Scientist Stories: David Bartel, Non-coding RNAs and Regulating Gene Expression

Thanks very much, Phil. I didn't mention earlier, but it is our tradition at the symposium not to have questions following the talks. There's a variety of reasons for that, and one of them is that the speakers simply can't see the audience from here. There's blazing lights. So if you have questions for the speakers, I invite you to come up at the end of each session. They'll be hanging around and be happy to talk to you. So we're going to get into the symposium proper after a wonderful opening overview by Phil, who really provided color to what's described as the RNA world. And the RNA world is just getting richer and richer and more interesting by the day. Our next speaker actually helped convince us that there once was an RNA world through his efforts as a graduate student with Jack Shostak, where he demonstrated that one could evolve RNA molecules with catalytic activity, including polymerization. Giving life to the idea that once upon a time before there were cells, there were RNA molecules that could make more of themselves. And that's David Bartell, who's now a professor in the Department of Biology at MIT and a member of the Whitehead Institute for Biomedical Research. Following his graduate work, David joined the Whitehead faculty and has continued on in his study of RNA in various ways, including importantly to that explosion of information about microRNAs through deep sequencing, first in C. elegans and then mammalian cells and plants. David has gone on to study the mechanism of action of microRNAs and their targets, and he's expanded his reach to study other regulatory RNAs, including long non-coding RNAs. David has won many awards and honors in his career. He's a Howard Hughes Medical Institute investigator. And a member of the National Academy of Sciences.

David. Thank you, Tyler, and I appreciate the opportunity to tell you what we're learning about microRNAs and poly(A) tails and translation. So as Phil told you, these microRNAs are these small RNAs that are loaded into Argonaut proteins coming from these hairpin precursors. And in the Argonaut, if they have very extensive complementarity to their targets, like what we found in plants, And if the Argonaut has the activity, cleavage activity, they will direct cleavage of the mRNA, as siRNAs will do entering this pathway. And that happens a lot in plants. But in animals, you have primarily this other type of targeting, although there is a little bit of this happening in animals. But primarily what's happening is that microRNAs are regulating their targets through predominantly through these interactions to what we call the seed of the microRNA nucleotides 2 through 7, often supplemented by a pair here to nucleotide 8 or an A here across from nucleotide 1. Okay, so these 7 or 8 nucleotide sites, surprisingly little role for the 3' part of the microRNA. So this— these insights were actually found predominantly by doing computational analysis, looking at what are the types— looking in the UTRs at what are the types of pairing that are most conserved in evolution. And our most recent analysis of this has shown that for the very highly conserved microRNAs, the microRNAs conserved at least from human to fish, that they each have many conserved targets. After you subtract out what you'd expect by chance, each of these microRNAs— there's about 87 of these families, microRNA families— each of them have over 400, or on average at least, more than 400 conserved targets per family. Okay, now if an mRNA is targeted by one microRNA, it's often targeted by others, and so on average what we see is about 4 to 5 conserved sites per conserved target, but this still adds up to more than half of all the mRNAs with this conserved pairing to microRNAs. And then there's additional non-conserved pairing that can also be functional. And so if you are studying a gene in mammalian cells, which many of you are, it's clear that at some point in development, you know, that the microRNAs are going to be influencing the expression of that gene. And so it's going to be hard to find a disease or developmental process that isn't somehow influenced by microRNAs. And so to help people sort this out, you can just Google TargetScan, get to our website, and find these interactions that are preferentially conserved. And we've also done work showing— predicting which are going to be the interactions that are most— going to be most responsive to the microRNA, predicting those compared to all the other more minor interactions. But overall, many conserved sites per family. So how do these— this targeting of the AGO to the 3' UTR, how does that give downregulation? So work from many labs has shown that— and this is the current view of how this works— is you have Argonaut being recruited here to the 3' UTR. This recruits a protein that in flies is called GW182, TNRC6 in mammals. GW182 interacts directly with the poly-A binding protein. And most importantly, it recruits deadenylase complexes, PAN2 or PAN3, PAN2-PAN3, or more importantly, the CCR4 NOT deadenylase complex. And that