|Type||Kitchen Table Talk|
On June 23, UC Merced’s Shahar Sukenik treated Dewpoint and Condensates.com to a fascinating Kitchen Table Talk on his lab’s work. Shahar has been studying the behaviors of proteins in cells since his PhD with Daniel Harries at the Hebrew University of Jerusalem and his postdoc with Martin Gruebele at Urbana-Champaign. In 2018 he opened his own lab at UC Merced to study the interaction between proteins and their surrounding environment both in vitro and in cells, with a special emphasis on the breakdown of homeostasis, which is relevant for many pathological conditions including cancer, viral infection, and other metabolically linked diseases. In this research, his lab deploys a wide range of methods such as live cell imaging, spectroscopy, and computational modeling.
In his talk, Shahar discussed how his lab identifies the hidden structures of IDRs and how these hidden structures alter the ability of proteins to form specific, functional, and stable condensates. We were all brimming with questions afterward, and Shahar kindly provided written answers to the ones he didn’t have time to answer during the talk. You can find those here, and the recording of his talk below.
Create an Account or Sign In to view the video.
Mark Murcko (00:00:00):
Hi everybody. Good to see you all again. Yeah, it’s interesting because we’ve heard so many great comments about the whole lecture series, this whole idea of the Kitchen Table Talks for the whole community. We’re getting a lot of great feedback from people around the world about how they really enjoy these lectures, find them fun and educational and so we’re really glad to hear that. That’s why we’re doing this, we’re doing this because we think that the whole condensate community can benefit. We’re all wrestling with the same scientific challenges and so it’s good to be thinking about these things together with everyone.
Mark Murcko (00:00:34):
So today we have Shahar Sukenik. He’s from University of California at Merced. I’m really excited about having Shahar here. Got his PhD at the Hebrew University of Jerusalem, and then did a postdoc at Urbana-Champaign, really focusing on microscopy, really thinking carefully about the interactions of proteins in live cells. And obviously over the years has gotten attracted by the condensate bug, like all the rest of us. And three years ago took his faculty position at Merced where he is now an assistant professor.
Mark Murcko (00:01:19):
And really interesting work because it highlights the interactions between proteins and their environments, both in vitro and in vivo in a cellular context. And of course, Shahar does this with an amazing range of technologies, live cell imaging and spectroscopy, computational modeling, other things and so really applying a multi-disciplinary approach. And that’s a common theme, isn’t it? A lot of the condensate lectures we’ve heard really rely on such an interesting blend of technology to tackle these challenging problems. And so in particular, he’s been focused on homeostasis and we all know the critical nature of that and, and how the metabolic rewiring affects proteins in a variety of ways and how relevant that is for cancer and viral infection and lots of other pathology.
Mark Murcko (00:02:22):
And of course that leads to IDRs, which are such an essential part of this whole process. And so naturally a lot of the work that Shahar does is related to understanding the way the IDRs and the other aspects of proteins lead them to form these very specific and functional and stable condensates. And so, very exciting to have you here today and your title this morning is, Turning Protein Disorder into Function: Is the Solution the Solution? Floor is yours.
Shahar Sukenik (00:02:54):
Thanks so much, Mark. So first I’d just like to start by thanking Jill for this invite and Mark for that fantastic intro, which basically turns my intro obsolete, but I’m going to try doing it anyway. So it’s really great to be here this morning. And so I think we’ll just jump right into it. And the place I’d like to start with is a slight apology in that while there will be some discussion of condensates most of this talk, we’re going to zoom in and really focus not on the collective behavior of proteins, but rather on the behavior of the monomer. And like Mark said before, to understand the collective behavior of these molecules, it’s really important to first form a good understanding of how they behave when they’re on their own. So that’s kind of our entry point…
Shahar Sukenik (00:04:00):
And so the first thing I want us to start thinking about is the structure of proteins. And so usually when we go to a textbook and now it seems like, right, this animation is working? Because it didn’t before, right. So when we look at a textbook and we think about these traditional well folded proteins, they all have a very specific shape. And these shapes, or structures, inform on the function of these molecules. And an absolutely beautiful example is this pilus assembly mechanism that’s been resolved using cryo-electron tomography by the Jensen Lab way back in 2016. But you can really see how this is a molecular machine, where you see stuff that looks like cogs and parts; and they’re moving in synchrony and they’re acting together to form this pilus, which is this harpoon that bacteria use to move around.
Shahar Sukenik (00:05:01):
And so this idea that structure turns into function is heavily relied in our own understanding, right? That’s what we know from our day to day life. And that’s usually how a lot of the proteins in our cells work. So it doesn’t have to be these massive assemblies. It can also be smaller complexes, like polymerases. It can be metabolic enzymes or other proteins. But all of these have this specific structure that essentially facilitates their function. So this is good. And the reason why this is good is because the cellular environment, the environment these proteins actually work in, it doesn’t look like this blank space, it’s actually completely jam-packed with other molecules. And this environment is constantly changing.
Shahar Sukenik (00:05:54):
So over here we see some images of cells doing their thing, splitting, moving, blebbing, doing all kinds of things that cells do. And it’s clear, even though we don’t have a single molecule resolution in these images, it’s very clear that the environment undergoes dynamic changes over short timescales. And so if we think about the physical chemistry of what’s happening inside the cell, we can think about changes to crowding, ionic strength, osmotic pressure, pH, viscosity, and even the composition of the cell, the concentration of the solutes that exist inside the cell.
Shahar Sukenik (00:06:31):
And so those folded proteins, the proteins that have this specific structure have a strong hydrophobic core held together by thousands of intramolecular bonds, have a surface area that’s optimized for the surroundings and so isn’t really perturbed strongly. And so these folded proteins may not care. This makes sense because decades of biochemistry done outside the cell taught us everything we know about these enzymes and the buffer environments where these experiments were done is very distinct from the cell. But not all proteins are like this.
Shahar Sukenik (00:07:07):
And so intrinsically disorded proteins, which I abbreviate here as IDPs, but also IDRs, which stands for intrinsically disordered regions, are outliers in this respect. Even though they’re outliers, they compose a large chunk of the proteome. So it’s estimated that about 40% of the human proteome is composed of these intrinsically disordered regions. They play a really critical roles in cell regulation, in homeostasis, in signaling and of course in liquid-liquid phase separation.
