Scientist Stories: Anna Pyle, RNA structure and RNA recognition by protein enzymes

Well, I want to thank the Society for this invitation to come and talk to you guys about a topic I'm really excited about right now. I also want to thank Steve for that nice introduction and for his optimism that in fact the approaches you'll see today are going to help us understand what lincRNAs are doing. So— and what I'll try to do is I'll walk you through the kinds of rationale, the kinds of methods that we have developed to study these RNAs and how they're based on our early studies of catalytic RNAs. Catalytic RNAs are easier to study because you sort of have a crutch, which is that they do chemical reactions, so they can report on their structure through their activity like any other enzyme. And we've used them to develop a toolkit that we think will enable us to progress on much more tougher targets. So what I hope I'll have time for today is to tell you at least two stories about important lncRNAs that are functionally validated, and how we found that secondary and even tertiary structure in those molecules affects how they function. So first I'll tell you about how we came to these ideas., and then I'll take you through those stories. So before we begin, I'd like to introduce you to a couple of different ways to think about RNA molecules and how they interact with their environment. RNA has many ways to communicate with the outside world and with the cell. There's the primary sequence. There are sequences in here that bind proteins or do other things. That dictate their function, and this is how most people think about RNA. And yet most sequences can fold into a secondary structure, and oftentimes a functionally essential secondary structure. And then in many cases, this secondary structure can fold in upon itself to form elaborate tertiary structures, just as a protein will do. These are globular, and they're essential for the function of many catalytic RNAs. And what you're looking at right now is the progression of information on a Group II self-splicing intron, which is the largest ribozyme in nature, if you exclude the ribosome, which is also a ribozyme. But it's certainly the largest free RNA molecule that's ever been refolded and studied using the techniques of biochemistry and structural biology that are so important to all of us and members of this society. So I've focused on these kinds of problems for the last 20 years. And this, using Group II introns as a focal point, has helped us develop thinking about RNA and develop methods. But before I show you how we have come to think about its structure, let me just briefly tell you what a Group II intron does. This is also a lncRNA, a long non-coding RNA. It's essential for metabolism in plants, fungi, and yeast, and it catalyzes a self-splicing reaction. In other words, it's an enzyme that has to do a two-part reaction. So here is an intron surrounded by two exons, a 5' exon and a 3' exon. And this Group II intron, when it folds into its catalytically competent structure, activates a single 2' hydroxyl group. That means a single sugar within the whole structure. This is a big deal. Group II introns are usually 1,000 nucleotides in length. One sugar is activated, and the 2' hydroxyl group on it acts as a nucleophile attacking the 5' splice site that liberates this strand, and the 3' hydroxyl of the cut strand attacks at this other splice junction, liberating the free intron and ligating the exons. So that's a splicing reaction, and it's catalyzed by the RNA that you saw on the previous slide. So we have studied this for many, many years, and I'll show you a very recent result we have, which is that We've been able to take the molecule apart and study its individual domains. Group II introns have a bunch of domains, and the first domain, Domain 1, also folds first and forms a scaffold for the other things. We recently crystallized Domain 1 in isolation, and you can see it adopts kind of a more closed form. I'm showing it opening, and then we find that the other domains support that opening by docking sequentially into receptors within the Domain 1 scaffold. So there's a very orderly process to assembly of this seemingly really complicated and scary-looking RNA. I'd love to tell you all about all the cool interactions within this RNA, but we don't have time, and it's not really what I came here to tell you about. And we've published many papers on it, so if you're curious, you can look these up. But what I'm showing you here is just one of our crystal structures of a group 2 intron right before it does the 5' splice site cleavage reaction. Here is the 5' splice site inserted. That's the cisphosphate with the cyan dot. And what you can sort of see here without knowing much is that this is a really globular molecule. It's got a very well-defined form. So we've learned a lot about tertiary interactions, what they're made of, how they're stabilized. We've learned a lot about how RNA catalyzes chemical reactions. In work I won't