VIDEO: Christine Mayr on Condensate Networks

Christine Mayr (02:54):
And more recently, we found that mRNAs can also contribute to the generation of local environments at the site of translation. And these local environments can also regulate protein functions. So why I’m so excited about 3’UTRs? So when the human genome was sequenced, people were very surprised to find that the number of protein coding genes is actually quite similar between worms and humans. So this led to the idea that it’s mostly regulation that has increased in the evolution of multi-cellular organisms. And so this regulation is basically more enhancers, more non-coding RNAs. But one part of this increased regulation is actually contained in mRNAs, and it’s due to longer 3’UTRs.

Christine Mayr (03:47):
So what you see here, I’m drawing typical mRNAs from different organisms. And you see that the distribution of coding region length actually has been quite constant. But 3’UTR length, here drawn in blue, you see has expanded quite a lot. And if you look at human mRNAs, the 3’UTR for a typical mRNA is as long as the coding region. And you will see on the next slide, half of all human genes make alternative UTRs. And in those genes, UTRs are usually twice as long as the coding region, at least. And so UTRs are much less conserved than the coding region, but still about a third of the nucleotides are conserved in UTRs. And so therefore we think, because every protein is generated by an mRNA, and so you can basically put a lot of regulatory elements into 3’UTRs that then can influence protein functions. My lab is really interested how 3’UTRs do this and how they basically influence protein and cellular functions.

Christine Mayr (04:57):
And so when I started my lab, I wanted to study 3’UTRs, but at the time, 3’UTR boundaries were not very well annotated. So we developed a sequencing method that allows us to identify all 3′ ends of the transcriptome. So that’s a method called 3′-seq. And we applied it to many different cells types and tissues. And what we get is… Here at the boundary between the 3’UTR and the poly-A tail, we get a peak. And so half of all genes have basically one peak. This means they make one 3’UTR. But the other half, and this was a surprising finding, basically has at least two peaks in the 3’UTR. So this peak is generated when a proximal poly-A signal is used, and it makes an mRNA with a short UTR. And this peak here is made from the usage of this poly-A signal. So here the mRNA has a long 3’UTR. In both cases, the protein that is encoded is the same. So the only difference is the 3’UTR sequence.

Christine Mayr (06:06):
And at the time I was actually very surprised about this. And so I thought it was very puzzling. So I thought, “Why would half of all human proteins be generated by these mRNAs that differ only in the 3’UTR?” And therefore, we basically set out to study what alternative 3’UTRs do to the protein. But the same study also revealed another surprising finding. At the time, people thought that the major functions of 3’UTRs are the regulation of mRNA stability and protein abundance. We also, of course, thought that this would be the case. So when we now compare two conditions… and I’m showing you, only here, B cells before and after Epstein-Barr virus transformation. You would think that mRNAs that changed the UTR isoform expression would also change the mRNA levels, because you now get rid of stability elements and now you change how much of the mRNA is present.

Christine Mayr (07:14):
But we actually didn’t see that. And so, instead what we found was that basically there is a bunch of mRNAs that change abundance in these two conditions, and there is a bunch of mRNAs that change UTR isoform usage. But there is very little overlap. So it really suggests that these two parameters of gene expression are orthogonal parameters, and they are not basically doing the same thing. So in the meantime, we have actually done this on 120 cell types, 50 pair-wise comparisons, and we always find the same features. So, some mRNAs change abundance, some mRNAs change UTR isoform usage, but very few change both.

Christine Mayr (07:57):
So what that means is, there is maybe 5% of mRNAs that really use the 3’UTR to change protein expression. And those mRNAs mostly encode cytokines and cell cycle regulators. However, the rest of the mRNAs, at least the ones that have alternative 3’UTRs, do not use the 3’UTR to make large changes in protein abundance. So that really led to the question, “What do 3’UTRs then do?” And so therefore, we looked a little bit more closely, and we found that 3’UTRs can actually mediate protein-protein interactions. And so we looked at CD47, because CD47 protein can be encoded from an mRNA with a short or a long UTR. And so we made GFP fusion constructs. So we fused GFP to the coding region, and then we either added its short UTR or its long UTR. And at the time, we really didn’t know what would happen. So we transfected them, these constructs, into cells. And so, CD47 is known as the plasma membrane protein. And when we use the construct with the long UTR, we see very nice GFP localization to the plasma membrane, as expected.

Christine Mayr (09:15):
However, the construct with the short UTR leads to a much more even distribution, and CD47 is mostly located intracellularly. So that was surprising. However, when we then… So those are HeLa cells. So when we looked at a different cell type and we transfect in CD47 made from a long or short UTR isoform, you see actually that the cell morphology has changed. So in the presence of the long 3’UTR, we see lamellipodia formation. And this doesn’t occur in the short isoform. So lamellipodia formation is a sign of active RAC. And therefore, we then basically looked with these cells that, where we transfected in CD47 with a long UTR, actually have more active RAC, and you can see, yes, they do.