Shahar Sukenik (00:07:38):
And so these proteins or these proteins regions exist in this rapidly interchanging ensemble, where the ensemble is kind of the collection of conformations that these sequences can take. And these ensembles, especially compared to well folded proteins have very few intermolecular bonds and have a high degree of surface area as you can see from this trajectory here on the right-hand side. And so it’s this confluence of few intermolecular bonds and high degree of surface area that really makes these sequences poised to sense and respond to their surroundings.
Shahar Sukenik (00:08:20):
And so this is really our entry point. This is really the question that we’re interested in. It is, how does the cellular environment interact with these disordered ensembles? And so when we think about this, since we’re, or I am at least trained as a chemist, we start thinking about the physical chemistry of this. And when you think about that, you realize that these disordered regions sense the environment through their surface area, and then can respond to their environment by changes in their ensemble. So how does this take place? Well, first of all, we can about the global dimensions of the ensemble. So these hairballs that you see here are essentially, a few hundred conformations of disordered protein that they’re overlaid one on top of the other, and depending on the environment, the protein is in this can cause this ensemble to expand or to contract.
Shahar Sukenik (00:09:18):
And so when we think about the physical chemical model or mechanism behind this, you can think about the interaction of the solution with the protein backbone or the volume taken up by the protein. And this has been characterized for many years by people like Record and Timasheff and Bolen and Minton through the action of osmolytes, which act relatively uniformly on the peptide backbone by repelling the peptide backbone through the action of denaturants, which act relatively uniformly on proteins by attracting to the peptide backbone, exposing it–that’s why they denature it. Or it could be action of crowding where the protein tends to take up smaller volume when in a crowded environment. So that’s one aspect, these global dimensions of the chain.
Shahar Sukenik (00:10:12):
And the other aspect relates to the local dimension of the chain. When I talk about local dimensions, I mean specific interactions within the chain that can be modulated by a change in solution. And so some classic examples of this would be electrostatic interactions that can be screened away by adding salt to the solutions, could be attractions or repulsions. It could be hydrophobic patches or aromatics that exist within the sequence and can bring together different regions of the sequence depending on what is the solution around them.
Shahar Sukenik (00:10:44):
And so this is the theoretical framework that we have to understand how ensembles might change. And the question we had starting out is, how the different IDP sequences respond to similar changes in solution? And the idea we had is that we expect that there’s going to be some kind of sequence dependence here, because these kind of specific interactions vary with the identity of the protein.
Shahar Sukenik (00:11:12):
And so we started out by just looking at simulations. And so we used the Campari simulation suite with the ABSINTH force field that was developed by Rohit Pappu. And together with Alex, with Alex Holehouse, we’ve developed a method by which we can essentially tune the interactions of the solution around the peptide. And so over here, what you can see is how the global dimensions change as we turn the interaction between the solution and the peptide backbone into something that’s more repulsive. So the more repulsive the solution gets the smaller our ensemble becomes because the protein wants to compact it on itself to reduce the surface area, reduce the amount of interaction with this repulsive solution.
Shahar Sukenik (00:12:04):
When we started simulating more and more sequences here, so this is work done by a graduate student, Feng Yu, in the lab. So what Feng did is, he simulated about a hundred different intrinsically disordered regions, and we normalize them so that all the radius of gyration start from the same point. And then we just turn the solution into, or tune the solution more and more repulsive.
Shahar Sukenik (00:12:30):
And what’s completely clear is that there’s a wide range of behaviors that are sequence dependent. And so we have some kind of average behavior, where the sequence undergoes some intermediate collapse, as the solution becomes more repulsive. But then there’s going to be other sequences that show almost no sensitivity, and then other sequences that are exceptionally sensitive. And so we found this really intriguing. This implies that there’s some kind of sensitivity that’s encoded into the sequence, at least as far as this global dimension goes. Of course we also wanted to look at how these local interactions or local structure plays a part. And so for that, Feng has developed an algorithm to help us visualize the contacts within an intrinsically disordered chain.
Shahar Sukenik (00:13:21):
So over here, what we see is an intrinsically disordered protein sequence and the red lines imply repulsive interactions, the green lines imply attractive interactions. We’re not going to go into how we calculate this. If anyone’s interested, I’m happy to talk more about this later. But then when we take this sequence, which over here is simulated in water and put it into different solutions, we’ll see that these patterns change. And so when we have a solution that’s attractive to the peptide backbone, like a urea solution, what we see is almost all of the attractive interactions go away. When we put this in a solution that’s attractive to hydrophobic patches that becomes even more pronounced, but if we put this in a solution that’s attractive to polar patches, then you can see that a lot of these contacts actually persist. And so even though all of these solutions have the same global interaction, you can see that very different structures may arise when the specific interactions are different.
Shahar Sukenik (00:14:32):
And so this doesn’t just happen in this sequence. We can see that these long range interactions are very prevalent. And here’s just a collection of a few of the disordered proteins that we’ve simulated. Each one of these boxes has a different sequence, and you can see that the attractions and repulsions in each of these sequences are very different and also often span the entire range of the sequence itself.
Shahar Sukenik (00:14:54):
So this is all great, but like any computational person worth their salt, the question always exists, is this real, right? This actually happened in the real world, or is this just an artifact that’s a result of whatever idiosyncrasies exists in the force field that we’re using. And so to answer this question, of course, we needed to take this concept and build an experimental system that will test it specifically.
Shahar Sukenik (00:15:22):
And so the way we chose to do this is using ensemble FRET. And so before I describe the system, I’m going to pause here and just say that ensemble FRET has been for decades gotten a really bad rep. And I’m really hoping to bring it a little bit back into vogue. Because single molecule FRET, time resolved FRET, these are fantastic methods but these are methods that are relatively expensive to implement. They can be extremely low throughput and what we lose in ensemble FRET in quantitative accuracy, we gain many fold in throughput. And we also have the added advantage in that we don’t need really fancy fluophores to do these ensembles, and so this allows us to use fluorescent proteins. And these are fluorescent proteins, it’s just such a huge advantage. There’s no extra labeling steps, it’s amenable to just expressions in the cell. I mean, we’ve really been capitalizing on the ability to use these.