tell you about, we've even found that small molecules can target pockets on this, and you can inhibit the splicing reaction just as you would a protein enzyme. And as it turns out, the structural work has revealed a lot of new things about the evolution of RNA processing. And has supported the notion that Group II introns share a common evolutionary ancestor with the eukaryotic spliceosome. And the spliceosome is this machine in your nucleus that is important for processing all of your mRNAs. And in case you don't think that's important, most of your pre-mRNAs contain at least 8 introns. So you do a lot of splicing, and it's catalyzed by the spliceosome, and it's thought that the active site of it looks almost just like this, and it does the same reaction. Anyway, by focusing on this one RNA, we've been able to learn a lot of things. And here are some of them. We've learned that multi-domain RNAs can fold accurately and faithfully to a specific structure. Just like a protein enzyme, the sequence contains the information needed for faithful folding, at least in certain RNAs. Probably not all, just as in proteins sometimes chaperones are needed, but many RNAs contain the information they need to get where they're going. We find in many cases that upstream regions, in other words the regions that are first transcribed, they tend to fold first, and they often will template the folding of the entire molecule. And a grad student in my lab who is very smart named Chen Zhao says, "First comes, first folds." And this is kind of a useful way to be thinking about RNAs that that you encounter, that transcription matters, and the timing of transcription relative to RNA folding is often important. So we think that there's sort of, in many cases, a folding frame that's hardwired into the design of many RNAs. In other words, that when you're transcribed is part of a process that's related to your structure. RNA tertiary structures can be conserved and they can be targeted, and you can use crystal structures, or now we're also increasingly using cryo-EM. We talked about that earlier in the meeting. To reveal the building blocks of tertiary structure and provide the basis for prediction. And what we'll talk about today is that this work has also led to things that will aid in the characterization of much harder problems, like the plethora of non-coding RNAs that are being discovered in genomic studies. So put in sort of a pictorial representation, can we leverage what we've learned from molecules like this to study molecules as complicated and intimidating as this? And what I'd like to show you today, I'll give you a couple of examples where we actually think the answer to that is yes. It's slow going, but I think the precedents are there, and we're going to make fast progress over the next decade. What you're seeing now is a molecule I'll be showing you. It's a long intragenic non-coding RNA called hot air. Just a few words about what lincRNAs are. Before I do that, let me get back to one of Steve's comments, which alludes to the fact that we now know that non-coding RNAs are abundant and very important, especially in mammalian biology. And this is the result of the ENCODE project, where the transcriptome of mammals and other organisms was evaluated. And they reported the very shocking finding that not only are our coding sequences transcribed actively, but all of our genome is transcribed, and all these non-coding regions that don't encode for anything are made into RNA. So there's all this RNA in the cell, much of it is very important, but we have no idea what it does in most cases, and we don't know anything about its structure. So today I'm going to be telling you about one subclass of that RNA dark matter, as people like to call it. I actually don't like that term because I think it's not so dark, but anyway, it is a sort of vast universe of new things we're going to have to characterize. So we're biting off a chunk of that to study something called long intragenic non-coding RNAs. And these are RNAs that are actively transcribed from intragenic regions, or so-called regulatory regions between genes. People like John Rinn and Eric Lander have shown that there are about 3,000 of these expressed and in common amongst mammals, so almost everything we'll be talking about is a mammalian RNA today. These genes have little or no protein coding potential. Some of those that had been previously flagged as non-coding have now been shown to, in fact, encode small peptides. So when you come across a non-coding RNA, it can often be a deficiency of programs that are used to detect stop codons and other things. But still, even standing the test of time, there are at least several thousand that are upheld as non-coding. They are big. They are usually more than 1,000 nucleotides in length, and their typical function is in chromatin activation, repression, direct action on our genome to influence expression. And I'll show you a couple of different ways this works. Okay, so now I've just sort of in the