Christine Mayr (10:05):
So active RAC is RAC1 bound to GTP. So, you do a RAC IP, and so you see that there is the same amount of total RAC. But these cells where we transfected in CD47-LU have more active RAC. So this was a really remarkable finding, because remember, we put in a construct that makes the exact same protein, and the only difference is the 3’UTR. In one case, it leads to activation of downstream signaling pathways. Whereas in the other case, it does not. So that means that 3’UTRs can actually really determine protein functions.

Christine Mayr (10:46):
And of course, we wondered how this works. And so we found that there is an RNA-binding protein called HuR that can only bind to the long 3’UTR, but not to the short UTR. And this RNA-binding protein then recruits an effector protein called SET. And so, recruitment of this protein to the site of translation resides in SET binding to CD47.

Christine Mayr (11:11):
So when SET is not recruited to the site of translation, SET cannot bind. And so the experimental evidence of this is shown here on the right-hand side, so where we did a Co-IP. So we pull down on CD47, either made by the construct with the long UTR or the short UTR. And you see that only in the case of the long UTR, endogenous SET can bind to the protein, whereas here, in the presence of the short UTR, SET cannot interact with the protein.

Christine Mayr (11:44):
So SET seems to be an important trafficking factor. So it does not regulate protein translation, because total CD47 levels are the same. However, surface expression of CD47 changes when we knock down SET. And so SET actually, and I come back to this, binds to positively charged amino acids in the cytoplasmic part of CD47. And SET actually regulates, also, plasma membrane trafficking of other membrane proteins, because they all have an increased amount of positively-charged amino acids in their cytoplasmic domains.

Christine Mayr (12:27):
Okay. So from this, we basically concluded that 3’UTRs contain genetic information on protein functions that can be transmitted to proteins. And so then, of course, we wondered, “How is the information that is contained in 3’UTRs transmitted to proteins?” And here, in the case of CD47, we thought it’s quite easy. So the information is contained in SET binding. So we’re basically… SET, first, associates with the 3’UTR and later binds to the protein. And we wanted to know, how is SET transferred from the 3’UTR to the protein. And we already knew that HuR is important.

Christine Mayr (13:13):
However, we hypothesized that there is a second RNA-binding protein that is required for efficient SET transfer. And the reason for this was that, when we used the long 3’UTR of CD47, we always get very nice SET binding to CD47. However, if, instead, we used a UTR that contains a very good HuR binding site but nothing else, we get a little bit of SET binding but not as much.

Christine Mayr (13:42):
So therefore, we hypothesized that there might be a second RNA-binding protein that cooperates with HuR in SET transfer. And we wanted to find the second RNA-binding protein. And so we did, actually, two screens, and in the very end, we identified an RNA-binding protein called TIS11B. And so this protein is an RNA-binding protein, and when we, here, knock it down, we lose SET binding to CD47. So you see, when we knock down QR, we lose binding of SET to CD47 but also when we knock down TIS11B. So this told us that TIS11B is required for SET transfer to CD47 protein.

Christine Mayr (14:29):
Okay. So when we knock down TIS11B, we looked at protein expression levels of all the molecules that are important in this process, and protein expression levels didn’t change. So therefore, we basically hypothesized maybe TIS11B makes a certain environment that is important for SET transfer. And because it’s an environment, we thought, “Let’s do imaging.” And this was basically the start.

Christine Mayr (14:57):
And so we did imaging, and we were very surprised. So we basically saw these granular assemblies that are in the vicinity of the ER. So I’m showing you here, in the first panel, HeLa cells… So here is the nucleus in blue. In green is the ER, and in magenta is TIS11B, and it forms this reticular network in the vicinity of the ER. And this is endogenous TIS11B protein. So then, of course, we looked at other cell lines and basically, in every cell line that we have looked so far, we see these granular assemblies. Also, mRNA levels of TIS11B are widely expressed, so we haven’t found a cell type that does not express TIS11B.

Christine Mayr (15:42):
Okay. So the only difference that we found was that, in some cells, there is really a very high expression, and so it leads to this ring in the perinuclear region. So basically, the whole perinuclear ER is covered by TIS11B. But there are other cell types, like fibroblasts, where maybe only a quarter of the ER is covered by TIS11B granules.