Shahar Sukenik (00:16:30):
All right. So after that, the construct itself contains two fluorescent protein, a donor which is mTurquoise2 and an acceptor, which is mNeongreen and then between these we ligate disordered protein sequence. And the idea is that when we excite the donor here, we’re going to measure a spectrum. And we do these measurements in well plates so it’s relatively high throughput. We can do this in 384-well plates and just get many of these experiments done quickly.
Shahar Sukenik (00:17:05):
And so what we get is generally two peaks, one belongs to the donor or the other to the acceptor. We can get rid of artifacts by looking at the donor alone and the acceptor alone. And over here, the shaded areas show the fluorescence of each of these fluorescent proteins on their own. And then we can deconvolute the measured FRET spectra using these two base spectra and get at the FRET efficiency in a specific disordered protein sequence. And so if the solution is repulsive and contracts the sequence, for example, what we’re going to see is that the acceptor fluorescence increases and vice versa if the solution expands the sequence, so the acceptor emission is going to decrease. And so what we’re going to be measuring is this either E_FRET or FRET efficiency, which is essentially the fraction of photons emitted from the acceptor over all the photons emitted total.
Shahar Sukenik (00:18:18):
Okay. So this is great, but it’s not without problems. And one of the main problems with any FRET measurement really is that they’re really sensitive to the length of the sequence that we’re interested in, really sensitive to the distance between the two proteins. And so our solution for this is to calibrate the FRET based on a sequence that we can characterize very well. The sequence we picked was these glycine-serine repeats. The reason why we chose these is that there’s evidence, especially coming from Magnus Kjærgaard’s lab, showing that these glycine-serine repeat polymers behave like theta state homopolymers. So these are polymers whose behavior has been well-characterized experimentally previously, and we kind of understand how they work.
Shahar Sukenik (00:19:15):
And so when we go ahead and measure the FRET efficiency of different lengths of glycine-serine repeats, we see this really nice linear dependence. For the aficionados of polymer physics, this is not what’s expected from just a plain homopolymer, but remember that this is not a plain homopolymer. Remember we have our two fluorescent proteins on either side. And because of that, we get this linear behavior. And what this linear behavior lets us do, is it lets us interpolate what the FRET efficiency of a GS linker would be for a sequence of any given length, assuming we have that length.
Shahar Sukenik (00:19:57):
And so we can now essentially predict what the FRET efficiency of a GS linker would be for any length here and use that as a comparison. And so when we do that and we replace our GS linker with another disordered protein, we can always say, is this linker now longer than a glycine-serine repeat or is it shorter than a glycine-serine repeat? At least as far as the end to end distance measured by the FRET labels can tell us. And so that’s going to be our point of reference.
Shahar Sukenik (00:20:31):
So we go ahead and we started doing these experiments. And so over here this is the basic experiment that we started doing. So on the X axis, we have the concentration of a solute, in this case it’s sarcosine, which is actually a pretty prevalent cosolute inside cells, and it’s a precursor of several amino acids. And then on the Y axis, we have this normalized FRET efficiency. So this would be essentially the FRET efficiency divided by the FRET efficiency of a GS linker of the appropriate size minus one. So when it’s zero, it’s exactly the same dimensions as a GS linker of the same size. When it’s positive, it means the sequence is more expanded. Negative means it’s more compact. This over here the FRET efficiency of the 32 GS repeats of course starts from zero because it’s exactly the same dimensions as the GS 32. But as we start adding sarcosine into this mixture, we see that the dimensions start changing. And so we were really encouraged by this. And of course here is where we can really capitalize on the throughput of the system. So we don’t have to do this for just one sequence or one solute, we can do this for lots of sequences and lots of solutes.
Shahar Sukenik (00:21:54):
So we call these graphs solution space scans. As you go across columns, you see that we change the identity of the solute that we’re adding into the solution. And then as we go across the rows, we see that in this case, we’re just taking longer and longer glycine-serine repeats.
Shahar Sukenik (00:22:16):
And so you can see that in solution space, different solids are going to have very different properties. And some of these can kind of be predicted. So polyethylene glycol and ficoll, which are generally thought of as polymeric crowders, tend to compact the sequence. Urea and guanidinium, which denature proteins, generally expand the sequence. And then we see some more interesting things with others, especially notable is the fact that for example, salts and these sequences have this non-monotonic effect. But overall you can see that regardless of their length, the GS linker is behaving the same way.
Shahar Sukenik (00:23:01):
So the response to the solution space scan for glycine-serine linker is the same regardless of length. And we’re really happy to see this. It made sense because whatever local structures you can think of exist in the sequence, they’re going to be kind of the same, regardless of the length of the sequence. By the way, I forgot to mention the color coding in the background is simply a proportional to the slope. So if the slope is really strong, then it’s going to be a stronger color and vice versa.
Shahar Sukenik (00:23:32):
Okay. So this is great, but of course we want to move beyond these glycine-serine repeats into real disordered proteins. And so the question is, does this happen in real IDPs? And so we took a few classic examples, and we pick these examples specifically because in each one of these, in almost all of these, the structure of the ensemble has been shown to be linked to the function of the protein. And so when we took these four different sequences and we did the same solution space scan, we saw something very different. So of course, each of them still responds differently to different solutes but now as we go across the rows, we see that the intensity of color is very different. And while some sequences are, for example, strongly compacted by PEG, others are barely affected at all. And so the response to chemical environment, in this case, is sequence dependent. And so we immediately thought, maybe this could be this evidence of hidden structure that we saw in our simulation.