first couple of slides shown you how I think about RNA structure. I'm going to show you the prevailing view of link RNA structure. So this is what most people think of when they think of link RNAs. This is the cell, and this is the link RNAs. And, you know, they have a point. I think many people working on transcription say the very act of transcribing these RNAs could itself be sufficient for gene activation at other promoters because it can— these RNAs can kind of attract transcriptional machinery to different loci and thereby influence gene expression without even having any information content in the RNA itself. So these are the RNA doesn't matter, but it its sequence and structure don't matter, but its very presence does, people. And so they're thinking like this. So I say to people like that, well, you know, almost any RNA has a structure. And they'll say, okay, so spaghetti, fine, you can fold it in a knot, but that doesn't change its function. It's just, you know, it's food. It's going to serve the same function whether it's like that or it's in a much more disorganized form when it hits your stomach. And they would be right. In many cases, I think these RNAs are going to have structures and they don't matter. But RNAs almost never just flop like spaghetti. They often adopt a structure of some kind. So I've been— what we'll be talking about today is, do they have structure? And does it matter? And the only way to tell is first to figure out if there's structures in there, and then to test their role in functions of these RNAs. So our very first work on this was done on an RNA, as I said before, called hot air. This is an RNA that was first identified by Howard Chang and John Rinn in 2007. It was isolated from different kinds of human fibroblasts. As it turns out, it plays a really important role in differentiating what kind of skin you have in different parts of your body. So if you look at yourself, The skin in different parts of you is actually quite different, and HOTAIR is one of the genes, and one of the— it's an RNA that helps make that happen. It's a 2,148-nucleotide spliced and polyadenylated antisense RNA that's transcribed from the HOXC locus, so these are pattern— these are patterning genes, and it represses transcription of HOX proteins from the HOXD locus. And Wren and Guttmann have hypothesized that it's likely to be highly modular and have specialized domains for drawing it to chromatin and doing different things. Okay, so that's pretty much where it stood. Over time, there has been a lot of validation that HOT AIR is involved in progression of cancer. It's certainly important for differentiation into different kinds of tissues. But does it have any structure? So how do you even work with a lncRNA? That's an important question. So during our work on very large self-splicing ribozymes, we learned that sometimes it's not a good idea to treat RNAs in the way that is historically most common, which is to take the RNA, heat it up to 95 degrees after you transcribe it— oh, I should back up. When we work with RNAs in vitro, we're usually making them in a test tube in a test tube. So we take a plasmid that encodes that RNA, we put a T7 RNA polymerase promoter in front of it, and we throw in the polymerase as a workhorse, and we make many, many copies of this RNA, and then we isolate it. So we can make milligrams of any RNA that we're interested in, and we can do biophysical analysis on it just as you would do on a protein. But we have an added benefit. We don't have to make it express in an organism. So our work is actually easier than protein biochemistry. But one of the things we learned is that it's a bad idea to take those RNAs and then purify them in a way that denatures them and then try to reassemble them if they happen to be very large. So we learned that it's best to do in vitro transcription, to then digest the template with RNase-free DNase. You get rid of the polymerase by digesting all of that with proteinase K, which is actually a really small enzyme. You use ultrafiltration through an Amicon column to get out all the little bits, and you only keep the large molecules on the top of the filter, and you purify using FPLC. So we're using FPLC, not gels, to purify giant RNAs. So this has helped a lot. It's really hard to run a gel on a giant RNA, unless it's an agarose gel, and then you don't have the resolution to really pull out well-defined things. So we treat our RNAs very gently, and I urge other people to do so too. They're never subjected to denaturing conditions after transcription. And you can even see this when you look at FPLC traces of the same RNA that's been subjected to different treatments. This is a natively purified and transcribed RNA in red. Here are samples of the same RNA that's been subjected to rapid heat-cool treatments of different flavors.. And you can see that it's really, in this case, only homogeneous if it's natively purified. This kind of procedure also lets you take the fraction of the most sort of homogeneous