Christine Mayr (16:13):
And so then we wanted to have a closer look, and we turned to high-resolution live-cell imaging. So, again, in green is the ER, and in magenta is the TIS11B. So here is one cell. The black hole is the nucleus, and you really see it forms these reticular assemblies that are intertwined with the ER, so. At first, we thought it actually looks quite similar to the ER, but then, after a while, you notice it’s a little more bulky. But it really shares a lot of surface with the ER, and it’s mostly in the perinuclear region, which is the rough ER, which is the major site of protein synthesis in cells. So we were very happy when we saw this, because we have never actually seen anything like it before. But the imaging didn’t really help us to find out what TIS11B actually does.

Christine Mayr (17:12):
Okay. So, what we then thought is, okay, so TIS11B is an RNA-binding protein that binds to AU-rich elements. So what is the relationship between TIS11B and CD47 mRNAs? So we did FISH, so single-molecule FISH, on our construct, so CD47 with GFP, followed by the short UTR or by the long UTR. So on the top panel, we have the long UTR. So in red, this is TIS11B in a HeLa cell, and you see that… And in green is the RNA. You see that there is very nice colocalization. So basically, all transcripts of CD47 with a long UTR colocalize with TIS11B granules. And when we zoom in, you can see it even better. And to basically be able to do quantification, we made these line diagrams, where you see very nice colocalization and a very high correlation coefficient.

Christine Mayr (18:12):
So when we look at CD47 with a short UTR… So CD47 short also needs to be translated at the ER because it has 5 transmembrane domains. But the RNA basically localizes mostly outside of TIS granules, so you can see that the green dots are mostly adjacent to the red dots, and there is very little colocalization. And many green dots are in areas where there is no TIS granules. So that means that CD47 with a short UTR localizes to the ER in a different area than CD47 with a long 3’UTR.

Christine Mayr (18:50):
Then, of course, we thought, “Okay. Does this only happen for CD47 or is this widespread?” So we tested 15 candidates, and we learned two rules. And so basically, if you have a membrane protein that has a lot of AU-rich elements in the 3’UTR, nearly all of the transcripts localized to TIS granules. However, if you get rid of the 3’UTR, they don’t go to TIS granules. So I want to guide you through this.

Christine Mayr (19:24):
So here’s BCL2, and it has a very long 3’UTR with many AU-rich elements. So when we use the construct with the long UTR, you can see here there’s very nice colocalization, also here in this line diagram. If we now take the same protein, but we delete the 3’UTR, now, in green, BCL2 mRNA localizes to the cytoplasm and is actually excluded from the region of the granules. And so here’s a selection of the other candidates that we looked at. And the bottom line is that, as I said, mRNAs that encode membrane proteins that have a lot of AU-rich elements, they go to TIS granules.

Christine Mayr (20:12):
All right. In the meantime, we actually purified TIS granules, and we are able to confirm this finding. And so basically, what we found is that all mRNAs that have a very high AU content, and so we have over a thousand mRNAs here that are highly enriched in TIS granules, whereas mRNAs with a low AU content are depleted from TIS granules. So it seems that TIS granules is a subcellular compartment that allows translation of AU-rich mRNAs.

Christine Mayr (20:51):
Okay. Okay. So now we knew that RNAs can differentially localize to TIS granules. But, in the end, we wanted to know, what’s the advantage of being translated in TIS granules? And luckily, because we found TIS granules, because we were studying the interaction between SET and CD47, we looked at the interaction, at the protein-protein interaction between SET and CD47, to see, is the granule actually required for this protein-protein interaction?

Christine Mayr (21:30):
Okay. So here you see a cartoon of CD47. So it’s translated on the ER and has five transmembrane domains. And in the cytoplasmic part, it has several stretches of positively-charged amino acids. SET is highly negatively charged, and it binds to these positively-charged amino acids, because if we mutate them, SET doesn’t bind any more. So, and I already showed you this, if we knock down TIS11B protein, then SET can no longer bind to CD47, shown here. But here we basically knocked down the protein. So at the time, we didn’t know, is the protein required, or is the TIS granule actually required?

Christine Mayr (22:16):
So luckily we have a mutant. And so this mutant is still the wild-type TIS11B. However, we made a few point mutations, and now it can no longer make assemblies, and so this mutant is shown here. And so now, we basically ask, is the granule required, or do you only need the RNA-binding protein TIS11B? And as a readout, we used surface expression of CD47 because, when SET binds, CD47 traffics much better to the plasma membrane. And you can see, we get a nice increase in surface expression in the case of the wild-type TIS11B, so that means in the presence of granules. However, when we used the mutant TIS11B that binds perfectly well to the RNA, then we have very little increase. So this really suggests that it’s the TIS11B granule that is required for the protein-protein interaction between SET and CD47.

