|Type||Kitchen Table Talk|
On October 14, Dewpoint and Condensates.com hosted a rising star in the condensates field, Kansas State University’s Jeremy Schmit, as part of our Kitchen Table Talk series. Over the course of his career, Jeremy has been tackling questions about complex protein behavior, which has naturally led him to tackle questions about the complexities of condensates. Most recently, he has worked toward understanding gel-to-liquid phase transitions, amyloid assembly, and all the levels of disorder inside condensates. This talk draws on all of these experiences to delve into the physical principles that enable liquid structures to encode all of the amazing and specific functions that are so critical for life.
Jeremy got the audience thinking and asking questions—so much so that we ran out of time, but he was nice enough to finish answering questions via email. We also got some great answers from Tanja Mittag, who collaborated with Jeremy on some of the work he presented. You can see both of their comments below. We hope the talk gets you thinking too. And if you want yet another way to think about the complex physical behaviors of proteins and amyloids, check out this cool video that Jeremy and his colleagues made. Spoiler alert: Legos.
Create an Account or Sign In to view the video.
Mark Murcko (00:00:00):
It’s obviously, a great pleasure to welcome Jeremy Schmit today. Jeremy comes from the physics department at Kansas State University. And Jeremy’s whole career, really has been a masterclass at studying the complex behaviors of proteins. And this of course, quite naturally, has led him to phase behavior and thence to cellular condensates. All roads as everyone on this call knows, all roads lead to condensates.
Mark Murcko (00:00:24):
And so, as we now know, condensates are able to carry out many diverse functions. But the question then becomes, what are the common features of condensates? And on the other hand, what makes condensates unique so that one type of condensate is able to carry out certain specific functions that are different from the functions that are carried out by other condensates. And then that leads you to the question of what are the underlying physical properties of the proteins and the nucleic acids at the nanoscale, that then lead to the macroscale behavior that we’re observing in cells? And so these are the kinds of questions that Jeremy has been tackling with great success.
Mark Murcko (00:01:08):
And he’s published on amyloid and huntingtin, looking for the triggers of aggregation and how that connects to disease. He’s collaborated with Tanja Mittag, on the transition between liquid and gel-like states. That’s a JACS paper that came out I think, last year.
Jeremy Schmit (00:01:24):
Beginning of this year, yeah.
Mark Murcko (00:01:26):
Oh, beginning of this year. And then there’s the work with Mike Rosen, which I think is still on bioRxiv. And that helps us to better understand this really interesting way that biological function can emerge out of these condensates, despite the fact that there is disorder in the structural states of the proteins inside of condensates, which is just a fascinating area.
Mark Murcko (00:01:49):
He also on his website, at Kansas State, he has these wonderful little movies that use Legos to explain some of these incredibly complex physical behaviors of proteins. And everyone should definitely check those movies out if you haven’t seen them already.
Mark Murcko (00:02:05):
So Jeremy’s talk today is on the Biophysical Basis of Condensate Functionality and he’ll cover the physical principles that enable these liquid structures to encode all of these amazing and specific functions that are so critical for life. So Jeremy, the floor is yours. And thanks again for doing this.
Jeremy Schmit (00:02:25):
Oh, thanks a lot. And thank you for a wonderful introduction. So the way I want to start out this talk other than showing one of the nicer views that we have here in Kansas, is by confronting my biases here, and maybe confronting is the wrong word, I’m at least going to confess them.
Jeremy Schmit (00:02:50):
And as Mark mentioned, I’m in a physics department. And I very much think like a physicist. And to illustrate that line of thinking here, I’m going to use these two molecules as a test case here. If I’m interested in describing what’s going on with molecules like this, I’m going to start looking at certain physical properties here. And some of the first properties that might pop to mind would be things likehow big are these…
And these are both sort of small-ish proteins on the order of 100 amino acids. They’re also weakly charged; we expect proteins to behave roughly like colloids, and these guys are relatively weak, or at least weakly charged or uncharged.