contains enzymes that shorten the poly-A tail. And as the poly-A tail gets shorter, you end up with decapping and 5'-3' exonuclease activity degradation of the mRNA. So that's what's thought now to be the major mode of downregulation. And on top of that, there is also some translational repression. And that is also thought to be mediated by recruitment of the deadenylase complex. So in thinking about this, we thought that it would be really nice to have a high-throughput method for looking at the lengths of poly(A) tails, just to be able to monitor what the microRNAs are doing. And it'd also be interesting for other reasons. So we know that poly(A) tails are very important. The presence of a poly(A) tail is important for the stability of the mRNA and also for its translation, to recruit the translation initiation factors. But the question we had was, does it matter how long the tail is? Does it matter if it's 80 nucleotides or 120 nucleotides or 40 nucleotides? And that wasn't so clear. You could think about your experimental system. Do people know whether the length of the tail matters? Not just the presence of the tail, but does it matter how long it is? And so many systems unclear, but unless, of course, you're working in early embryos. And in early embryos, like in frog and flies, there you have certain messages where their tails get extended through what's called cytoplasmic polyadenylation. So in this case, we have cyclin B1 mRNA that gets longer, and when that happens, increased protein production. And that's important at this stage in the maturing oocyte. This is thought to be causal. It's not just— yeah, so there's evidence then that not only this correlation between longer tail and more protein translation, but it's It's also thought to be causal in that Marv Wickens' lab showed that you could append, through hybridization, tails onto a cMOSS mRNA. And when these tails are long enough, you actually get maturation of the oocyte. And so this is kind of an indirect readout of the protein translation, but evidence that increasing the poly(A) tail length gives you more translation in these systems. But what about in other systems? And so to look at this, Alex Subtony in the lab has worked out a way to use high-throughput sequencing to measure lengths of millions of individual tails. And he does some molecular biology here to get a cDNA that is the reverse complement of the tail attached to a little bit of the 3' UTR attached to adapters. He takes that cDNA without amplifying it puts it onto an Illumina cluster station where he generates clusters. Just showing two clusters here. This one has a relatively short tail compared to this one. And then while it's still in the cluster generation station, he'll do a primer extension reaction, putting in TTP and also biotinylated UTP, extend through the tail, And then he puts it on the sequencing machine and starts reading the sequence of the 3' UTR. OK, so now for each cluster, we know the sequence of the 3' UTR. And then we can flow in fluorescent streptavidin, take another fluorescent reading, and get a fluorescence reading that's proportional to the length of the tail. Longer tail gives you more fluorescent molecules in the cluster than shorter tail. And he puts in synthetic standards, and that allows him to get a standard curve here where he can just read out this fluorescence intensity as a tail length. So now he has these sequences of the UTR coupled with tail length, and then we get millions of those. And then we can start to look at poly(A)-tail distributions for different genes. So here's just 10 examples from 3T3 cells. Some genes have generally quite short tails. You can see the distribution here. Other distributions— here you have a gene with sort of centering around a tail of about 200. So this is just 10 examples. He, of course, has this for thousands of different genes. And so here in 3T3 cells, he's got these stacked up with the dotted line here being the average. Tail length, and each row here is the distribution for that particular gene. So this gives you a sense of what's happening for tails in 3D3 cells. And then he did this in other contexts. Here's budding yeast, fission yeast, Arabidopsis thaliana, mouse liver, these cell lines, and then the embryonic samples. These, I think, are the most interesting, these embryonic samples. Where you have a rather broad range of average tail length. But when you look at each row, the distribution is quite narrow. So lots of white space here and here. And these tail lengths sort of increase and get a little broader during the course of embryonic development. All right, so this gives us an overall view of what the tails look like in these various systems. One thing that Alex found was that the tail lengths tend to be conserved. So if you have a rather long tail here in HeLa, the human cell line, you get also a similarly long tail in mouse cell line. OK, so some conservation of the tail lengths. What about the relationship between tail length and stability? Well, when we look at just the mean poly(A) tail length— so each of these dots is a different gene, the mRNA from a