Shahar Sukenik (00:24:43):
So this is great, but we were still a little bit skeptical. And part of the reason why we’re skeptical is the reason that I started out with, and that is that in ensemble FRET is problematic. So even if we get rid of all the artifacts in ensemble FRET, all the bleed-throughs and the cross-excitation, and we managed to really calculate the FRET efficiency accurately, which is difficult, there’s also some model dependence. And this model dependent specifically comes when we want to move from FRET efficiency into end to end distance, we need to use this Ro factor. And the Ro is generally treated as a constant, but there’s a lot of different factors that fold into it. But it actually depends on many parameters, and saying that it can just be held as a constant, especially in different solutions is a strong statement.
Shahar Sukenik (00:25:38):
So another method that we can use to talk about that the dimensions of a sequence is small angle X-ray scattering or SAXS. And the big advantage of SAXS even though it’s much lower throughput than our FRET experiments is that interpreting SAXS measurements in terms of a characteristic radius and specifically the radius of gyration of an ensemble is really model free. So we can take a scattering curves over here on the Y axis, we have the logarithm of the scattering intensity as function of the reciprocal angle of scattering. And when we take this and we fit this to a line over a certain region, it’s called the Guinier region, we get out as a slope, the Rg. So there’s absolutely no model fitting used here. And so we’re really excited about the possibility of treating these Rg measurements as the ground truth. As something we know that works, assuming our measurements are sufficiently high quality.
Shahar Sukenik (00:26:40):
In the background here is a long-term, I guess, controversy dealing with the agreement between FRET measurements and SAXS measurements in intrinsically disordered proteins. Again, there’s a lot to talk about there. I’m not going to go into it. I’m just going to say that at least in our constructs, this has been resolved and I’ll show you exactly how that happens now.
Shahar Sukenik (00:27:07):
So we start with independent measures of Re through FRET and Rg through SAXS in our glycine-serine repeats. This is what it looks like. Okay. So the blue symbols here are the end-to-end distance, the orange ones are the radius of gyration. And together with Alex Holehouse what we did was we essentially said, can we make an all-atom simulation that’s going to simultaneously match both of these observables. Because if we can, then that’s at least one possibility of an ensemble that can exist for these measurements.
Shahar Sukenik (00:27:49):
And the answer is yes. So we managed to do it. So Alex, using an all-atom homopolymer simulation has managed to tune the parameters in such a way that the simulation Rg and Re match almost quantitatively with our experiments. And this was really exciting because the way these simulations worked, it turned out that the behavior of these simulations is exactly how we expect glycine-serine repeats to behave. So this was really encouraging. And the next question was, can we match this also in actual IDPs? These are just homopolymers. So what happens if we now take a heteropolymer, change the sequence, can we still get this interesting relationship?
Shahar Sukenik (00:28:40):
So what we do next is just plot the end to end distance versus the radius of gyration. And you can see that the glycine-serine repeats just fall in this nice diagonal here. And what we do next is we measure a couple of heteropolymers. So we measure p53, NTAD, and the BH3 domain of PUMA, which we’ve seen before. And when we put those on the graph, well, not really. So not only are they not on the diagonal, which okay, I guess that can be expected, they also have almost the same Rg, but very, very different end-to-end distances. And one is below the diagonal line of the homopolymers and one is above. This is not really what we expected to see, right? This implies that something here is very different from the glycine-serine repeats. So again, our go-to are these long range, hidden structures that alter the ratio that… sorry, that exists in these heteropolymers, but not in a homopolymer and alter the ratio between Re and Rg.
Shahar Sukenik (00:29:43):
And so we thought, how can we test this hypothesis? Well, one way to do that would be to take the sequence of one of these IDPs and scramble it up. So we took the PUMA BH3 domain, you see the sequence over here, we scrambled it up. Our visualization implied that the long range interactions differ dramatically between the different sequences. And what we did next is we repeat this correlated FRET and SAXS experiments. What we see is that we see the wildtype here, and then we see the three scrambles here. And even though the amino acid composition is exactly the same, the length is exactly the same, we get very, very different behaviors of Re and Rg.
Shahar Sukenik (00:30:34):
So not only that, but the scrambles also show very different solution sensitivities. So if we do our solution space scan, which is now a little bit more limited, but once again, we see that, for example, the wild type has a very high sensitivity to crowding and denaturants whereas some of the scrambles are essentially unaffected by anything.
Shahar Sukenik (00:31:01):
All right. So to conclude, I think, or I hope that I’ve managed to convince you that by combining FRET and SAXS, we are starting to reveal these hidden structures that exist in intrinsically disordered protein ensembles. I’ve shown you that these long-range interactions can be prevalent in these ensembles and often largely determine their shapes. I’ve shown you that these long-range interactions can respond to changes in the solution and also that the changes in the solution can in turn inform on what hidden structures are there. And so the bottom line is that at least in vitro, the physical chemical environment matters, at least as far as the conformational ensemble of the IDP is concerned. And so what we’re doing now is we’re adapting this methodology for functional studies. So we’re looking at how these changes in ensemble translate into changes in protein binding, and phase separation, and post-translational modifications, and so on.
Shahar Sukenik (00:32:09):
So this is all great, but here’s the place where we remind herself that this all happened in vitro. And I started out my discussion about how the cellular environment changes and it matters. And so really the question is, does this kind of effect happen inside the cell? And so here again, we capitalize on the fact that we’re using fluorescent proteins as FRET labels. And so we can simply transiently transfect these plasmids into mammalian cells and this is what we do. So generally we use U2OS cells though we’ve now used a few cell lines, the results are to a large extent, similar, even though there are notable differences sometimes between cell lines. And what we do is we look at the emission of the donor and the emission of the acceptor through filters on our microscope. And this, of course, also allows direct comparison with in vitro experiments because while we don’t have the full spectrum in our microscope, we can just take specific windows in our spectra and then calculate the donor and acceptor fluorescence in the same way we do in our microscope.
Shahar Sukenik (00:33:22):
Okay. So now let’s start looking at it. And I should point out that these experiments are actually much, much easier to perform than the in vitro experiments because there’s no purification or expression purification process necessary here.