Jeremy Schmit (00:03:37):
Finally, we could go in and look a little bit more at the structure. And these are both globular folds, with a mixture of alpha and beta, subunit or, elements in here. And so looking at these together, I’d feel quite confident that anything that I said about the physics of one of these molecules is likely to be equally valid for both of them. Physically, these molecules are very, very similar.
Jeremy Schmit (00:04:06):
However, biologically, that’s not true at all. Because in biology, what we care about is the function of these molecules. And if we put the function up on there, ubiquitin is of course, a molecule that signals a protein for degradation. Whereas barnase is an enzyme that goes and chops up bonds.
Jeremy Schmit (00:04:26):
So biologically, these molecules are incredibly different. And so this is going to set up a theme that’s going to recur throughout the talk here. This contrast between physical principles that apply to large numbers of molecules in biological function, which is incredibly system specific.
Jeremy Schmit (00:04:45):
And so of course, we’re not talking about individual globular molecules here, we’re talking about condensates. So the first thing I want to do is sort of lay out the case for condensates on both ends of the spectrum here, which is sort of repetitive from Mark’s introduction.
Jeremy Schmit (00:05:02):
But if we’re looking at condensates from the physics point of view, there’s a lot of good reasons to expect that we’re going to see universal physical behavior here. And for somebody like me, this is incredibly attractive. We saw a couple months ago, a number of wonderful talks from Rohit, talking about the polymer physics and the stickers-and-spacers morphologies that pop up over and over and over again. We have things like the client scaffold framework that shows up over and over again. And so there’s just huge elements of commonality in these condensates here that’s very strongly suggestive that we can say stuff universally on the physics front.
Jeremy Schmit (00:05:47):
There’s also phase transitions. We love universality in physics and phase transitions is a place where universality is incredibly strong. It turns out that we use almost the exact same mathematical models to describe the spontaneous alignment as a magnet as we do for the separation of oil and water. This is very remarkable, that such a remarkably different systems, can be described by the same underlying mathematics.
Jeremy Schmit (00:06:16):
And finally, as I’ll get into the talk later, the very fact that we have liquidity on these things is suggestive that there’s some sort of universal behavior. Liquids are inherently disordered, and this would seem to make it the fine details of given systems less important. And so based on these sorts of things, we might expect that existing theories, things like Flory–Huggins might be great descriptions of these sorts of systems that we can use.
Jeremy Schmit (00:06:48):
Now, the counter argument that these are all different is very succinctly summarized by this slide. And that is we’ve identified lots of these condensates, they show up throughout the cell, lots of different locations and contexts. And most of these functions have not been fully resolved. But the ones that have been are showing a remarkable diversity. There’s lots of different things that these condensates can do.
Jeremy Schmit (00:07:15):
Okay, so this sets up the story I want to tell here. And to give a brief overview of where I want to go with this, what I’ll do is talk in detail about … I’ll tell two short stories of systems that we’ve looked at in detail. And we’ll look at some of the interesting features that come out of each one of those systems. And then the last third of the talk, what I want to do is compare and contrast those two systems to see if any universal messages start to emerge.
Jeremy Schmit (00:07:45):
So the first system I want to talk about is condensates formed by SPOP and DAXX. And of course, as a physicist, I’m going to start off with the physical properties of SPOP. SPOP is a globular protein, it has three domains. Let me make this guy smaller so that I can see what I’m doing. Okay. So it has three domains, the red and blue domains like to dimerize with each other. And so this leads to the formation of these rod-like assembly’s here. And then the green domains here, these guys sort of dangle off to the side here. And what the green domains do is they bind to DAXX. So DAXX is an intrinsically disordered protein, and it’s got roughly five sites that recognize and bind to these green domains on SPOP here. And so the cartoon that this evokes is something like this, where we have these sort of disordered linkers that are going to join together these long rods here.
Jeremy Schmit (00:08:58):
Now, SPOP is an interesting protein. Biologically, it’s the most commonly mutated protein in prostate cancer, a lot of the mutation sites seem to be associated with these green domains here. So there’s good reason to think that the binding behavior of SPOP and DAXX, their ability to co-localize, it might be playing a role in cancer progression.