different gene— we don't really see a relationship between the half-life, as reported previously by others, and Alex's tail length measurements. If anything, it's a little bit negative here, but pretty weak overall relationship. But this is looking at steady state. And so what you can imagine could be happening here is that MRNAs with long half-lives could be transitioning through these tail lengths, you know, the tail's just getting shorter and shorter over time, more slowly than the mRNAs with short half-lives. Okay, so, and if you looked at steady state in this situation, the tail lengths would average to be exactly the same. Okay, even though there is something happening when you look at dynamics of tail length. And so Alex looked by metabolic labeling at the dynamics of poly(A)-tail deadenylation. And what he saw then is when he looked and plotted deadenylation rate rather than steady-state tail length, he did find a very clear relationship where the mRNAs with short half-lives tend to have a very fast deadenylation rate. Exactly what than we would have expected. What about the relationship between tail length and translation? So to look at translational efficiency, we used the method developed by Nick Ingolia when he was in Jonathan Weissman's lab. And this is called ribosome footprint profiling, where the ribosomes are arrested, and then the mRNA is digested away. That leaves these ribosome-protected fragments. And then we could just purify these protected fragments and sequence them. And that tells us where all the ribosomes are. You can count the number of ribosomes per mRNA. We do in parallel fragment the mRNA, sequence it, and that gives us the amount of mRNA. You normalize one to the other, and that gives the number of ribosomes per mRNA on average. And that's what we're calling the translational efficiency. And this is a method that was worked out in yeast. And Jonathan and Nick helped us adapt this to mammalian cells. And we used this initially for looking at the effects of microRNAs. And that's work I'll just briefly mention later. And that was work done by Hui Li. And so more recently, Stephen Eichorn has tweaked this method some more. And he began to look at some of the samples where Alex had measured the tail lengths. So these are Stephen's data on translational efficiency compared to Alex's data in mean tail length. And now we're looking here in the zebrafish, either at the cleavage stage or at the blastula stage. And what they found was that, as you'd expect based on the Xenopus experiments, there's a really striking relationship between the length of the tail and the translation efficiency. You increase tail length about 2-fold here, and you get about 5, 6, 7, 8-fold increase in translation efficiency. So that's what we see at cleavage in blastula, and then at gastrulation, it disappears. And the same thing happened when we looked in frog. Okay, this is the comparable stages in Xenopus. Cleavage blastocyst at gastrulation, pretty much gone. And also when we looked at post-embryonic samples, whoops, a similar— oh, I forgot to say that what we're looking here is at sample data. So you might say that, well, maybe the reason that this this relationship is missing here is because the range of tail lengths is now more narrow. And that is true. It is somewhat more narrow here in these later stages in development. But even if we sample these data using the tail lengths from this distribution, we still end up with a very clear relationship. And the same thing is true. In the frog. Then looking at post-embryonic samples, we see that in yeast, in budding yeast, fission yeast, mouse liver, r here of 0.00, not a whole lot happening in terms of the relationship between tail length and translation efficiency. Same thing in these cell lines. So how do we think about this? Well, first thing to point out is, as you all know, in these embryonic samples, there's this transition in transcription, where you start off with all the mRNAs here being inherited. So these are maternally inherited. And then at this point, somewhere between 2 and 4 hours, the zygotic genes begin to be transcribed. And now the gene regulation is more under zygotic control here. And so what we're seeing is that after this transition in transcriptional control, there's a transition in translational control, where translation starts off here as coupled to tail length. And then it becomes decoupled. And in all those post-embryonic samples, decoupled, although we suspect there'll be some places, maybe in synapses, where it'll be coupled again. So how do we think about this? Well, again, at this stage, the regulation's under transcriptional control. There's plenty of transcription going on. Here, there's none. Here, there's also a lot of regulation happening at the level of differential mRNA stability, okay? Whereas at this stage, if there's no transcription, there's less opportunity for this differential mRNA stability because if the mRNA goes away, the cell isn't going to see it again until transcription starts, okay? So at this stage, without transcription and with limited utility for differential mRNA