species. We take the middle. And then you can begin to do biophysical characterization on that. And we use the same techniques we've been using for years on proteins. We use analytical ultracentrifugation techniques. And we use one technique that is unique to RNA molecules, at least for now, although it's a lot like HD exchange in protein structural studies, and that is chemical probing of structure. So you take a pure fraction of a homogeneously folded RNA, and if you want to figure out where the regions of secondary structure are— so secondary structure in RNA are the regions of base pairing versus single-stranded— and what you have to do is subject this to chemical probes that will that only attack the regions that are single-stranded and leave the base-paired regions protected. There are a number of different chemical entities that can do this. There's one approach called SHAPE that was developed by Kevin Weeks that's particularly powerful, in which the 2' hydroxyl group of any floppy and unbase-paired ribose will form this acyl linkage here, And this will flag regions that are single-stranded versus double-stranded. A much older technique that we always use in parallel assesses whether a base is actually base paired. And this is where you take your nucleotide and you combine it with dimethyl sulfate, and that will tend to put a methyl group on the N1 position here of A, and any place where it modifies on the Watson-Crick base pairing face of a nucleobase will block the action of reverse transcriptase, which is how you figure out where these lesions are. So using single-hit kinetics, you modify your RNA of interest under conditions in which it's folded, and then you use reverse transcriptase to look at where those sites are, and those appear as strong stops that you can see by capillary electrophoresis, and this is just a trace showing on the hot air RNA the positions of modification here, and then you have markers that are basically individual nucleotides, and that allows you to construct a histogram. This is basically like a sequencing profile of unstructured and structured regions of an RNA. So if you have high shape reactivity, it means that that region is not base paired. Because we're not going to talk about hot air for long, I'm just going to cut right to the chase and show you the secondary structure that we devised for the hot air RNA through a number of experiments, all of which are described in this paper. This is the secondary structure of hot air. It appears in 4 domains. These were independently validated by just making these RNAs as isolated transcripts, in other words, just making this piece of RNA, and showing it has the same folding pattern as the intact molecule. The other thing that we did is, once you get a secondary structure that's experimentally determined by chemical probing, you can then build better alignments of primary sequence and determine if you have regions of covariation and conservation, and in that way, finally identify functional regions of the molecule. So when we started this work, people said that the hot air RNA was not conserved in sequence and structure between any different mammals. However, once you know that the secondary structure of a given species has a certain form, you can map all the sequences onto this, and we were able to show, for example, that in fact this stem this H10 stem actually has even sequence conservation that's conserved from humans all the way through mouse. So there are indeed regions of conserved sequence, but you can't find them in this sort of mass of linear sequence unless you have some anchor points that are provided by the secondary structure. So that was our first study. And we learned from it that lncRNAs can contain modules of conserved structure. Biophysical and chemical methods give you maps to navigate through these large link RNA structures, and these maps then allow you to build good alignments from different species, do phylogenetic analysis, and identify regions of conservation, covariation, and potentially function. So one way to think about this is that when you're confronted with a new non-coding RNA molecule, it's like you've had somebody shoot you with silly string— remember those cans and you spray them— and all this string comes out. At least this has a sequence, but you have no idea what to do with it unless you have a scaffold on which you can kind of drape it, and that is what's provided by the experimentally determined secondary structure. Before moving on to our next story, I'll tell you that a lot of people have also developed ways to do this chemical probing in vivo, especially groups like Howard Chang, and now increasingly Kevin Weeks and others. So you can also do these kinds of approaches in a cell, especially if you can amplify the products, and begin to get information not just on a natively folded molecule that you've been studying in a test tube, but in the living organism. So those studies are very helpful, and in many cases I'll be telling you about today, we're doing