Jeremy Schmit (00:09:24):
So Jill, who introduced this talk, measured the phase diagram of SPOP and DAXX, and this is what it looks like here. And what we see here is sort of three distinctly different phases here. In the low DAXX concentration here, so this is when we have SPOP but no DAXX. What we see here is basically a dilute system of these SPOP rods. Okay, so the SPOP rods are there, but they don’t associate with themselves and so this would be something that looks to me like a gas or a vapor of SPOP.
Jeremy Schmit (00:10:02):
As we start to introduce DAXX into the system, or Jill introduces DAXX to the system, what you see is a gel-type network showing up here. Okay, so we’ve got this sort of fibrous network showing up here. And then as you add more DAXX to the system, what happens is the gel dissolves, and you form these liquid droplets here.
Jeremy Schmit (00:10:30):
Now, this to me looks weird because I look at this phase here, this to me, it looks a lot like a solid. Gels are very often just molecules that are condensing into a solid state, but they’re kinetically arrested on the way there. And so this looks to me like a kinetically, arrested solid. So what we have here is a vapor, a solid, and then a liquid. And of course, this is inverted from the usual; you expect to see vapor, liquid then solid. So we’ve got an inversion here. And that’s what was I found interesting about this data that got me involved in it.
Jeremy Schmit (00:11:06):
Now, the last piece of the story that I should mention here is that at very high concentrations, DAXX will phase separate into liquid droplets by itself, it doesn’t need SPOP around to do it. So there’s a self-association due to DAXX, but it’s fairly weak, because you need to go to high concentrations for that to show up.
Jeremy Schmit (00:11:26):
So we set to work trying to figure out what was going on here, and putting together physical models to describe that. And I should mention at this point that everything … Well, everything that I’m going to show you here is pen and paper theory, there’s not a single simulation in this talk. This is all just writing down statistical mechanical models, and testing them against the data and seeing if they can quantitatively describe the data or not. And if it doesn’t, you throw it away, and you start again. And so there’s a lot of trial and error involved here. And skipping over all of that, this is the model that ended up working here.
Jeremy Schmit (00:12:03):
So what do we have here? So this gel phase here, this isn’t going to surprise anyone, this is basically just networked, or cross-linked bundles of SPOP. So in my cartoons here, SPOP is these rods here with the green dots represent binding sites for DAXX. DAXX are these little black lines here. And I’m coarse-graining DAXX as being bivalent. And that actually ended up working very well, we can discuss that later, if it pops up in the questions here. But basically, we’ve got DAXX forming these cross-links that are joining these rods together and forming these rigid bundles that are going to form the backbone of this gel phase.
Jeremy Schmit (00:12:44):
This is your vapor phase, you still have the DAXX present, but there’s not enough DAXX to join the rods together. Basically, what’s happening is that you’ve got a significant translational entropy cost to bring the rods together, and you need a driving force to overcome that entropy. And interestingly, if you compare this phase to this phase, you notice that all of your DAXX is bound up. And to a pretty good approximation, that’s true.
Jeremy Schmit (00:13:11):
So there is no change in the binding energy going from this state to this state. And it turns out that the driving force for this is actually the entropy of the number of ways that you can arrange DAXX on the system. So this DAXX molecule can form bonds in all these different directions, whereas this guy here can only form bonds with the neighboring sites here. So there’s a big entropic benefit to forming this. And that’s the thing that condenses your rods.
Jeremy Schmit (00:13:41):
Now, the more difficult one to explain is what’s going on in here in the liquid phase. And the model that we converged on here was this one here: And what’s happening out there is that DAXX has essentially saturated all of the potential binding sites on SPOP. And so what happens is this molecule here, cannot form a cross-link with a neighboring rod because there’s just simply no sites for it to bind to, they’re already occupied with other DAXX molecules here. And so what happens is instead, it forms these weak interactions with the neighboring DAXX molecules. And as I mentioned before, these interactions are very weak. But we’ve got this large assembly with lots of DAXX molecules on there. And so we’ve got this amplification of weak interactions. And this helps drive it together and to form a liquid phase at much lower concentrations than DAXX would be able to do so on its own.