stability, the cells have another way of regulating the genes, and that is this— remember the cytoplasmic polyadenylation. They have very robust cytoplasmic polyadenylation changing the lengths of tails. And, of course, when that happens, it'd be very nice to have a way of reading that out. And so tail length and translation is coupled. OK, and that's kind of the picture that we had originally. I think what's the real insight here comes in the loss of this coupling. And the way we think about that is that if many genes are being controlled by differential mRNA stability, there's actually a disadvantage to coupling tail length to translation. Because if the mRNAs have been in the cytoplasm for a long time, their tails are just going to get shorter and shorter. And so for an mRNA molecule that's been in the cell for a very long time with a rather short tail, if it wasn't translated, what would be the point of having it there? So when the cell is using differential mRNA stability and has many of these mRNAs that that have been around for a long time in the cell and still wants to use them, it does not want to do this age discrimination against the mRNAs that have been there a long time that have the shorter tails. And so the tail length and translation are uncoupled. So that's the way we can rationalize these observations. And then here, you still have— despite having some transcription, you still have this coupling. But it just may take some time to get rid of it, and so— but this is what we see. So how then do we relate this to the microRNAs? Well, for some time we've been looking at the molecular consequences of microRNA targeting, and we've known, even though this was kind of a surprise at the time, that there's clearly a decrease in the mRNA, okay, when the microRNA is induced, okay? So this wasn't originally thought to be the case. Originally, it was thought that all the regulation was at the level of translation. In fact, even in Phil's introduction slide, it still had microRNAs as translational repressors. But it turns out that there's quite a bit of repression happening just at the level of the mRNA level, okay? So here you have a gene that has 4 mRNAs. You add the microRNA, and now you have 2 mRNAs. And then the question has been, well, on top of that decrease in mRNA, is there also a decrease in the amount of protein being made, the number of ribosomes on that mRNA? For the mRNA that's left, are they left efficiently translated? OK, and so we've been looking at that using the ribosome profiling that I described, and originally using SILAC proteomics in the ribosome profiling. And what we see is that in general, the ribosome and the protein changes mostly are reflected by changes in mRNA. OK, so here you have 16 ribosomes making protein. Here you have 8. And the reason for that is that the number of mRNAs has gone from 4 to 2. And we don't really see any evidence for some targets being repressed only at the protein level. But with these very sensitive methods, we can detect some decrease in the translational efficiency. So really, instead of having 8 here, maybe you have 7. OK, so you go from 16 ribosomes to 7 ribosomes, mostly because you have less mRNA, but also a little bit of the repression because there's a little bit less translational efficiency. And we've looked at this now in a lot of unpublished work by Hui Li, but mostly Stephen Eichhorn. And what he finds is that there's some cell-to-cell differences, but overall, 70% to 80% of the repression is explained by the decrease in the mRNA. It doesn't matter if the cells are dividing, stressed, transformed, not transformed, if the cells are dividing, arrested, stressed, not stressed. It doesn't matter if the level of target's high or low. This is what we see. This is all, of course, looking at steady state measurements. But we've also done time courses looking at the dynamics of regulation. And in mammalian cells, we just have not been able to find a setting in which the dynamics of microRNA repression imparts some sort of substantial effect observed only at the protein or translation level— protein or ribosome level. So this is actually great news. For those of you wanting to know what a microRNA is doing in your system, you can get a quite quantitative readout just by looking— perturbing the microRNA and looking at what's happening to the mRNA, which is a lot easier to do, as Phil was saying, than looking at protein or ribosomes. So this is great. But there is an exception. And this is found by Antonio Geraldi's lab in the early a zebrafish embryo. And he was looking at the regulation by miR-430. This is a microRNA that comes up very strongly between 2 and 4 hours. And what they found using the same ribosome profiling, mRNA measurements, was that the target ribosome-protected fragments, the ribosomes, went way down here after the microRNA came up. The mRNA also went down, but there was a— substantial delay, so that the mRNA went down at this transition. So here, all the effects of the microRNAs could only be seen by looking at the ribosomes. If you were looking at the changes at the mRNA at 4 hours post-fertilization, you'd