Shahar Sukenik (00:33:36):
So the first thing I want to show you is that glycine-serine repeats inside the cell behave kind of like they do in vitro. So we’re now going to transition to a Y axis that shows the donor to acceptor ratio. So as this increases, that means the donor is further away from the acceptor. And you can see a linear scaling of D/A, as we increase the number of residues in our glycine-serine repeats. This is what the same thing looks like in vitro. You noticed there’s a change in scales. This is expected simply because the sensitivities of the different imaging devices varies.
Shahar Sukenik (00:34:17):
We can also start looking at IDPs. So for example, we repeated our PUMA and PUMA scrambles experiment inside the cell. So again, you look at D/A here, you see the wild type, S1, S2, S3 and here’s the comparison to the same thing in vitro. And you can see that the trends persist. So wild type is the most compact here, S1 is the most expanded and so on. And this is by the way, the average of over 10,000 cells in each of these. So these hidden structures that exist or that we think we’re measuring in vitro also persist inside the cell. Of course, we’re not limited just to this. We can now look at a whole range of different proteins.
Shahar Sukenik (00:35:02):
And so we tried different naturally occurring IDPs inside the cell. You’re going to see them like this over here. This is kind of a violin plot. The little points, the little circles in the middle are individual cells, we’re taking wholesale averages. The squares is the average, and you get the standard deviation here and the distribution denoted by the violin. And then again the D/A, the FRET signal is normalized to the GS linker of the equivalent length, so that when it’s zero, it’s going to be more… sorry, it’s the same as a GS linker, when it’s negative, it’s more compact when it’s positive, it’s more expanded. And so this is the picture we see for a few different disordered proteins that we tried out.
Shahar Sukenik (00:35:48):
Over here on the left-hand panel is just the length of that protein. So it’s arranged by length, the shortest one at the bottom, the longest one at the top. And you can see that it’s very much length independent. So some of these sequences just in cells that are in normal conditions behave as the GS homopolymer equivalent would, but others show more compaction. We actually have not yet observed the case with more expansion and that’s interesting. And again, we have thoughts on why that happens. But what it means is that these different sequences are going to have in certain cases, some hidden structure that would cause them to compact more than a homopolymer otherwise would.
Shahar Sukenik (00:36:38):
Okay. So this is a good start, but what we really want to do is do our solution space scan? We want to kind of kick the cellular environment around a little bit and see what these sequences do in response. And so how can we actually do that? The cellular environment is very complex, like we already said, but it’s also adaptive. So when we do changes to the cellular environment, the cell is going to try to counteract them. And we’re really interested in, again, the protein behavior. So kind of devoid of the context of the regulation machinery of the homeostasis machinery of the cell. And how can we decouple the physical response of the cell… sorry, the physical response of the proteins from the biological response of the cell itself?
Shahar Sukenik (00:37:22):
So the best way to do this is the realization that the timescales are drastically different. The physical-chemical adaptation of a protein, the change in ensemble structure happens in milliseconds or less. Especially in disordered proteins, the conformations are sampled extremely rapidly. Whereas biological adaptation, for example, phosphorylating the protein, or causing some cascade that would counteract the stress or the perturbation that we’re doing to the cell, that takes something on the order of minutes, 20 minutes or more, especially if we’re thinking about transcriptional response.
Shahar Sukenik (00:38:03):
So what we do is we try to resolve our experiments very quickly. So something like two minutes or less, generally, we actually do this in less than two minutes, we actually do this in less than one minute. And our perturbations are very rapid. The perturbation I’m going to talk about today is osmotic pressure perturbation, which is historically something that I’ve been really interested in and it’s actually gaining a lot of attention recently as a fantastic way to probe the biophysics of the cell. You can see here that these osmotic changes happen very reproducibly and very quickly. So it’s just a really good handle to perturb the intracellular environment.
Shahar Sukenik (00:38:42):
So what we do is we measure the D/A right, the FRET signal inside the cell before and after osmotic perturbation and we plot it like this. So over here on the Y axis is the full change in D/A. So the full change would be after perturbation and divided by before perturbation. And on the X axis, we see the osmotic pressure after the perturbation. So on the left-hand side, we have hypoosmotic pressures, so the cell actually expands under those conditions. And to the right of 0.3, we have hyperosmotic conditions, so the cell will actually compact. And isosmotic conditions are a 0.3 osmolars. Once again, we have this calibration to the base level and so if it’s above the zero line, the osmotic challenge will expand the sequence and if it’s below the osmotic challenge will compact the sequence.
Shahar Sukenik (00:39:43):
So we start out with just our glycine-serine repeats. So over here, we have increasing lengths of glycine-serine repeats. You can see in all of them when the cell volume expands, the environment becomes more aqueous and overall, it looks like the polymers expand. And then when the environment turns hyperosmotic, the cell compacts, water is pushed out, everything becomes more crowded, the glycine-serine repeats contract.
Shahar Sukenik (00:40:16):
So this is kind of what we expect. We were very happy to see this. This is also the trend we’ve seen in vitro, I don’t show that here, but it was apparent in the solution space scanning slide. And we can even get some insight showing that for very short glycine-serine repeats, the slope exists, but doesn’t really change, and the change really happens as sequences grow longer.
Shahar Sukenik (00:40:44):
But then this picture gets really interesting when we start looking at heteropolymers. And so we do that, we see trends that not only are extremely sequence dependent, but also simply can not be explained by any current biophysical theory that we usually use to explain in vitro observables. So you can see sequences that behave somewhat similar to the GS linkers like Nup15 and–things are in the way–and cMyc. But then we have stuff that show non-linear response like p27. We have stuff that’s extremely insensitive, like Nup49, and maybe most interesting, sometimes we get effects that simply contradict what we expect from crowding. So we have sequences that compact under hypoosmotic conditions, but expand under hyperosmotic conditions. This is absolutely fascinating to us. So since crowding models, and even osmolyte or denaturant models, can’t really explain this, what we really need to look at now is the biology, and that’s what we’re doing now. So we’re looking at post-translational modifications, we’re looking at protein-protein interactions that might actually affect the dimensions of these ensembles.
Shahar Sukenik (00:42:18):
So to conclude this part, disordered proteins maintain their hidden structure inside the cellular environment. These ensembles do respond to physical-chemical changes inside the cellular environment, to changes in the cellular solution. The sensitivity of these IDPs can sometimes depend on sequences, but sometimes depends on more than sequences. And so this is, like I said, the stuff that we’re starting to look at now.