Jeremy Schmit (00:14:43):
Okay, so what Jill then did is she went and measured the binding affinities for DAXX in these two configurations. So one configuration is where DAXX forms two bonds with SPOP and one is where it just forms one bond, and we can measure equilibrium association constants for both of those. And then we can feed those into the theory. And we can calculate a phase diagram from it. And lo and behold, it agrees very well. These are sort of the phases that we observed on that plot that I showed you on the three slides ago or something. So these are your sort of gel states, these are the liquid states, and these down here would be the vapor states. And there’s a log plot. So of course, there’s going to be more vapor states over here that don’t show up unless you go to a crazy scale.
Jeremy Schmit (00:15:34):
Okay, so this is a real nice agreement with experiments. And there’s not a lot of free parameters in this model here. In fact, there’s very few because Jill measured the association constants for us. But there’s a few tightly constrained parameters where there is wiggle room there. So it’s nice to see if we can get an independent verification of this model.
Jeremy Schmit (00:15:56):
And the way we got that was looking at the densities, the molecule concentrations in the two phases here. So this over here is the gel phase. And we’ve got this sort of disordered network of SPOP fibers here. And we don’t really know what that is, but Jill measured those concentrations here. But based on that concentration of SPOP, we can calculate how much DAXX should be present based on her equilibrium constants. And that fits very well.
Jeremy Schmit (00:16:30):
Next one, when we go into the liquid phase. Well, the density in the liquid phase is basically set by the volume of this assembly right here. And this looks to me kind of like a cylindrical brush, okay? We’ve got this central axis here with bristles hanging off to the side here. And so the radius of this cylindrical brush here should be given by the radius of gyration of the DAXX. And if we plug that in, it turns out that it gets the concentration of SPOP and the concentration of DAXX, pretty darn close here.
Jeremy Schmit (00:17:04):
And so we can calculate the concentration of DAXX increasing as we start to saturate the binding sites here in the liquid phase. And then eventually, what happens is you start to form a dilute or a DAXX phase by itself when you get to very high DAXX concentrations. And what happens here is that these SPOP rods start to dissolve in this pure DAXX fluid here. And so your SPOP concentration drops down dramatically while this DAXX concentration levels out at its equilibrium value. And so your limiting behavior is going to look like this, where you’ve got all DAXX with this occasional SPOP rod dissolved among it.
Jeremy Schmit (00:17:46):
Okay, so with that, let’s summarize what we’ve learned from this SPOP/DAXX story. We have two phases here, one of them is a gel phase. And it’s driven by these strong SPOP-DAXX interactions, okay? So these are your strong interactions to the system. And there’s this interesting feature that the driving force comes from the entropy of arranging molecules on there, not necessarily from the actual binding energy itself.
Jeremy Schmit (00:18:15):
Then we have this entirely separate liquid phase here that’s driven by these weaker DAXX-DAXX interactions. That’s down here, these DAXX-DAXX interactions. And in this case, SPOP is helping to assist this assembly by collecting large number of DAXX molecules together. And in fact, amplifying this effect of these weaker interactions here. And so we can transition between these two phases, this is reversible, both of these two phases are quasi-equilibrium. And so we can revert back and forth between them by altering the stoichiometry of the system. So this is your SPOP-enriched region, and this is your DAXX-enriched region.
Jeremy Schmit (00:19:01):
Okay, so let’s move on to the second system here. And this system here is condensates formed by poly-SUMO and poly-SIM. And so let me just describe that briefly here. So this is a study by Mike Rosen and his co-workers there. And what they did is they made these assemblies here where you took repeated units of SUMO, which is, as far as I’m concerned, it’s just a small globular protein. And they linked it together with these 12 amino acid linkers. So this is a completely synthetic system. And I’ll get into why they did this in just a moment here.