Christine Mayr (23:19):
We have other examples, like PD-L1. So PD-L1 has the same architecture in their cytoplasmic domains. So basically, it has a few regions that are positively charged that can bind to SET. So when we use a construct that contains the UTR with the AU-rich elements, then PD-L1 RNA perfectly colocalizes with TIS granules. When we delete the UTR, now PD-L1 RNA still localizes to the ER but no longer to TIS granules. And you can see it’s actually depleted here, in this region surrounding the nucleus. Again, in the presence of the UTR, it binds SET. In the absence of the UTR, it doesn’t bind SET. And then in the presence of the UTR, it goes to the plasma membrane, but in the absence of the UTR, it doesn’t.

Christine Mayr (24:12):
Okay. So again, so there is this correlation. If you have a UTR with AU-rich elements, you localize to TIS granules, and you can bind to SET. So now the question was, is it the UTR that is required, or is translation in TIS granules the important event for the protein-protein interaction between SET and CD47 to form? So to address this, we turn to CD47-SU, so CD47 that was made from a short isoform, does not bind to SET, and does not go the TIS granule. Okay, so we wondered, can we force CD47-SU to be translated in TIS granules? How do we do that?

Christine Mayr (25:00):
So most of the time, we express TIS11B in a way that it looks very similar to the endogenous expression level. So basically, not all regions of the perinuclear ER are covered by TIS11B. However, as I showed you before, there are cells that have a high endogenous TIS11B expression, and then they basically have this ring of TIS11B in the perinuclear region, so where basically the whole perinuclear ER is covered by TIS granules. And we call the ER domain that is covered by TIS granules the TIGER domain, for TIS granule ER domain. So basically, in this cell here, all the perinuclear ER is the TIGER domain.

Christine Mayr (25:48):
And so now, CD47 with a short UTR still needs to be translated on the ER, and the only place where this can happen is in the TIS granule region. And so, when we look at that, now we see that… So in cells that have a lot of TIS11B expression, we see that also CD47 made from the short UTR can bind to SET, and it behaves equally like CD47 with a long UTR, because it can travel very well to the plasma membrane. So that shows you the deciding event is not the UTR. It’s basically, if a protein is translated in TIS granules, it can form the protein-protein interaction between SET and CD47. And this protein-protein interaction cannot occur when the protein is translated outside of TIS granules.

Christine Mayr (26:43):
So the UTR is still important because it basically allows the protein to be translated in a specific environment. And the UTR probably contributes to the environment. But here, it’s really translation in a specific subcellular compartment determines the protein-protein interaction between SET and CD47.

Christine Mayr (27:10):
So, in summary of this part, so we found that TIS granules form this reticular network that is intertwined with the ER, and we call the ER that is covered by TIS granules the TIGER domain. So an mRNA that contains many AU-rich elements will localize to this domain. And it’s translated on the ER and now can basically form a protein-protein interaction. And in the case of CD47, this leads to higher surface expression. CD47 made with a short UTR does not contain the AU-rich elements, so therefore it’s translated outside of TIS granules, and therefore it cannot bind to SET. And therefore it cannot efficiently localize to the plasma membrane and is expressed intracellularly.

Christine Mayr (28:03):
So we still don’t know what’s special about the translation environment generated by TIS granules, but we know we can visualize it. So what I’m showing you here is live cells in the top row, where we transfected in TIS11B and SET, and you see SET is really everywhere in the cell. Then we fixed the cells, and then we permeabilized them and wait for one hour. And then we looked, what happens to SET. And so I’m showing you a zoom-in of the cell here. You see that… So SET is a very small molecule that diffuses out, so we make holes into the plasma membrane, and it can diffuse out. But SET that is localized in the TIS granule region stays. So, as I said, we don’t know what is different in the TIS granule environment, but it’s definitely different from the cytoplasm, because SET has a very different behavior.

Christine Mayr (29:06):
Okay. So those are our current questions that we are trying to address. So we would like to characterize the TIS11B granular environment. We also would like to know what are the major functions of TIS granules because we identified TIS granules… We came, basically, through the back door because we were studying the interaction between SET and CD47, and this led us to find TIS granules. But I showed you there is more than a thousand mRNAs that are translated in TIS granules, and we think that the interaction between SET and CD47 is, of course, not the most important one. And we would really like to know, what is the major function of TIS granules? And to get at this, our first step was to purify TIS granules and to identify the granule-enriched and -depleted mRNAs. Then we want to know what other reactions are promoted or inhibited when proteins are translated in TIS granules.

Christine Mayr (30:04):
And the question that I want to focus the rest of my talk on is at the bottom. So TIS granules look very different from most membrane-less organelles, like RNA granules. And we wanted to know what determines the 3D organization of TIS granules?