Shahar Sukenik (00:42:46):
So with that, I’d like to talk a little bit about the implications of all of this. And so the point we started out with, is that the intracellular physical chemistry changes routinely. This is exemplified beautifully when you look at any imaging of mitosis. The changes that happen in the cell, the nuclear envelope breakdown, the change in shape, the change in the cytoskeleton networks, all of these are dramatic changes to the physical chemistry of the cell. And they happen every single day, hundreds of millions of times. So we hypothesized that certain IDPs are really evolutionarily designed, not just poised to sense and respond to these kinds of changes. And when you think about it, this is such an elegant, rapid, and efficient way to sense and respond to changes in the cellular environment.
Shahar Sukenik (00:43:43):
So zooming in now and thinking about phase separation, condensate-forming IDPs may have long-range hidden structures. These structures can also exist in condensate-forming IDPs, and even in LCDs. And we actually show this for a couple of them, Fib1 and A1 LCD, where it’s clear that A1 LCD, despite being actually longer than Fib1, shows a more compact conformation here. Not only that these condensate-forming IDPs can have a wide variety of sensitivities to their surrounding environments. And these two sequences happened to show this beautifully where A1 LCD is extremely insensitive–So this is an outlying insensitivity to changes in cell volume, whereas Fib1 has an outlying sensitivity to this. Not only that Fib1 acts completely opposite of what we expect from crowding. As the crowding decreases the sequence compacts, as the crowding increases the sequence expands.
Shahar Sukenik (00:44:46):
So this is really fascinating stuff and we’re trying to correlate this now with the Csat and other parameters in phase separation of these two proteins. But what’s important to remember, is that many condensates are formed in response to some perturbation to the cellular environment. And over here are two collaborations, one with Aaron Gitler and Seung Rhee regarding FLOE1, which is a plant protein that actually phase separates or condenses when the environment becomes hydrated. The other one is YAP, it is a transcriptional co-activator that phase separates again in response to osmotic changes around the cell in collaboration with Jennifer Lippincott-Schwartz and Danfeng Cai.
Shahar Sukenik (00:45:39):
So the fact that condensates are so responsive to these physical-chemical changes in the cellular environment raises a lot of open questions. How does the cellular environment actually regulate this liquid-liquid phase separation? How do the hidden structures inside these proteins determine the condensate properties? Whether it’s the tendency to form condensates, the material properties that condensates, or the specificity of these condensates? Another question is how did these ensembles change? How did these hidden structures change between the dense and the light phase? And how all of this is affected when the cellular environment becomes misregulated? And so that leads me to the next issue, which is that this finding can have really important implications when it comes to pathology.
Shahar Sukenik (00:46:31):
So pathology is a major driver of changes to the physical-chemical environment of the cell, and the examples are endless. Probably the most famous one is the Warburg effect in cancer cells, and that’s been characterized over a century ago and shown to change a whole bunch of different parameters inside the cell from the pH to the concentration of metabolites. But this kind of effect happens in viral infection, in neurodegenerative diseases, and many other different cases.
Shahar Sukenik (00:47:05):
And so if IDP sequences are meant to sense and respond to routine changes in the cell, we’ve got to ask, what happens to their function when the environment is strongly perturbed? A direction for us looking ahead would be to look into this as a mechanism by which diseases like cancer can progress, even in the absence of mutations, just by a misregulated intracellular environment.
Shahar Sukenik (00:47:34):
All right. So we’re almost done here, kind of a final summary: Long-range, hidden structures in IDPs are prevalent. The hidden structure can be tuned by changes to the intracellular solution, that’s why we’re thinking that the solution is the solution. That’s our initial pitch. Function, this is already known. The function can be regulated through local and global ensemble changes. And IDP function may be impaired in a pathological cellular environment.
Shahar Sukenik (00:48:07):
All right. So now come the thank yous. So I see the animations here are a little out of sync. I first of all, really want to thank the graduate students who did all the work here. So like Mark said, that the lab is relatively new. Most of the graduate students have been here, maybe two, two and a half years and the work that we’ve accomplished since then is really phenomenal. And so, Karina Guadalupe, a third year graduate student, has done almost single-handedly almost all of the live cell imaging data and analysis that I’ve shown you. Feng Yu has done all the computational work, including the visualization of long-range interactions. And David Moses has developed and taken all the in vitro data. Then these are other lab members who all contribute either intellectually or in actual help to the work. Our collaborators, Alex has done a lot of the simulation work that I’ve shown you and Erik Martin who’s helped extensively with some of the SAXS measurement.
Shahar Sukenik (00:49:21):
Again, a lot of this is done through infrastructure that you can apply to. So we’ve done SAXS measurements on SLAC, Argonne an ALS. Computational work on XSEDE and the Merced cluster, and funding from NIH and CCBM. Before I finish, I’m just going to say that the lab is looking for a postdoc. So sorry to put this shameless plug at the end, but if you’re interested, there’s a lot of freedom, diverse and supportive environment, four years of funding, super close to Yosemite. And by the way, here’s a recent lab trip where I almost killed my lab, that’s a story for another time to Yosemite. And yeah, with that I’d like to thank you for listening and take any questions.
Mark Murcko (00:50:13):
No, thanks very much. Wonderful talk. And by the way, if you could scarf up some tickets to Half Dome, I’d even sign up for the postdoc. It’s so hard to get them. Its so hard to get the Half Dome tickets, you know.
Shahar Sukenik (00:50:28):
Over here. A lot of the faculty have regular reminders on the day that the sign-ups start. That’s the only way to do it.
Mark Murcko (00:50:37):
Exactly. No, fantastic talk. Thank you. And lots of interesting questions coming in. I know we’re close to the top of the hour, so I’m not sure who’s able to hang around and who’s not. But maybe if Charlotte is still on, maybe… Your question I thought was a good one. Lets start with you.
Charlotte Fare (00:51:02):
Yeah. Hi. Can you hear me?
