. Jeanne Lee. I had followed Dr. Lee's work for many years. In fact, every time my lab and I looked at one of her papers, we couldn't believe the depth and breadth that she delved into one single long non-coding RNA pathway. Dr. Lee is a professor at Harvard Medical School. She's in the Department of Genetics and Pathology. She's a Howard Hughes Investigator, and she's also a member of the prestigious National Academy of Sciences. Dr. Lee.

It's a great pleasure to be here today, and I want to thank the Gardner Foundation for this kind invitation and also for the opportunity to tell you about our work on using X inactivation as a model, really, to study epigenetic regulation by long non-coding RNAs. Okay, so X inactivation, as many of you know, is the mechanism of dosage compensation for mammals— not flies, but mammals— that evolved to deal with the fact that females have two X chromosomes and males have one. And so to equalize gene expression across the sexes, what we do is we inactivate one X chromosome in the female mammal. And the reason this is such a great system to study long non-coding RNAs is because this process is controlled by the X inactivation center, which is an X-linked locus that's both necessary and sufficient to drive the X inactivation process. And so, in fact, if you were to take this X inactivation center and put it on an autosome, the autosome becomes subject to the same kind of silencing that the X chromosome is. So all of the regulatory elements are really present here at the inactivation center. So there's this Xist gene, which initiates the silencing, coats the chromosome, and silences it. Its antisense partner, psiX, which is a repressor of exist. So exist designates the inactive X, whereas the antisense designates the active X chromosome. So two non-coding RNAs regulating that process. And a short repeat transcript, RepA, which targets polycomb complexes to the X chromosome. And then this upstream activator, JPX, which turns on exist expression. Okay, so here's the famous exist gene, 17 kilobases. It's expressed only from the inactive X chromosome and is absolutely required to initiate this process. And it's been said to coat the X chromosome. So here's the RNA fluorescence and C2 hybridization. You see that the RNA is plastered across the X chromosome. Now, our more recent studies using super-resolution microscopy tells us that exist is really not a coat per se, but is really more of a cluster, a bunch of dots of exist particles across the X chromosome. So here's one inactive X, and here's another inactive X, and you can see that these exist particles, they form about 100 dots, with each dot being approximately 1 to 2 copies of the RNA. So the exist stoichiometry is actually very low on the X chromosome, there only being about 100 copies outside 200 copies of the RNA on the inactive X, and that implies something about the way the RNA silences the chromosome. We now know that not only is Xist required to initiate the X inactivation process, but it's also required to maintain organismal health, because if we were to delete Xist from the blood compartment in mice, what we see is a fulminant blood cancer in female mice, essentially with 100% penetrance. So Xist is very important not only early in embryogenesis when X inactivation takes place, but also throughout adult life. And so what we and others have been very interested in for the past really two decades is how does this long non-coding RNA, Xist, and its various non-coding partners initiate silencing of a whole chromosome. We now know that Xist serves as a recruiting platform for a bunch of epigenetic complexes, including Polycomb repressive complex 2, which is a very important complex that trimethylates histone H3 at lysine 27, generally represses gene expression, and has many, many important developmental roles, particularly also in the pathogenesis of diseases like cancer. Now, on the polycomb side, there's been this major question in the field about how you target this complex to thousands of positions in the genome when none of the core subunits of PRC2 is a transcription factor. It doesn't have a sequence-specific DNA binding subunit. So I think the answer is going to be that there are going to be many different mechanisms by which polycomb can be targeted, and that may even be species dependent. But in mammals, we know that one of the important recruiting elements is going to be RNA., as we showed many, many years ago now, that exist, and this RepA RNA can both directly bind to PRC2, and then target that complex in a locus-specific fashion to the inactive X chromosome. Okay, so this is a very fast-moving field, and a lot of things have happened over the last few years. I don't have time to tell you or show you all of the data, so what I'm going to do in this slide is summarize summarize what we and others have done over the last several years. All right, so in the beginning, CTCF and another transcription factor called Rex1, which is a pluripotency factor, will sit on the