Christine Mayr (30:22):
So when you look here in the top row, you see that nucleoli, P bodies, stress granules, P granules, they’re all sphere-like. But TIS granules have this mesh-like morphology that is intertwined with the ER. And we think that, for the trafficking of membrane proteins, it’s really important to have that mesh-like morphology, because then it can share a lot of surface area with the ER. And so therefore, we want… So we think that this morphology actually contributes to the function. So therefore, we wanted to know how is this mesh-like morphology generated. And so, TIS11B is an RNA-binding protein, and we thought we would, just basically, first identify the region of TIS11B that is required for formation of this 3D organization. So it has a double zinc finger domain.

Christine Mayr (31:17):
First, we only only basically used this RNA-binding domain, and you can see the RNA-binding domain, alone, is not sufficient for this mesh-like morphology. But then we fused the RNA-binding domain to multivalent domains. First, we used SUMO-SIM. So we used SUMO-SIM because it forms really nice sphere-like granules in the cytoplasm. However, if we now add the RNA-binding domain of TIS11B to SUMO-SIM, then we get a very different morphology. You see that now it makes much more filamentous condensates. But these filamentous condensates are not intertwined with the ER. So this was a real eye-opener, because up to that point, we always thought that this mesh-like morphology requires ER. And this experiment showed us that, actually, the morphology is independent of the ER, because you can still make a filamentous condensate that is not intertwined with the ER.

Christine Mayr (32:24):
All right. So then, instead of SUMO-SIM, we also used the IDR of FUS fused to the RNA-binding domain of TIS11B. So FUS IDR alone does not form condensates in cells. However, if we fuse it to the RNA-binding domain of TIS11B, you can see it forms these reticular granules that look very, very similar to wild-type TIS granules, and they’re very nicely intertwined with the ER.

Christine Mayr (32:58):
So then we did some FRAP, and then we saw something very surprising. So we found that both FUS-TIS and SUMO-SIM-TIS, the protein components are actually quite dynamic. So basically, you can see they recover very fast, so that tells you that these network-like condensates are not aggregates. So basically, their protein components are highly mobile, but they are not sphere-like. And so therefore, we basically think we wanted to address a much broader question, because now we can basically we can make mesh-like RNA granules in cells that are not sphere-like, but they still have highly dynamic protein components. And we wanted to know, what, basically, leads to the formation of mesh-like condensates that are liquid-like?

Christine Mayr (33:54):
Okay. So it seems that the RNA-binding domain of TIS11B plays a crucial role because when we fuse it to SUMO-SIM or to FIS-TIS, in both cases we got filamentous condensates. So then we did the reverse. So now we take wild-type TIS, so full-length TIS11B, and we mutate the RNA-binding domain. And then also, we now see here that now the mutant TIS11B forms sphere-like granules. So now there’s two possibilities. Either it’s the RNAs that are recruited or that are bound by the RNA-binding domain, or it’s an intact zinc finger that is required for the mesh-like morphology. So to address this, we now swapped the RNA-binding domain, and we used the HuR RNA-binding domain. So both HuR and TIS11B bind to AU-rich elements, so they basically have a strong overlap in RNA targets. And so, now you see if we use the RNA-binding domain of HuR, we also get a mesh-like condensates. So this strongly suggested that it’s the RNAs that are bound by either HuR or by TIS11B that are responsible for this mesh-like morphology.

Christine Mayr (35:15):
And now we wanted to know, what about these RNAs… So what are the determinants that lead to the mesh-like morphology? And we turned to an in vitro approach. And we in vitro purified FUS-TIS, and you can see it forms condensates, so it undergoes phase separation. And you see that they are highly liquid, and so they can fuse.

Christine Mayr (35:42):
Okay. Fine. So then, the interesting experiment was… So now we add to FUS-TIS different RNAs, and we wanted to know what happens. Okay, so on the top, it’s just FUS-TIS. This was after overnight incubation. It forms sphere-like condensates. However, when we add in vitro-transcribed 3’UTRs, and so we added CD47 or PD-L1 or FUS, then in those two cases, we get a really nice network. But in this case, we only got spheres. And so those three were the first three that we tried. And at the time we thought, okay, so CD47 and PD-L1 are much longer in length, so maybe it’s the length that is important. So then, my post doc, Weirui Ma, from that moment on, he always did size-matched pairs, and so… But you see the size is not the important determinant, because we have 3’UTRs that are around a thousand nucleotides that can form mesh, and others cannot. Of course, we wanted to know what’s the difference between them.

Christine Mayr (36:59):
But first I want to show you some FRAP, and you see that these condensates are not aggregates because the protein components are actually highly mobile. So, after two hours, and after sixteen hours, the protein is still liquid-like, so it basically can move in and out of the condensate quite fast. Okay.

Christine Mayr (37:24):
So then, so Weirui, in the very end, so he did this experiment at two times points for 47 different mRNAs of different sizes and different concentrations. And bottom line is, network formation is an intrinsic property of the RNA. So there are RNAs that always form a network, and there are RNAs that never form a network. And, so what’s the difference?