Shahar Sukenik (00:51:03):
Yeah. Hi, Charlotte.
Charlotte Fare (00:51:05):
Hi. Great talk. So I guess I have a couple of questions, so I’ll try and go through them a little bit quickly. But the first question is, have you looked at if there’s any relationship between how proteins respond in cells and their cellular function, like if proteins in one compartment or one function have a consistent activity?
Shahar Sukenik (00:51:35):
Right. So, I mean, you’re talking about some kind of GO ontology or something like that and see if it breaks down. Yeah. I mean, that’s a fantastic suggestion and one that we’re looking into. I think that right now, the data set that we have is probably not extensive enough to really get good inferences there. I mean, usually you get strong correlations when you have large genome scale- or proteome scale- data. Over here we’re ramping it up, but we’re still on 10 to 15. So I don’t think that currently there’s any strong correlations there, if that makes sense.
Charlotte Fare (00:52:20):
And then my other questions were about the in vitro experiments, because you talk about compaction a lot, but I was wondering if you had any insights into what the nature of that compaction is. Like are you forming similar types of structures within the intrinsically disorder domain and then relatedly are… Can you induce and reverse compaction and then are those cycles similar in any way?
Shahar Sukenik (00:52:54):
Right. So I’ll answer the first part. So the compaction from what we see at this point can be driven by different motifs. There could be different combinations that would bring about this compaction. Our way to know what it is, I think really hinges on our interpretation of the solution space scans. Because say the compaction is abolished by adding salt into the solution, that would mean that it’s driven by electrostatics. Whereas if the compaction happens with… or sorry, the compaction is abolished by even a small amount of say urea or guanidinium, it may imply that these are more dispersed along the chain and not as strong as electrostatic interactions would be. So this is really a direction that we’re looking into, zooming into the specific sequences and maybe with the use of mutations, even, trying to predict what the driving force is for compaction or expansion maybe. And so in terms of the cycles, do you mean cycles, like you talking about live cell or maybe help me understand exactly.
Charlotte Fare (00:54:13):
Yeah, I guess it might make more sense to do experiments in a live cell but even in a FRET type situation where you maybe replace the buffer with one that promotes compaction and then reverse it. And If you see maybe it becomes less dynamic when you switch buffers back and forth [inaudible 00:54:40]
Shahar Sukenik (00:54:39):
So kind of like hysteresis as a result of what we see when we do thermal denaturation, right?
Charlotte Fare (00:54:44):
Shahar Sukenik (00:54:44):
Yeah. That’s actually a fantastic idea. We have not done this. The way we do this is on a plate reader, so we just do one and done. But that’s great and we should probably look at that. Thanks.
Mark Murcko (00:54:58):
A couple of good questions also from Amayra. The first one, I don’t know if you’re still on the line, but I love especially the first one there, the age-old challenge of distinguishing results in cells from in vitro in particular in the context of the work that you described today. You could probably see it in the chat.
Shahar Sukenik (00:55:22):
Okay. Let me see because I haven’t looked at the chat at all.
Mark Murcko (00:55:24):
About the in vitro versus…
Jill Bouchard (00:55:26):
I’ve unmuted her. She’s ready to talk.
Mark Murcko (00:55:29):
Shahar Sukenik (00:55:30):
Amayra Hernandez Vega (00:55:30):
Hi. Great talk.
Shahar Sukenik (00:55:30):
Amayra Hernandez Vega (00:55:30):
Hi. And so the first question it was regarding the pertubation as you have done in vitro and in cells. Because in cells you have interactions of this proteins with other molecules, DNA or the proteins. So do you have any example that behaves completely different in vitro after perturbations and in cells?
Shahar Sukenik (00:55:50):
Yeah. So essentially every single time we’ve seen a protein expand when the cell volume decreases, that’s a result that cannot be reproduced in vitro. At least as far as we expect a PEG solution to be a more crowded solution, which there’s problems with that assumption. But, like I said, there are solutions that would expand the protein, but not in the context of a compacting cell. And so, what you said is absolutely right. That’s the directions we’re looking at. So we’re starting to run these in-situ cross-linking experiments where we cross-link the entire contents of the entire proteome of the cell and then pull down on our specific proteins and see what gets attached. Then we can do that in different osmotic perturbations and see how that gets modulated. Maybe in the context of that, I mean we had a 2017 PNAS paper where we showed that these changes in volume can actually cause protein-protein complexes to form more readily. And so that very well might be something that’s happening.
Amayra Hernandez Vega (00:57:15):
Do I have time for another question or?
Mark Murcko (00:57:17):
Yeah. Your other question is also one of those classic questions, cytosol versus nucleus. It’s a great question. Why don’t you go ahead.
Amayra Hernandez Vega (00:57:26):
And so I was wondering with the GS repeat [inaudible 00:57:29] in cells, if you target to the nuclei or cytoplasm, if you see a difference with the osmotic pressure [inaudible 00:57:33]
Shahar Sukenik (00:57:35):
So actually Dianna, one of the undergrads in the lab is doing that analysis now. So we’ll have that soon.
Amayra Hernandez Vega (00:57:43):
Mark Murcko (00:57:44):
Beautiful. Beautiful. How about Kibeom Hong? If you’re still on. Maybe not. But the the question in the chat is interesting. The FP-fused IDPs, might they exist as multimeric forms?
Shahar Sukenik (00:58:07):
Mark Murcko (00:58:07):
in those measurements you’re seeing?
Shahar Sukenik (00:58:10):
Right. So that’s a good question and definitely something that we were concerned about… Let me just find this slide. Yeah. So over here this is the untethered fluorescent proteins. You can see that… So what we do is we use them in the same concentration that they would be as a tethered construct. And you see that when we break that link the FRET efficiency is zero. I don’t have this here, but we also do some concentration dependence of the FRET efficiency and there is none. Again, as we go to higher and higher concentrations, this might happen. Inside the cell, again, we do similar controls, and we do get some artifacts and cells that have extremely high expression levels, again remember this is transient expression so expression levels are varied. We generally filter out over a window of expression so we want to intermediate levels of expressions. Cells that are expressing a lot, we generally filter those out. And that’s kind of our way to circumvent this. Our main control is–inside the cell at least–is looking at correlations between the FRET signal and the direct acceptor signal, which is only proportional to concentration. Our basal hypothesis is that, that’s kind of like concentration series when you think about it. Our basal assumption is that we’re measuring the monomer and so there should be none. So that’s kind of the control we have on that.