exist promoter and repress exist transcription. And then what happens during cell differentiation is that this short RNA, RepA, is transcribed, and it co-transcriptionally recruits PRC2 to the X inactivation center. And what we showed just this past year is that the ability of PRC2 to recognize this repeat A motif depends on another epigenetic factor called ATRX. And ATRX, we believe, acts as an RNA helicase to change the shape of repeat A RNA to a form that can be recognized by PRC2, and therefore determining the specificity of PRC2 for this RNA. Okay, so another thing that we've learned is that the process of recruiting PRC2 is not the same thing as loading of that complex, because what you see, at least for X inactivation, is that as long as this antisense transcript, psi-X, that I was talking about before, that designates the active X, okay, so as long as that RNA is expressed, this polycomb complex doesn't touch down onto the chromatin. So recruiting is not the same thing as loading of the epigenetic complex. And only when the antisense RNA goes away do we see the complex load onto the 5' end of Xist. And equally interesting, loading of the complex, at least in the biochemical sense, does not equate with its catalytic activity. So what we showed is that, first of all, this recruitment requires Jarrah2, and what we showed is that when the RNA engages PRC2, it prevents the methyltransferase activity of EZH2, which is the catalytic subunit. And when the RNA engages Jarrah2, the RNA's affinity for EZH2 decreases. And so the RNA probably partially disengages, unleashing the methyltransferase activity of EZH2 onto the chromosome. Okay. So now, while all of this is happening for PRC2 and various transcripts, another thing is going on at the same time, and that is this JPX activator is transcriptionally expressed-induced tenfold.. And when it crosses a certain threshold, particularly in female cells, Jpx will bind to CTCF and evict CTCF from one X chromosome. So that happens in female cells because the concentration of PFC2— concentration of Jpx is twice as great in female cells. So it's a threshold response. Okay. So this combination of losing the repressor, which is psiX,. And gaining the activator, which is JPX, allows Xist to be expressed for the first time. And Xist, because it also has a binding site for PRC2, will bind to the PRC2 with the aid of ATRX again. And this RNA-protein complex then touches down on another site that we call the nucleation site, mediated by transcription factor YY1.. And once it touches down on this nucleation site, the RNA-protein complex has the license to then spread across the whole X chromosome. Okay. So, that is how we're thinking about the initiation of X inactivation. And now, another very important question in the field is, aside from recruiting PRC2 and interacting with ATRX and loading onto the YY1, nucleation site, what else is existRNA actually doing to the X chromosome? And you can imagine this is very large non-coding RNA with a lot of real estate to bind other complexes, and we imagine that it has to interact with other complexes in order to silence a chromosome which is 5% of our genome. And this is just a portent of things to come, that the nature of this interaction could be very different for each and every one of these factors. Okay, so as some of you know, we've embarked on an Xist proteomic analysis to define a comprehensive Xist interactome. So asking the question, what are the other proteins that interact with Xist RNA, not only to understand the mechanism of silencing, but also in looking ahead to ask the question about whether we can unlock the potential of the inactive X, which has, you know, 1,000 very good genes. To treat X-linked diseases such as Rett syndrome or Fragile X syndrome or Duchenne muscular dystrophy, just to mention a few. Okay, so what Anand Mehta Ji has done in the lab is to develop a new protocol that we call iDRIP, in which he UV-crosslinks female cells that have an inactive X chromosome, and he uses UV-crosslinking to obtain directly interacting proteins. UV cross-links, and then, of course, isolates nuclei, DNases to solubilize the chromatin, and then trap these exist complexes using oligonucleotides that are complementary to exist RNA, subjects these complexes to denaturing conditions so that we identify only those interactions that are directly cross-linked covalently to exist RNA. And then in collaboration with our critical partner here, Willie Haas at the MGH, we performed a quantitative proteomic analysis to pull down approximately, depending on your enrichment cutoff, somewhere between 100 to 200 interacting candidates. We were very pleased to see known interactors, including PRC2, HNRMP2 and ATRX, and also this nuclear matrix factor, HNRMPU. We were also delighted to see macroH2A, which is a variant histone, and SMAD1, which is a condensin protein, both of which have been described to be enriched on the inactive X but not known to be directly recruited there by existRNA. A lot