Christine Mayr (37:49):
And so basically… So, here’s the result that basically led us to the right direction. So when we folded these 3’UTRs using RNAfold, you see here two examples of RNAs that are not able to form networks, and here are two that form networks. So the biggest difference is the color. So on the right-hand side, you see a lot of red, whereas on the left-hand side, you see more… Oops. More yellow, blue, and green. So the color code is the base-pairing probability. So red means… So these RNAs have a strong tendency to form intramolecular interactions and form strong structures here. Whereas on this side, there is large regions that are basically not structured. And, when they form a structure, it’s actually quite weak. Of course, both groups of RNAs always have both regions. Also, these RNAs have unstructured elements. However, it’s about the relative abundance. So, on the right-hand side, there is more red than on the left-hand side.

Christine Mayr (38:58):
And then we wondered, can we basically assign numerical values to the colors? And we noticed that there is a parameter called ensemble diversity that correlates really well with network formation. So ensemble diversity is the number of potential RNA structures that are predicted for a given mRNA. So if you have an mRNA that is highly structured, it cannot take on many different conformations, because it always wants to find that conformation where there is most base pairing. However, RNAs that are mostly unstructured can basically take on many different conformations. And so, this parameter correlates with length of the RNA, and therefore we are using a normalized ensemble diversity value. And you see it correlates really well with network formation R0.

Christine Mayr (39:54):
And therefore, we wondered, can we use the NED value to predict if an RNA will form a network or not? And so Weirui tested another 24 new RNAs, and he chose them only based on their NED values. And you see that nearly 80% of them are predicted correctly, just by using this single parameter. And so that means that mRNAs that have large unstructured regions basically are able to form mesh-like condensates.

Christine Mayr (40:26):
But we still wanted to know how this all works. Okay. And so we looked again at these structures. And so then we thought, okay, so the red RNAs basically seem to form really strong intramolecular RNA-RNA interactions. And therefore, we hypothesized that the unstructured RNAs might predominantly form intermolecular RNA-RNA interactions, and we wanted to test this.

Christine Mayr (40:54):
So, what we took is, we took two structured RNAs that will not form networks, but they can actually interact here. And then we added, at the 5′ and 3′ ends, RNA dimerization elements. So now we can create, in vitro, a higher-order RNA network. And we chose these RNA dimerization elements in a way that we can actually really form a really large network.

Christine Mayr (41:25):
And you can see this here on the gel. So these are the two RNAs that we used. This is their size. When we add them together, they form a dimer. However, if we then add the dimerization elements… You see, here is the dimer. But then they actually form a much higher-order network. And when we add this to FUS-TIS, you see, if we add the RNA dimerization elements… So when there is extensive intermolecular RNA-RNA interactions, we obtain a network. Whereas, when these RNAs can bind to each other but cannot form a large interaction network, they basically form spheres. So this shows you clearly, when there is a multivalent intermolecular RNA-RNA interaction network, this is basically the basis for this network condensate.

Christine Mayr (42:22):
But still, we… So, I mean, this was all beautiful. But we still didn’t really understand how the network is formed. And to look at this, we turned to 3D time-lapse imaging. So we really wanted to image fusion events that occur very early. So on the top, I’m showing you FUS-TIS where we added an RNA that is not able to form a network. And you can see when there are two condensates that now basically touch, and they fuse, and in the end, they form a round, a sphere-like condensate.

Christine Mayr (43:03):
In contrast, when we use a network-forming RNA like the CD47 3’UTR… So now, these two, they touch, and they connect at the contact site. But basically the morphology of the condensate does not change. Now, of course, most people, when they see this first, they think this is because it’s no longer liquid-like. They think this is gel-like or something. But when we do the FRAP, this is all… So the protein component can really move. It’s highly mobile. So basically the protein can still exchange with the surroundings, but it still makes not a sphere. And so basically, at a later time point, so you can see how these two molecules, they touch, but they don’t, basically, fully fuse… I mean, they touch and are connected at the contact site, but they don’t fully fuse.

Christine Mayr (44:08):
So maybe this becomes clearer in our model. So how we look at this is… So when we have structured RNAs, they can be basically viewed as globules. And so now we have two condensates that fuse with each other, and so the surface tension promotes a high fusion force. And so, this leads to mixing of the components, and because these structured RNAs are relatively free to move, so now they can basically take up the new space.

Christine Mayr (44:43):
In contrast, when we have disordered RNAs, they have some globular parts, but they also have these disordered regions that can then do intermolecular interactions and form an RNA skeleton, or what we call an RNA matrix. So now, when these two condensates touch, they can fuse at the contact sites. And the protein components shown in magenta, they can basically freely move. But the RNA skeleton is quite rigid, and therefore it can basically not change its shape dramatically. And so therefore, it forms these large filamentous condensates. And so, so you see here, these condensates, they line up as beads on a string, and then there is some remodeling to really then form a more filamentous condensate. But again, the protein components are highly mobile.