Mark Murcko (00:59:57):
That’s great. So Emery has an interesting question about how the properties of the IDPs affect your measurements. Why don’t you go ahead, Emery.
Emery Usher (01:00:06):
Hi. Awesome talk. So I was just curious if you’ve done any, looking into how the sequence space of the IDPs you selected and the study, how those may scale or trend with the measurements of Rg in the different conditions. So if you’re looking at net charge per residue or hydropathy, how does that, if at all, does that impact what the Rg looks like?
Shahar Sukenik (01:00:31):
Right. Thanks. That’s a great question. So, basically the best way to answer this is to think about the scrambles. The global sequence properties of the scrambles are all the same. Net charge per residue, the average hydropathy, that’s all going to be the same. I’ve already shown you that those get very different… Again, I’m trying to look for that slide. Yeah. So those give very different, both Rg and Re results. Then if we try to correlate this with patterning behavior, usually that is more successful. I don’t have this because it’s very preliminary yet still, but for example, this was surprising, but the behavior in the cell seems to correlate a lot with actually hydrophilic patches. So the hydrophilic patches showed the strongest correlation with the basal dimensions of the sequences inside the cell. So this it’s definitely something we’re thinking about in a fantastic question that I still don’t have a straight answer for.
Emery Usher (01:01:44):
Cool. Thank you so much.
Mark Murcko (01:01:46):
Great. And how about Yifan?
Yifan Dai (01:01:50):
Yeah. Thanks for the talk. So my questions is just a few, for some protein that’s supposed to increase the solubility for your IDP fusions, therefore how that can impact the IDP phase transition behavior.
Shahar Sukenik (01:02:05):
Right. That’s a great question. So the way we’re approaching this right now, and you can think of this as a disadvantage, but I actually think of this as an advantage. The reason is that… and here, let’s bring up this slide here. The reason is that adding those FPs, kind of ensures that we’re looking at the monomer. And so if our question is, what is the behavior, or how does the behavior of the monomer then get translated into Csat or some collective variable, then we can simply look at the behavior of the monomer and then do some correlative experiments with the unlabeled sequence to see if it makes sense. Does that answer the question?
Yifan Dai (01:02:51):
Yeah, kind of. I have a follow-up question. So when you do this kind of FRET measurement, you are investigating the intra-protein interaction. So how this intra-protein interaction can be correlated to this kind of phase separation behavior specifically? Are they competitive mechanism?
Shahar Sukenik (01:03:10):
Right. So that’s a great question, I think. So first of all, the dimensions of a sequence can correlate with Csat. I think Erik Martin and Alex Holehouse showed this in their Science paper last year. But in this case, especially when you think about these long range interactions, if we think about the sticker and spacer model, then it’s definitely possible that under certain conditions stickers will stick to stickers intramolecularly and compete out with intermolecular interactions. And that’s one of the directions where we’re thinking about.
Yifan Dai (01:03:52):
Mark Murcko (01:03:58):
I think maybe we’re at a good stopping point. So we’re about 10 minutes over and we are starting to lose folks. So, maybe this is a good spot. What often happens, Shahar is that people will have questions later. So don’t be surprised if you get pinged by different people saying, “Hey, I loved your talk, but I was thinking some more about it and…” So that happens quite often.
Shahar Sukenik (01:04:27):
All right. Yeah, looking for it.
Mark Murcko (01:04:27):
Bede just said he’s going to email you 50.
Shahar Sukenik (01:04:28):
Yeah. I saw that.
Jill Bouchard (01:04:32):
And we’ll have you email some of the answers to some of the other questions that you didn’t have time to answer too. So, we’ll get back to you soon with the chat.
Shahar Sukenik (01:04:41):
All right. Fantastic.
Mark Murcko (01:04:41):
Beautiful. Well, thanks again. Thanks, Shahar and thanks everyone.
Shahar Sukenik (01:04:43):
Thank so much for having me. This was absolutely great.
Mark Murcko (01:04:45):
Really stimulating, I’m thinking. Lots of good stuff here. All right. Thank you everybody.
Jill Bouchard (01:04:51):
Mark Murcko (01:04:52):
Okay. Take care.
Shahar Sukenik (01:04:54):
Mark Murcko (01:04:55):
Question from Gaurav Chauhan: Could the sequence-dependent changes (in IDPs, not in GS homopolymer) in persistence length affect the values of R_e one would get from FRET experiments?
Shahar’s Response: So this was exactly the concern that made us decide to correlate FRET and SAXS. We hypothesized that the changes won’t be only from distance, but also for example from preferred orientation of the dipole transition moment in the fluorescent proteins. However, this type of orientation is only possible when the sequence is non-isotropic – in other words, when the sequence has some structure to it. The fact that our experiments corroborated this effect is further evidence of hidden structure…
Question from Hamed Kooshapur: How much do the fluorescent proteins affect your results? How can you exclude that the IDP is not interacting with these proteins (at least transiently)?
Shahar’s Response: The short answer is that we don’t know. However, the way our experiments are designed means that the differences in FRET measured between different proteins is a property of the sequence rather than FP interactions, and so even if such interactions exist, our conclusions still hold. It’s also important in this context to remember that most IDRs are in fact attached to well-folded domains, and in these cases the question of how the sequence functions untethered may be just as problematic (if not more!).
Question from Kamran Rizzolo: Heat soluble proteins tend to be enriched in IDRs, did you happen to check any sequence specificity by heating your peptides in vitro?
Shahar’s Response: This is a great question – unlike solutes, which interact specifically with specific residues, heat acts uniformly on the entire chain. We do expect heat capacities and structural changes to be affected by the hidden structure in IDPs. We are currently working on running this assay in qPCR instruments to get these high-throughput “melting curves” and see if they correlate with predicted structure.