Jeremy Schmit (00:19:40):
And so then they did the same thing with SUMO’s binding partner here. This is SIM, it’s just the 23 amino acid sequence that likes to bind to SUMO. And so both of these cartoons show tetramers, they actually did their experiments with decamers. My cartoons are all the simpler version. But every time you see a tetramer here, imagine that there’s in fact, 10 binding sites.
Jeremy Schmit (00:20:05):
So to me as a physicist looking for universal properties, these two systems look very much the same. Okay, once we have a globular protein here with these repeated units at regular spacings, and then we have an intrinsically disordered binding partner with discrete binding sites here. And so there’s every expectation, at least in me, that we’re going to see some sort of universal behavior.
Jeremy Schmit (00:20:30):
But before we get into that, let’s motivate these experiments a little bit. And what they were trying to do here was test this client-scaffold framework here. And many of you probably know this, but just for completeness, let’s run through this here. So the idea here is how do you build an organelle? Well, if I want to build a membrane bound organelle, of course, the first thing I do is I set up a membrane. Okay, so this is my organelle. On this, of course, doesn’t do anything. And so in order to give it functionality, I need to stuff it full of enzymes. Okay, so there’s a physicist picture of an organelle here that has functionality because I’ve got enzymes stuffed inside of it.
Jeremy Schmit (00:21:13):
So if we want to do the same thing for a membraneless organelle. Well, by analogy, what we’re going to do is we’re going to form some sort of network that holds this thing together. And these are your scaffold molecules here. And these basically serve like the membrane here, in just establishing the fact that you’ve got an organelle.
Jeremy Schmit (00:21:33):
But this thing here is going to be somewhat inert. And so in order to provide functionality to it, we’ve got to decorate it with enzymes, okay? And so this is the way we get a functional organelle here. And so this was their motivation in designing these molecules. And so the idea is that this poly-SUMO and poly-SIM, these would serve as the scaffolds these are the sort of black lines that define where my condensate is, and where I can provide functionality. And then they’re going to test the ability of these scaffolds to recruit small molecules by probing them with these low valence versions of the same molecule. So this would be just a monovalent or a divalent SIM here. And this would be a monovalent SUMO, and they’ve fluorescently tagged these molecules, so that they can tell the difference between the clients and the scaffolds here.
Jeremy Schmit (00:22:28):
So as I alluded to earlier, I expected to see a lot of universal behavior between these two systems. These are both just different versions of your sticker-and-spacer morphology here. And we’ve already worked out a nice theory over here that describes the condensation of this sort of sticker-and-spacer network here. And so my expectation here is that the only thing we’re doing is we’re changing the valence of the molecules going from this system to this system, and maybe changing the lengths of the linkers. And then we’re off to the races.
Jeremy Schmit (00:23:05):
And it turned out that that failed catastrophically. And the basic thing is that when we did this, what we found is that the condensates that would have arisen out of this theory, were darn near close-packed. We were getting theoretical volume fractions on the order of 90%. Well, there’s a conversion there between theory units and experimental units, but basically, you’re expecting these molecules are darn near bumping up against each other. Whereas experimentally, the volume fraction observed in these condensates was on the order of 2%. So no matter how you change your physics units here, you’re still orders of magnitude too low in concentration.
Jeremy Schmit (00:23:50):
Okay, so what else could we try here? Well, we tried throwing every other type of model we thought we could add them. We tried three component Flory–Huggins, we tried this, that the other thing. They all gave very, very similar results here.
Jeremy Schmit (00:24:03):
And it’s fairly easy to understand why we would get this result. If you’re going to form some kind of three-dimensional network. Okay, there we go. There’s our three-dimensional network. There’s a characteristic length scale right here, distinguishing how big the porosity in this network is.
Jeremy Schmit (00:24:21):
And that characteristic length scale is just going to come from the distance separating your repeat units here. And so basically, no matter what we did with these theories there, they’re saying that you’ve got this driving force to satisfy all these binding interactions here. And this is going to cause these things to condense into these very, very dense states.