of general RNA-binding proteins, including HNRMP2. PK and SPEN, which Howard and Mitch Guttmann have identified. And then— and we got a little bit more excited when we saw the other polycomb complex, PRC1, the maintenance methyltransferase DNMT1, these topoisomerases, TOPE1 and TOPE2, which are important for so many epigenetic processes, including DNA replication, transcription, and so forth, chromatin remodelers of the SWI/SNF family. And then these factors in red, many of which are chromo— architectural factors such as the cohesins, which I'll tell you more about in a few minutes. But we believe that these are real interactions because we can see the reciprocity of that interaction. So if we bait with existRNA, we pull down these proteins, and then reciprocally, if we bait with these proteins, we can pull down existRNA. Interaction is reciprocal. All right, so now if these are real factors, then we might expect to see some perturbation to gene expression on the inactive X. And so we asked this question first by looking at a transgenic GFP marker which has been placed on the inactive X. Okay? So interestingly, what you see here is that if we were to hit only one of the components of the interactome, such as DNMT1 or TOPE1 or TOPE2, we saw very little, if any, reactivation, 5x reactivation, which is essentially nothing. But if we were to combine two interactors, we saw anywhere between 50 to 100-fold reactivation. So that's getting a little bit better. If we combine three interactors, we saw anywhere from 100 to 200-fold reactivation, which is becoming extremely respectable, because as you can see here, the active X chromosome is expressing GFP only at 150-fold of the inactive X chromosome. Okay, so we can get high-level reactivation that's essentially matching that on the active X chromosome by targeting multiple interactors, which supports this idea that the inactive X chromosome is repressed by multiple parallel and synergistic pathways. Okay, but of course what we really want to know is what happens to native genes on the inactive X. And some of you will say the GFP doesn't really count because it's only been there for a few years, not millions of years over which many of these X-linked genes would have had to evolve with X inactivation. So we performed a different kind of analysis, allele-specific RNA-seq analysis, using a hybrid cell line that always inactivates the Mus musculus, one kind of mouse X chromosome, whereas the Mus castaneus chromosome is always kept as the active X chromosome. There are lots of polymorphisms between the two that allows us to distinguish between the XA and the XI. And the net effect of this is that there is increased expression, relatively speaking, speaking, from the inactive X chromosome, anywhere from 10 to 25% greater expression from the Xi. Now, at the single gene level, of course, we can see anywhere from zero reactivation to zero X reactivation to twofold reactivation to tenfold, fiftyfold, to hundredfold. So every gene does its own thing. Now, what I want to stress here is that we're not getting Xi-specific reactivation, which is, of course, impossible because all of these factors are general epigenetic factors that regulate many different processes in the cell. What we see instead is an XI preferential reactivation, and that's because we're targeting synergistic components of the existent interactome. Okay? So that's an important point. Here's an example of a reactivated gene, G6PD, which is actually a disease gene, and you see increased number of reads over the exons that contain a SNP that allows us to distinguish XA from XI relative to our controls. Okay, so then we got interested in mechanism, and we were particularly interested in CTCF and the cohesins because these are, as many of you know, general architectural factors that bring together distant genetic elements. And so it's thought that CTCF may serve as a glue that brings together, say, enhancers or promoters, or maybe even something from two different genes altogether. And the cohesins appear to form a ring around that structure to lock in that conformation. Okay, so to look at the mechanism of action, John Froberg in the lab asked what it is that cohesins and CTCF are actually doing. And when we performed RNA immunofish, we got our first surprising finding. Which is that these proteins, unlike PRC2, which forms a very strong focus over the inactive X— unlike PRC2, CTCF and cohesins really aren't enriched on the inactive X chromosome. So here's exist in red, and you can see that cohesins are really, if anything, depleted over the inactive X territory. So that already tells us that the nature of interaction between exist and cohesins, or CTCF, is going to be very different from the nature of interaction between PRC2 and exist. Okay. To get to the bottom of this, we performed allele-specific ChIP-seq, and just to summarize a lot of data here, what we saw is that cohesins are indeed depleted from the XI but not absent altogether. Okay, so on the XA, there are