that as well.. But we find that the in vitro experiments are highly complementary because you can actually get a lot more signal and you can get a primary determination of structure. Okay, so that's sort of all background. This is how we came to this point in our work. And now I'll show you the complete story of how we've applied it to a very important link RNA that has been studied by a young woman named Nadia Dimitrova, who's now a faculty member at Yale in my department. And this RNA is really interesting. It's called lincRNA p21. It's interesting because this is a long non-coding RNA that is induced upon stress, and it's p53-dependent. So if a cell gets stressed out, you induce p53 expression of genes, and one of these is this lincRNA p21, and this lincRNA binds to proteins like HNRNPK, PK and other proteins, likely in the nucleus, and comes back and participates in the induction of other sort of survival genes that are made in response to stress. This has been most well studied in the mouse, and there's very little known about the human variant. In the mouse, it's thought that there is a function in the nucleus, but also there's export into the cytoplasm., and there have been many models for how this lncRNA is actually functioning in both compartments. We set out to figure out what the human RNA is doing and if it has a structure that is germane to its function. So just a little introduction to this RNA. This is the mouse variant. It has 2 exons, and it's about 2,200 nucleotides long. In this particular case, or at least this exon is. And the mouse gene actually does undergo a splicing event, giving you a 3,000-nucleotide-long fragment. The 5' end, that's actually only the terminal couple hundred nucleotides, is homologous to the human variant. Okay, so we go to the human variant. It seems to have a different gene architecture. The exon 1 appears longer, and it had been hypothesized that these are splice sites, and that this gene is actually spliced in humans too to give you this 2,000-nucleotide RNA. But nobody had gone in and actually fished out and mapped the splice sites in the 3' end of this RNA in humans, so we decided to do that. So first we decided to determine the splicing of this RNA. And so we just used simple primers to different parts of this RNA, including the 5' splice site and the putative splice junction. And we got an interesting surprise. These are primers that would tell you whether or not this RNA got spliced, this SP. And you can see in any of the different very diverse cell lines we've looked at, you don't get any PCR products across the putative splice site. You only see PCR products that span the intron, exon, or reputative intron-exon boundaries and within the exons. So this gene, this RNA, is not spliced. It's a single long exon, as it were. So as a result of this, nobody had looked hard at the sequence in here. They just assumed it was an intron and it was trash and it was thrown out. So we took a good hard look at that, and we had a big surprise. It actually has was conserved. The intronic region was sequence conserved, very highly conserved. In fact, we even could tell what it was from the sequence. It's an Alu element. It contains this transposable element that's common in many RNAs, and like many transposons, it has been tamed by the system to help it function in many RNAs in certain kinds of subcellular localization. So Alu elements play a big role in many different RNAs, and this guy has one. Okay, so then we got a little concerned, and we looked harder at the information on this RNA, and we decided to sequence the 3' end using 3' RACE. And we found that this was not the proper 3' end. In fact, the 3' end— there are two variants of this RNA. One ends at this point, generating about a 4,000-nucleotide-long fragment. The most abundant fragment is about 3,000. And lo and behold, if you pull out the proper 3' end— and these RNAs have all been verified by northern blots— you actually have a second Alu element. You have two inverted repeat Alu elements in this RNA. And we have two isoforms of the RNA. Okay. So now that we understand the primary sequence, which is really important, Never take for granted that you know the sequence of the RNA that you're working with. You have to go in and fish it out, and I would say from multiple cell lines, and really use RACE and other things to figure out if you have the right termini and the right splicing isoform. We then use the pipeline I showed you before to map the secondary structure and understand the function of this RNA. So this is the shape map on the two anti— antisense Alu elements of human link RNA p21. And you can see from the shape reactivity maps, the reactive moieties are only in the loops and in the junctions, and there are regions of nice base pairing of this RNA in both arms. And really, very gratifyingly, these do indeed have the same secondary structure as has been characterized for SRP. And other elements that have been well characterized and even crystallized. Just as an aside, one