completely miss what the microRNA was doing. And then later at 6 hours, he saw what we see in all these mammalian systems where much of the change in the ribosomes is explained by changes in the mRNA. So the interpretation here was that this is due to the dynamics of microRNA regulation, that there's this initial translational repression, and then this lag phase, and then the mRNA destabilization. And this made a lot of sense. But that was before anybody knew about this transition in translational control that happens right also at this stage here. So we decided to look at this more in depth, transfecting— or in this case, injecting microRNAs, two different microRNAs that are not normally expressed at this stage in the zebrafish. And then just looking at the translation, the mRNA, but also the tail lengths at these 3 different stages. And this is with the help of Grace Chen and Hazel Siv. And so what we found is that— and in red here, or orange, are what's happening to the mRNAs that have a site to the injected microRNA. So you see that if they change, they go down. The ribosome-protected fragments from those mRNAs— this is ribosome-protected fragments, this axis here— go down, whereas the mRNAs without a site stay the same. It doesn't really matter plus or minus the microRNA. The mRNAs, though, do not change for either those that have sites or those that don't have sites. Same thing at 4 hours post-fertilization. Those with sites to the microRNA, if they change, they go down. And some of them don't change. And again, that's a whole other story that we have predictions of which are going to change and which not. And we're just looking here at those that have sites and those that don't. But here we see at 4 hours, again, we see a change in the ribosomes. The ribosomes go down in the presence of the microRNA for the mRNAs that have sites, but they don't for the mRNAs that don't have sites. And there's no change in the mRNA. And then at 6 hours, now for those mRNAs that have fewer ribosomes, there's also less mRNA. OK. So a lot— very quite similar to what Gerald has found with miR-430, where you have this early situation where you only see a change in the ribosomes, no change in the mRNA, and then between 4 and 6 hours, decrease in the mRNA corresponding to the change in the ribosomes. But there's an important difference here. In this case, the microRNA was there from the very beginning, yet this transition happened at the same time. Okay, so this told us that it wasn't the dynamics of microRNA regulation that was doing this. It's probably this transition that we had discovered earlier. And so we looked then at the tail lengths. And what Steven and Alex found was that for those mRNAs with sites where the ribosomes were going down, you had a corresponding change in the tail length at 2 hours and at 4 hours. But at 6 hours, again, the ribosomes would be going down, but with really not much of a tail length difference. Remember, here you have this mRNA difference. Now we're looking at the RNA change, and there, that's what corresponds to the change in the ribosomes. So just the model to summarize this— at 4 hours, the microRNA is induced. It causes tail lengths to get shorter, and that causes less translation of those mRNAs. And this relationship is really quite strong. You decrease the tail length in half, and you get about a 5-fold decrease in the number of ribosomes. That's at 4 hours. And then between 4 and 6 hours, you get this uncoupling of translation and tail length. And what happens when the tail lengths get shorter? The microRNA is induced, the tail lengths get shorter. And in this case, in this regime, the mRNA is destabilized, whereas here it remains stable. And on top of that, there's a little bit of translational repression, which is not due to the differential tail lengths, because there is no correlation between tail length and translation here. And that's, again, due to this other consequence of bringing in the CCR4 NOT complex. All right, so this is how we explain what's happening here. And so this discovery of this switch in translational control, we think it's very interesting in its own right. It's also teaching us— of course, we'd like to know the mechanism of that. If there are questions, I'm sure that'd be the first one. And we don't know yet. But it's still teaching us some important things about microRNAs. It explains why there are very different consequences at 4 hours and at 6 hours. This is not because microRNAs are doing something different at 4 hours than at 6 hours. It's not because of the dynamics of microRNA regulation. It's because in both cases, what the microRNAs are doing is shortening the tails. And there's just different consequences of shortening the tails. At 4 hours, that causes less translation. At 6 hours, that causes mRNA destabilization. And this also explains why the early embryo is the only system in which translational repression is known to dominate. So I just want to close by thanking these people for things I mentioned, particularly Alex and Steven for the tail links and the ribosome profiling, and you for your attention.