Jeremy Schmit (00:24:44):
And so given that there was no fix that we could apply to this calculation to reconcile this behavior here, we were forced to abandon this model here. And so this resulted in quite a bit of head-desk contact, as we tried to figure out what the heck was going on here. And after a while it occurred to us that this sort of three-dimensional network is actually a very inefficient way to satisfy these binding sites. And in fact, it’s much more efficient instead of lining these things up randomly. If you line them up parallel to each other, it’ll allow them to stack with each other.
Jeremy Schmit (00:25:26):
So what does that look like? Okay, so here’s one scaffold here. And what I’m going to do is I’m going to try and satisfy as many binding sites on this as I can by aligning the next molecule with it. Now, of course, this is a statistical thermal system, so I’m not going to get the alignment perfect, but then I can stack another molecule on there. And I can continue to grow this thing by stacking more molecules on top of it. And so what’s happened here is I’ve built a one-dimensional filament here that comes from alternating poly-SUMO, poly-SIM, poly-SUMO, poly-SIM.
Jeremy Schmit (00:26:03):
Now, notice that the way I’ve drawn this thing, there’s defects all over the place on it. And in fact, there’s two of them that we can talk about specifically. One is that we have these sticky ends hanging off the outside. And this is what allows us to add new molecules to it and to elongate the filament.
Jeremy Schmit (00:26:22):
And the second type of defect is gaps. Okay, these are the spots in the middle here, where I have an unbound site. Okay, so now we have an explanation for why these networks are so much less dense than we would expect using a random network. And that is that the only places you can bring these filaments together are at these defects. Okay, so we’ve got a new mesh size that’s arisen here. It’s not set by the spacing between these individual modules, it’s set by the spacing between defects. And so what’s going to happen is I’m going to build a three-dimensional network of these filaments, and they can only cross-link each other at these places where you have these binding defects. And so now notice that my new pore size here is now this much bigger thing here. That comes due to the fact that anytime I have these zipper motifs here, this is basically inert to further binding.
Jeremy Schmit (00:27:29):
Okay, so what this means is that if I want to calculate the properties of this network, what I need to do is I need to calculate this defect density. And this is nice because what we’re doing here is we’re essentially building these one-dimensional elements here. And when you’re working in one-dimensional, there’s all kinds of just wonderful and super powerful analytic stat. mech. tools that we can use in order to calculate what’s going on. And I can go into those in detail if anybody is brave enough to ask me in the questions or in person by email, but I’m going to skip over all that now.
Jeremy Schmit (00:28:06):
So what we’re going to do is we’re going to calculate the defect density. And this is nice because the defect density also is going to tell us where we can bind clients. So we can get a two for one in this calculation. We can figure out how far apart these network cross-links are, and we can calculate how many clients we’re going to recruit to this network, which of course, is what Mike was originally interested in when he made this synthetic system.
Jeremy Schmit (00:28:34):
So this is what the experimental data looks like here. What we’re looking at is a fluorescently tagged client here. And on the axis here, we’re looking at the stoichiometry of the scaffolds. So this is I’m increasing my blue scaffold concentration, this is increasing my yellow scaffold concentration.
Jeremy Schmit (00:28:52):
And what shouldn’t be surprising at all, is that as I increase the concentration of blue scaffold, I get more fluorescence, okay? If I have an excess of blue over here, there’s more places for these yellow clients to bind, and so my fluorescence is going to go up.
Jeremy Schmit (00:29:09):
Now, that trend is not surprising, but what is surprising is when you look at these fluorescence quantitatively, okay? So look at this bottom row right here. And this bottom row would be along this chart right here. And what they saw in the experiments is that there’s actually a non-monotonic trend here. Okay, the fluorescence goes up at first, and then it starts to go back down.
Jeremy Schmit (00:29:38):
Okay, and this was very, very baffling to us. We had no idea what was going on there. But quite frankly, we were too frustrated by the concentration problem, to give it much thought. And so it came as a great surprise to us much, much further down the road in this project, when we realized that this filamentous morphology that we cooked up to describe the density also provided a very nice explanation for what was going on with this nonmonotonic client binding.