thing that was really cool when we went in and looked hard at this RNA is that there are many base modifications or editing sites in the RNA. So a lot of the RNA nucleotides have been transformed into something else relative to the genomic DNA. So there are apparent U to C transitions or transversions that are post-transcriptional in various parts of the single-stranded region of this RNA. We actually don't think they are U to C editing events because there is no known U to C editing enzyme in mammals. We think this is probably more like an unusual base modification that is put on, and if you saw Chuan He's talk yesterday, you'll see there are lots of different modifications that are put onto RNA, and we think these are just stopping the RT and looking like a base transversion. Anyway, so this RNA is subject to further processing. So we then— now that we have the secondary structures, we were able to use R2R, which is a program written by Ron Breaker's lab, to figure out regions of phylogenetic covariation in RNAs. So you can align— now you can align all the different mammalian lincRNA p21 sequences, and you can see that the structures that I just determined experimentally are phylogenetically true. And so therefore, this is likely to be the structure of the RNA in vivo. So you're probably saying to yourself, that's nice, that means you can probably solve the structure of this RNA. And that's true. I would love to, but actually Oliver Weichenreiter already succeeded in doing that and publishing it last year because our sense This element has a sequence that's almost identical to that of the Alu element that he recently crystallized. And I'm taking and showing you the PDB file of his structure. And given that it actually dimerizes, it looks a lot like what you'd see for an intact sense Alu element that we are working on. So I'm showing you this, and it's bound to SRP9, so you can see that these things that we pulled out Sure enough, they have a 3-dimensional architecture. It's not spaghetti. It's actually pretty well defined in secondary and tertiary structural space. Okay, fine. So we're back now with the spaghetti on the fork. Okay, so there's a structure in there. Does it mean anything? So now we're in a position to, for the first time, determine if a tertiary structure in a link RNA contributes to its function, and so we kind of seized this opportunity. First of all, we had to establish for the human RNA if indeed it is nuclear-localized and turned on by doxorubicin, which is basically turning on the stress response. So in a couple of different cell lines, and I'll just show you this for HCT116, you can see that you don't make any lincRNA p21 if you don't induce the stress response. If you do add a micromolar of doxorubicin, you do make the stress response., you see a lot of LINCRNAp21 produced, and it's mostly in the nucleus, but there's some of it in the cytoplasm. So this is actually a— well, this is sort of a ratio. I should have changed this. In higher amounts of doxorubicin, that ratio gets even bigger, so you get even more nuclear localization. So indeed, LINCRNAp21 is enriched in the nucleus in these cells, and it's induced by stress and DNA damage. Okay, so then, now that we know there's a structure in there, we can start to tinker with the structure and see if that affects nuclear localization. So here are a couple of the different— here's one of the Alu elements that we found. What we did is either in both of them or individually, we changed one of the pairing strands to something that's no longer complementary., and we transfected this RNA into cells, and what you see here is that for— we'll just look at one of the isoforms, this is the long isoform of the RNA— the wild-type RNA is very highly nuclear-localized if it's intact, but if you start banging up the base pairing, your nuclear localization radically changes. And you can see this by FISH, where you can see the wild-type RNA is localizing exclusively in the nucleus,, and actually it localizes even more to perispeckles, and we'll get back to what that is. That's another important subnuclear compartment. Our mutants, where we bang up the base pairing, these guys now start spreading out all over, and also into the cytoplasm of these cells. Okay, so what have we found at this point? We found that these inverted repeat Alu elements facilitate nuclear localization of lincRNA p21. So the structures are important for localizing this RNA where it needs to go. They also help to present the binding sites for this hnRNPK protein and thereby induce survival genes. And in our work recently, we've seen that they are also going into these perispeckles, and perispeckles have their own functional consequences. So we're excited because, yeah, this lincRNA has structure in it, and that structure does something. Now, we don't have much more time, so I'm going to really go fast through another one. And that is the other thing we've been working on is one that's been well characterized, and this is a lncRNA called RepA