about 500 to 600 specific sites. On the XI, about 10 to 20 specific sites. And then they share between them, XA and XI, about 200 cohesin binding sites. Okay, then to ask what happens when Well, I mean, to really test this idea that existin cohesins may be doing something important on the XI, we asked what happens when we ablate existin RNA. And the answer is that these cohesins come back to the inactive X chromosome. So you see here in wild-type cells, cohesin is binding to the inact— I'm sorry. Cohesin's binding to the active X, which is shown here in blue. But not the inactive X. But when a blade exists, a cohesin binding comes back to the inactive X chromosome. Cohesins come back to the inactive X chromosome. So what this tells you is that the cohesins seem to bind the X chromosome as the default. It's a default binding state. And the role of exist is really to keep those cohesins from binding to the XA-specific sites. Okay? And when we ablate exist, those cohesins come back en bloc onto the inactive X chromosome. Now, because these cohesins are architectural factors, we followed up and asked the question, what happens to the chromosome conformation when the cohesins come back as a result of ablating exist? Now, for this, I have to tell you that chromosomes are organized into these so-called topologically associated domains. The work from Job-Decker and Bing Ren and others show that they're organized into these megabase-size domains. And the inactive X, by contrast, as has been suggested by a number of different groups, don't have those same long-range interactions. John has performed his own version of Hi-C, and what he sees on chromosome 13 is exactly what other people see. See these topologically associated domains, these pyramids, with increased interactions within the pyramids and very little interaction outside of those domains. And here's the X chromosome, and you can see the same kind of so-called TADs on the active X, but a little fuzzier. And they're fuzzy because we're superimposing both the XA and the XI on top of each other. And so, he parsed out the XI by doing the analysis in an allele-specific fashion, and then called 112 topologically associated domains on the active X. So that active X looks just like autosomes. And on the inactive X, very few, if any, of these TAD-like structures. But the most important question here that we really want to address is what happens when we delete exist and the cohesins come back? And the answer is that the inactive X begins to look like an active X chromosome. So here's the inactive X in the wild-type state. You see very few interactions along the X chromosome. Here's over the same region on the active X chromosome, you see these upside-down pyramids, which are indicative of the these TADs. When we delete XIS and the cohesins come back, you see that these TADs reappear on the inactive X chromosome. And they're not reappearing in a random fashion. In fact, they look almost like exact mirror images of the active X chromosome. Okay. And then this is a whole chromosome map that shows you that where the TADs are— where the TADs are coming back, shown here in blue, is where the cohesins are being restored on the inactive X. Okay, so there's a large exist-interactome. Cohesins are important. It's an interacting protein. We can perturb the inactive X by targeting multiple components of the inactive X. And unexpectedly, exist is not attracting cohesins, but rather repelling these factors. And we called about 100 TADs on the active X and very few, if any, on the inactive X. And deleting exist will restore those cohesin sites, which in turn will restore the topologically associated domains, which we believe is the reason behind chromosome reactivation. Okay, so then in the second part of this talk, I want to give you a little teaser. Think about a problem we've been working on for 15 years, and a problem that we're still very much in the thick of things for. Now, up to this point, we've talked almost exclusively about how Xist is silencing the X chromosome. And we've said very little about what is actually controlling Xist itself. Now, you may be aware that there are actually two forms of X inactivation. There's one that we've been talking about, which is random X inactivation, in which either X chromosome, your mother's or father's X chromosome, can be inactivated. But there's another form which is equally interesting, and that's imprinted X inactivation, which has been serving as a really wonderful model to study another epigenetic process called transgenerational inheritance, where gene expression can be influenced by the parent of origin of a specific gene. Okay, and the reason this is such a great model to study transgenerational inheritance is that in imprinted X inactivation, the father's X chromosome is always inactivated. So we can see this in the placenta, of various eutherian mammals like the cow, the mouse, the rabbit, and so forth. But it is also seen in the whole body of the marsupial mammal like the kangaroo