Jeremy Schmit (00:30:08):
So here’s the story. Okay, so I’m building these one-dimensional filaments here. And this is where I’ve got roughly equal scaffold stoichiometries here. So I’ve got two blue scaffolds and two yellows. And so I get a sticky end at both sides.
Jeremy Schmit (00:30:24):
Now, if I’m going to titrate in more blue scaffold, well, there’s only one place it can go. It can only attach at this sticky end over here. And so what that means is that I’m going to end up with something that looks like this. So this of course, has increased the number of loose binding sites that I have available. And so this is going to recruit more monovalent client. And so my fluorescence goes up. And I’m sure some of you are wondering what would happen if I binded, my blue scaffolds at a right angle here. That is a correction of theory, we have accounted for that, we know exactly what’s going on with that. It turns out to be a minor correction until you get way out here. And again, I’d be happy to talk about that in the questions if it comes up.
Jeremy Schmit (00:31:13):
Okay, now, what happens now, if I continue to titrate in the blue scaffold, is I don’t have any more sticky ends to attach it to. And so the blue molecules don’t have any place to go. And so instead, what happens is they start to accumulate in free solution.
Jeremy Schmit (00:31:32):
And so what happens now is you’ve got these free blue scaffolds in solution, that now start to compete for the clients. And so they suck the clients out of the condensate, and the fluorescence of the condensate goes back down.
Jeremy Schmit (00:31:49):
Okay, now contrast that to what would happen if you had a random network. Okay, so here’s my random network thing. And now I start to titrate in my blue scaffolds. And the point is that a random network is going to be a lot more tolerant to the amount of blue scaffold that I’m going to get in. You are going to get this non monotonic behavior, but you’re going to have to go out to much higher scaffold concentrations, scaffold asymmetries, before you start to see it.
Jeremy Schmit (00:32:11):
And the sensitivity that we saw in this system here, comes due to the fact that we’re building these one-dimensional filaments. But the one-dimensional filaments have another important feature as well. And what happens is that they establish correlations in where the binding sites are located. So if I had a random network here, I’m again going to have these binding defects of unsatisfied sites. But these are going to be essentially scattered randomly throughout the solution. But what you can see here is that the way these filaments established is that I’m very likely to have these defects that are close by to each other, both in the gaps and especially at the sticky ends.
Jeremy Schmit (00:32:54):
Okay, so you’ve got these correlations in where your available binding sites are. And this becomes particularly important when you start to introduce clients of different valence in the system. So they use monovalent systems, they used divalent clients, and they even use trivalent clients. I’ll show you those on the next slide here. And so of course, these clients are going to be sensitive in different ways to the ways that these collections of binding sites are located.
Jeremy Schmit (00:33:23):
Okay, so let’s look at that. This is the partition coefficient. So this is the ratio of client in the condensate to a free solution. And we’re comparing for the trivalent client, the divalent client and the monovalent client here.
Jeremy Schmit (00:33:40):
Notice that, okay, the trivalent client is recruited much stronger, no surprises there. It’s got more binding sites, a stronger binding affinity. But the other thing is, notice that the peak recruitment shifts, this one has peak recruitment at this 50 to 60 ratio, this one here, the peak recruitment is at about 65. And over here, the peak recruitment, you can’t see it on the scale, I’ll show it to you in just a moment. It’s somewhere around 75, 80.
Jeremy Schmit (00:34:10):
Okay, so the valence establishes where this peak is. Now, let’s change the affinity of the modules. Okay, so if I change how strongly one of these binds to its neighboring site there, what that does is it changes the height of the peak, but it has very little effect on the location of the peak. So this is the trivalent peak. This is the monovalent peak. So it’s changing how high the peak is, but it’s not shifting it back and forth. And this is because we have those correlations in the binding site.
Jeremy Schmit (00:34:41):
Okay, so this makes it a fairly interesting prediction here. The next slide I’m going to show you–what we’ve done here is we’ve just dialed down the affinity of the trivalent client, and we’ve dialed up the affinity on the monovalent client to the point where they have the same peak recruitment, okay? So basically what we’re doing is we’re making the trivalent and the monovalent have essentially the same overall binding affinity.