Christine Mayr (45:44):
So in summary, what I showed you in the second part of my talk, we found a role for unstructured regions of mRNAs because they have a high capacity to form intermolecular RNA-RNA interactions. We found that 3’UTRs can actually fulfill structural roles in the cytoplasm and can act as skeletons for mesh-like condensates. So, what is the advantage of making mesh-like condensates? So basically, if you have a sphere, a sphere has a geometry that minimizes the surface area. And we think that, in mesh-like condensates, that’s not the case. So now, the mesh-like condensates actually have a large surface area. And, in the case of TIS granules and the ER, this might be really… This might give them an advantage because now they can share a lot of interface. And, more generally, this could be also true at the interface with other organelles. So you basically can create mesh-like condensates that have a liquid-like character, where basically two interfaces come together. And this might promote reactions that occur on the surface of the condensate.

Christine Mayr (47:02):
And so, we found that these organelle networks are driven by the interplay of two opposing forces. So one is a fusion force that is created by the surface tension. But the anti-fusion force is generated by the RNA skeleton that doesn’t allow full fusion. And we are actually quite excited about this, because this is very similar for the formation of membrane-bound networks. So when people reconstituted the ER in vitro, they only need two proteins and lipids. And so basically, the two proteins… One of them promotes fusion, whereas the other one inhibits fusion. So basically, it’s the same principle, how you generate membrane-enclosed networks or membrane-less networks.

Christine Mayr (47:58):
And with this, I will… I want to just have my last slide on what we think are the functions of TIS granules. So we think that they play a really important role in protein maturation. So basically, protein maturation is the step after protein synthesis, between protein synthesis and obtaining of a fully-functional protein because… So you synthesize a protein. Then it needs to fold. Then it needs to add post-translation modifications. It needs to find the protein interaction partners. And most people assume that this all happens, you know, by chance, on its own, or whatever. But we think that many of these processes are actually regulated. And we think that TIS granules have a strong influence on these processes. And this is basically what we are investigating right now.

Christine Mayr (48:55):
And with this, I want to thank the people who did the work. So, really nearly everything I told you today was done by a highly-talented postdoc, Weirui Ma. And so he had some computational help from Gang Zhen. And this is Ellen Horste. So she basically did the granular purification. But I also want to thank the rest of the people in my lab because they also worked on group projects. And I want to thank you for listening.

Mark (49:25):
Beautiful work. For me, as a protein person, really mind-bending to think about how much structural-

Christine Mayr (49:33):
Uh-huh (affirmative).

Mark (49:33):
-control these RNAs have. Really wonderful.

Christine Mayr (49:41):
Thanks.

Mark (49:42):
Yeah. I have questions, but I’m sure some other folks do, so I’ll just ask the group. Any questions for Christine? You’re all muted. Diana?

Diana Mitrea (49:58):
Hi. Yeah, that was beautiful work. Thank you. So it was really fascinating when you were showing the beads on the string, that they all seemed to be very homogeneous in their size, those, the beads. And I was wondering if you can comment a little bit on… So, I mean, you talked about how the RNA forms the skeleton and how there’s different proteins with effective… promoting fusion. And so…

Christine Mayr (50:30):
Yeah. Actually, yeah. So depending on what UTR you add, you can actually determine the size. So there are some RNAs that lead to droplets that have a smaller size than others. But because all I’m… So everything I showed here was always when we add one RNA. We also added two RNAs and stuff like that. But so, so therefore, the RNA determines also the size. And therefore, they always have the same size. But if you basically mixed two RNAs, they also have the same size. I mean, the only experiment we did is… because we mixed them up front and then we incubated them with the protein. So therefore, it’s again, it has the same feature. But the size is really important. If you actually make a very large droplet… I mean, so, in our prediction, only 80% were predicted right. And so there are some other features of the RNAs that can influence other parameters that actually influence the whole process. So, of course, I presented it in a very… straightforward, but it’s a little bit more complicated. But we don’t understand it.

Diana (51:45):
Yeah, I mean, I’m still trying to wrap my head around how the restriction in size is, in general, for condensates and cells. and why certain condensates need to be kept really small and others are really large, and how that connects to function.

Avinash Patel (52:11):
Christine, I have a question. I’m Avinash here from Dresden, Germany. It was a beautiful talk. So, I missed the first part of the talk, so I’m sorry if you have mentioned that already. So you mentioned, during the end, that TIS granules do also the protein maturation and stuff. So are you looking at other protein components in the granule that you envision might also help in this entire process, like other kinases, other chaperones, something like that, that forms in those granules?