and the opossum. And we know that imprinted X inactivation also requires these non-coding RNAs exist, which again designates the inactive X, the paternal inactive X, and psiX, which designates the maternal or the active X. But then very little, very little else is known about how imprinted X inactivation takes place. Now, there are two schools of thought. In the so-called de novo hypothesis, it is postulated that there is a mark placed during the spermatogenesis which isn't acted upon. So silencing doesn't actually take place until the next generation around the 4 to 8 stage, so that's mid-preimplantation development. Now, we and others have proposed the opposing hypothesis, the so-called pre-inactivation hypothesis, in which we suggest that the paternal germline is actually what's responsible for establishing silencing of its paternal X chromosome, and that in the germline gets silencing, and then that chromosome is passed on to the next generation as a pre-inactivated inactivated chromosome, at least partially pre-inactivated. So the idea here is that this process would depend on a process that you've probably never heard of before called meiotic sex chromosome inactivation. So the idea is the following: during the first meiotic prophase in the male germline, okay, so in the father's germline, homologous chromosomes pair. So you know that. Now, the X and Y chromosomes are only partially homologous, so they'll only partially pair, and non-homologous portions of those chromosomes undergo inactivation by this mechanism called meiotic silencing by— which you can't see, but it's called meiotic silencing by unpaired DNA. And it was an epigenetic phenomenon that was first described by Bob Metzenberg many, many years ago in Neurospora. But has been shown to be true in other species, including mammals, as well. Okay, so the idea here is what we showed, that contrary to what was known at the time, the X and Y chromosome get silenced, but not transiently. In fact, these sex chromosomes remain inactive through the 2 weeks of spermiogenesis.— so that's a long time. And then what we proposed is that that inactive chromosome gets passed on to daughters, daughters, as a partially pre-inactivated chromosome. Okay. So the idea here is that what happens during the first meiotic prophase in the father directly impacts the transcriptional state of the X chromosome in daughters in the subsequent generation. Now, as you can imagine, such a thing is very difficult to prove. So what we've been trying to do here is to develop a transgenic model for Xist imprinting to look at whether the unpaired state in the father's germline could somehow be responsible for silencing and then for imprinting of the Xist gene. And I'm going to show you the work of Sha Sun, who's a postdoc who's now moved on to her own position. At UC Irvine. Okay, so what Shaw showed is that this transgene here in pink, which includes most of the X inactivation center— here's Xist, psi-X, Jpx, as you'll note— this 200-kb transgene has all of the essential elements to imprint Xist. So when this transgene is paternally transmitted Xist turns on. When it is maternally transmitted, Xist is repressed. Okay. And then she established a number of founder lines and showed that there are profound phenotypes in the mice, as you might imagine, by silencing a whole chromosome. And I should say that these transgenes went in all over the genome, including in autosomes as well as the X chromosome. And as you can imagine, the female mice really do not tolerate this very well. Many of the mice are are embryonic lethal for the female. And as you might also imagine, the transgenes are also male lethal, but also male sterile, as shown here. Now this is something I don't really have time to get into right now, and we frankly don't understand very well ourselves, but I'll show you. Here's a wild-type testis, a transgenic testis,. And what you can see during pachytene in the wild-type male is a pairing of all the homologous chromosomes with the exception of the X and Y chromosome, which is here marked by gamma H2AX, signifying the beginning of transcriptional silencing. Now, in the transgenic animals, the pachytene spermatocytes are rather abnormal, so they have abnormal MSCI., and it appears abnormal ectopic accumulation of these silencing marks on autosomes. These males are sterile, so these spermatocytes do not mature to produce spermatozoa. However, we were lucky in the sense that one line was not male sterile and could in fact give rise to both viable male and female progeny, and so this gave us the opportunity to look at exist imprinting and whether the zygosity of that transgene affects imprinting. So these are somatic cells derived from this one mouse, and you can see— or one line— and you can see that males express exist in their body, and females express two foci of exist, one on the X and one on the autosome. They survive it somehow. Now, the germline is a different story altogether. So here's weird result number 1. In the male