from mice. Those of us who are female are born with two X chromosomes. We have to turn one off or we make too many of certain kinds of genes, and that's accomplished by transcription of a giant RNA, it's 20,000 nucleotides long, that coats one of your X chromosomes so that it's inactive. The existRNA, as that big RNA is called, is initially transcribed by the early expression of a small RNA called RepA that's analogous to the 5' end of your existRNA. So these guys are involved in dosage compensation. So we set out to figure out if RepA has a defined structure. And very briefly, we looked at its compaction into a discrete structure by sedimentation velocity experiments. And indeed, we found a magnesium concentration where it could basically stop folding. Under that condition, we mapped the secondary structure, as I've shown you before, and were able to superimpose on that regions of covariation. Through that, we've identified 3 domains in RepA. These were validated using a method I didn't show you earlier, but it's kind of cool, so I'll show you here. Here's full-length RepA. These are different fragments we made. I showed you there were 3 domains, so we made individual RNAs corresponding to those domains and did the same experiments I just showed you on those, and then on regions that span those domains that should be nonsense sequences. And you can see that the correlation in structures between the proper domains and the full length are pretty good, but not for the nonsense sequences. So what did we find, to cut to the chase? The domain 1, which is the domain that people have focused on a lot in studies of exist and RepA, has some really important regions in it, but they're not organized the way people thought. Everybody thought these little sequence repeats of GC-rich stems all looked like this, but we find that they don't. They are in different structural contexts. And what they really do is they present 3 highly conserved single-stranded regions, and these are linker region 3, 4, and 7. And you can see these long poly-U-rich regions are extremely conserved, including their surroundings, and we also find these are important for binding the proteins that help this RNA do its job. We then did something really weird that hasn't been done yet, which is we then did cross-linking studies to look and see if some of these regions were close together in space. And here's the secondary structure. We saw lots of cross-links. And to cut to the chase here, we've been able to use these together with a modeling program called Composer to show that regions on domains 2 and 3 form a tertiary structure, and that regions near those cross-links are again highly, highly conserved in sequence, and therefore likely to be important. So here is what we think is a structural module in far domains of the RepA RNA that nobody's been paying attention to, and they've all been assuming the action is at the 5' most end. Okay, so what have we learned from all of this stuff? We've learned that structure matters for link RNAs in primary sequence all the way to tertiary structure. So I don't think that in all cases the spaghetti model holds. We've shown that a native purification pipeline can help you study the native fold of link RNAs in vitro, and then this can set the stage for phylogenetic and subsequent in vivo studies. We've studied the in vivo secondary structure of many of these RNAs, and by and large they tend to be consistent with what we see in vitro. So there still is a place for in vitro studies of biophysical properties of RNA molecules, just as there are of proteins. Chemical probing can enable secondary structure prediction, leading to the identification of primary sequence and conserved secondary structure. Again, it gives you a scaffold or a shape to work from. And certain long non-coding RNAs do contain regions of elaborate tertiary structure, and this contributes to their function in regulating gene expression. So with that, I'll close by thanking the people who did this work. Almost all of what I showed you today was done by 3 people. Isabelle Chillon did the entire lncRNA p21 project by herself.— and it was her idea. She got in contact— actually Nadia Dimitrova got in contact with her, and now they've set up a collaboration so we can study all of this now in the mouse, which is where Nadia works. So we're very excited about that. The work on RepA and on hot air was done by Srinivas Samarthu and Fei Lu. Sri just got a job, and he's a professor now at Drexel, so we're really excited. These guys have bravely studied a very controversial and difficult problem. And we want to thank these people for participating and supporting us in these studies, including our faculty colleagues at Yale. And if you're curious about what you're doing— what we're doing— or you want to learn more about RNA folding, go to our website. And I also want to say I'm grateful for the support from these institutions, and I thank you for listening to this story, and And I'd like to take any questions you might have.