Christine Mayr (52:41):
Yeah. So basically, in the paper we showed that there are proteins that are enriched, like some chaperones are enriched. So we started by looking at what are the RNAs that are translated, to get a better idea of what’s going on.

Christine Mayr (52:58):
But we have another project where we look for rules. Of course, we can do mass spec, and we can see what are the proteins that are actually enriched in the granule. But we are a little bit more interested in learning the rules, why some proteins are enriched whereas others are excluded. And so we have set up experiments to get at that question. And because the idea is that, I mean, the RNAs that are in, plus the proteins that are in, will determine the environment. And then, the environment will basically determine what reactions are promoted and what reactions are inhibited. And so, we are basically going step by step.

Avinash (53:40):
Great. Thank you.

Mark (53:46):
So I’m still trying to understand. What are the sequence requirements of an RNA to enable it to form this more structured state that actually can overcome surface tension?

Christine Mayr (54:03):
Yeah. So it’s actually not sequence, because… So, I mean, so if you do… if you compare RNAs that have the same size, there is no correlation with AU content. So, really, it’s exactly the same. However, if you look genome-wide, there is a correlation with AU content because AU makes less-structured RNAs. And because, if you then have longer RNAs, they have a higher propensity to do it. So basically, nearly every RNA that we looked at over three kb basically forms a network.

Christine Mayr (54:40):
But so, we think it’s basically… It’s the composition. So you have regions that are actually structured, and then you have a region that is not structured. So you need a structured region to allow that this region is actually not structured. And then, basically, if you have two RNAs that come together, now they find base pairing in this unstructured region. And that, basically, then leads to these intermolecular interactions. And I think it’s a concentration issue. So basically, in the granule, there’s a higher concentration, and, I mean, Simon Alberti showed, in P bodies or in stress granules, that this prevents entanglement.

Christine Mayr (55:23):
And we think that, maybe, there is something special about the environment that allows RNAs to interact, but maybe not to do this in an irreversible manner. So somehow it’s still dynamic. And basically, nearly all the images that I showed are taken after sixteen hours, because the network looks more beautiful. But, I mean, the network looks very similar after two hours. It’s already all there. But there is remodeling. So, it’s not totally static. It actually… There is still stuff going on after sixteen hours, and it’s actually… So we think that within these networks, the RNAs still basically can find better interactions or so. But we don’t have the resolution to really look.

Mark (56:14):
And it’s still reversible.

Christine Mayr (56:16):
Yeah.

Mark (56:18):
Yeah. That’s impressive. That’s really nice. A real eye-opener. Other questions? No?

Christine Mayr (56:30):
So basically, Steven Boeynaems in his paper published in PNAS, maybe two years ago, he did this on homopolymers, where he had long stretches of As and Us. And if he puts those together, then he actually also gets these. But in his case, he didn’t have dynamic condensates, but they were actually aggregates. And I think that there is too much base pairing, so it was basically no longer dynamic. But we are using these physiological RNAs, and we don’t know, I mean, we don’t know why it’s different. But there’s obviously still stuff going on. And it must be widespread and weak interactions that are not permanently crosslinked, because otherwise, if it’s permanently crosslinked, I would say you’re actually getting a non-dynamic condensate. So there is… It’s probably a whole spectrum.

Mark (57:32):
Yup. And it’s also interesting to think about whether similar sorts of structures are occurring adjacent to other organelles.

Christine Mayr (57:44):
Right. Yeah. Of course. So, on the surface of the mitochondria or with the cytoskeleton, right.

Mark (57:53):
Exactly.

Christine Mayr (57:54):
Yeah, so yeah. We actually found a second network made by an RNA-binding protein that actually makes an even larger network in the cell. And we are also studying that.

Mark (58:06):
Mm-hmm (affirmative). Beautiful. Wonderful. Other questions? So if not, Christine, I’d just like to thank you again. Just beautiful work and wonderful to hear about it. And thank you for doing this for the whole community. I’m sure that many people will see this and enjoy watching it.

Christine Mayr (58:28):
Okay. Yeah. Thank you for having me.

Mark (58:31):
Of course. And often what happens is people, afterwards, when they’ve had a chance to think about a talk, they’ll come back later with additional questions. So that may happen here, as well.

Christine Mayr (58:41):
Okay. Sounds good.

Mark (58:43):
Right. So, wonderful. Thanks again, and appreciate your time. And stay safe.

Christine Mayr (58:47):
Okay. Okay. Thanks.

Mark (58:48):
Okay. Thank you, everybody.

Christine Mayr (58:51):
Bye.

Mark (58:54):
Bye-bye.

Avinash (58:54):
Thank you.

Mark (58:54):
Bye-bye.

Diana (58:54):
Thank you.