germline, what we see is that when the transgene is hemizygous, meaning unpaired, has no pairing partner, there's only one copy of the transgene in the male germline, the chromatin is highly condensed. Okay? But when we homozygose the transgene, this locus opens way up. And so we measured the area and so forth, and I'll tell you, there's a statistically significant difference. And that's the quantitation. Now, if you were a normal transcription person, you would expect the transgene to be expressed when there's open chromatin conformation, but not when it's condensed. But in fact, what we observed was the opposite. So exist expression was greater when the chromatin was condensed. Okay. And furthermore, now when we look at the next generation, at daughter cells— these are blastocyst cells— we see that when the daughters inherited the transgene from a hemizygous father— so unpaired in the germline— exist is very nicely expressed from the paternal chromosome. So exists as imprinted and expressed in the, in the daughters. But on the other hand, when the daughters or sons obtain that transgene from a homozygous father, so where the transgene would have been paired in the father's germline, we see no expression or very little expression of the transgene. Okay, so there's a little teaser. We don't understand the mechanism of this, but what we can say is that the unpaired state, the lack of pairing in the father's germline, somehow imparts a signal onto the exist gene for it to become imprinted and for it to be passed on to the next generation as an expressible construct. So I think that this provides perhaps some mechanism by which we can get a pre-inactivated paternal X chromosome. So the idea, again, is the following: that during the first meiotic prophase, when all homologous chromosomes are paired, the X and Y are not paired. And, of course, the exist gene is not paired, right, because there's no exist on the Y chromosome. And that absence of pairing is what places the mark, according to our hypothesis, and is what also silences the X chromosome, which then gets passed on to daughter cells as a pre-inactivated paternal chromosome. So two things are important for imprinting: the sequence itself of the 200-kilobase transgene, as well as the absence of pairing of that transgene in the paternal germline. Okay, so as I said, that was a little teaser. We have no mechanistic answer so far, but something to think about. I believe I've acknowledged almost everyone who's been involved in this work: Nanmi Najigi, John Froberg on the XIST proteomics, Sha Sun and Satoshi Namikawa earlier on the transgenic work, Kavitha Sarma, Catherine Cifuentes-Rojas did a lot of work on the specificity of RNA-protein interactions for PRC2.

Thank you. Thank you, Dr. Li.

We have time for some questions. Dr. Sherr? When the inactive X gets reactivated and you see TAD appear, you implied that TAD appeared before transcription in the reactivation of that chromosome. Could it be the other way around, that reactivation of transcription then drove the TAD organization, and that's why you recovered it?

Absolutely. What we see is TADs appearing at the same time that the genes are expressed, but we don't have the temporal separation to be able to say which actually came first, but we see that they're both there when measured.

Any other excellent questions? Okay. In this last unpairing chromosome—

Yes.

Transmitting inactivation. Is there any indication that the pi system is involved in that? In the pi RNA?

The pi RNA system is involved in that process? We are extremely interested in this question. And it's funny that you should ask about pi RNAs because what we— and I didn't have time to talk about this— is that in the next generation, so in the daughters of these fathers with either, you know, the hemi or homozygous, what we see is that the paternal X chromosome is already partially silenced, but particularly at repeats. So measured at LTRs, ERVs, signs, lines on the paternal X chromosome, we see a pre-inactivation, if you will. They are much more inactive than the coding genes. On the paternal X. So we think that there may be a connection, but it, you know, again, it's very difficult to prove because these Pee-wee mutants are sterile, just as many of our transgenic lines are sterile.

So very difficult to do the genetics here. You know, Dr. Lee, many of us have looked with admiration at how you delved into this one system in depth. As a broader question, maybe for the trainees, do you think that many of the long non-coding RNAs will have this level of complexity?

I think that any lncRNA out there that is ultimately will be shown to have function will have this level of complexity. They're going to interact with multiple proteins. I don't know that they're all going to be imprintable, but at least for the particular pathway that they're involved in, the proteins that are going to be recruited are going to complex, and the mechanism of silencing or activation, as the case may be, will be complex.