Jeremy Schmit (00:35:09):
And this is what we see when we do that. What happens is that the network can now specifically select and choose between these two clients that effectively bind to the exact same site. Okay, we have specific recognition in recruitment between these two clients here. And the system can choose which client it wants to recruit based on the scaffold stoichiometry.
Jeremy Schmit (00:35:38):
So that means if these clients had some sort of enzymatic property here, we could switch the functionality of this condensate just by adjusting the scaffold concentrations, okay? And I think that’s really pretty remarkable given that this is a liquid, okay? This is a liquid, it’s intrinsically disordered, yet there is enough structure in there to give you specific client recognition.
Jeremy Schmit (00:36:04):
Okay, so let’s summarize this part of the story here. What we have is we have filamentous assemblies, these are going to reduce the density of the assembly. And that gives us this sort of defect structure here, and this accounts for both the cross-linking and the ability to recruit clients. And there’s this really cool feature coming out that it gives us this tuneable and highly specific recruitment of clients.
Jeremy Schmit (00:36:31):
Okay, so let’s compare and contrast these two things. Okay, going back to the beginning, here, we came into it with these expectations that what we had was just two different manifestations of sticker-and-spacer here, and that we be able to use the same theory and just change a few parameters in the theory in order to describe what’s going on. And in fact, what we saw was anything but that. What we saw is not one, but three incredibly different network topologies here. We’ve got these bundle filaments, we’ve got these cylindrical brushes, and we’ve got these one-dimensional filaments here.
Jeremy Schmit (00:37:11):
Now more interesting than that, is the fact that these different microscopic topologies here encode these functional features to each condensate. So over here, the functional feature is the ability to switch back and forth between something that’s kinetically arrested, and something that’s dynamic. And over here, our functionality is the fact that we have this specific recruitment ability here that’s tuneable, it’s got a peak, and we can shift back and forth between clients if we so choose.
Jeremy Schmit (00:37:47):
So to summarize the story, we went into these projects looking for universal behavior. We went looking for universal physics principles, and instead what we found was structure/function. Okay, and so the bottom line here is that despite the fact that these are liquid states, there’s structure that’s hidden inside there, embedded within this apparent disorder. And this has functional features, okay?
Jeremy Schmit (00:38:17):
So my number one take-home message to you is that condensates are not just compartments, okay? These are not just a mechanism to concentrate and localize molecules, there are functional features embedded within them. Okay, and what those functional features are of course, going to change dramatically based on which condensate you’re interested in, and what its specific function is going to do.
Jeremy Schmit (00:38:43):
Okay, but then my physicist side comes back out again, and says, “Can we extract any physical principles, despite the fact that we’ve got these massively different structure-function relationships?”
Jeremy Schmit (00:38:58):
Okay, so let’s jam a square peg into that round hole and see what we can do with it. And in order to do that, what I want to start with doing is just summarizing why I find these results so surprising. Okay, now the point here is that liquids because they’re liquid, are inherently constantly dynamically changing, okay? They are rearranging, and in order to do that they have to have these relatively weak bonds that hold things together, but they have to break and reform rapidly. And the consequence of that is that liquids have a very high entropy, okay? There’s not a lot of structural order in the way these molecules are arranged.
Jeremy Schmit (00:39:39):
Now compare and contrast that to the other two phases. Okay, here’s a dilute liquid or your vapor phase as I always call it, and here’s your solid phase. And so the basic balance that we’ve got here is on this side of the scale, we’ve got these weak interactions. And so this is an entropy dominated system. And over here, we’ve got strong interactions that will overwhelm the entropy, and so we go into this low entropy state.
Jeremy Schmit (00:40:04):
Now let’s compare these three states based on their biochemical properties. And the first property we might look at is the concentration. And of course, it was realized as soon as the condensates were discovered that this provides a nice mechanism to concentrate your reagents, and low concentration is not as good for biochemical activity. Okay, so I’ve got these color coded, where blue means favorable properties, and red means unfavorable.
Jeremy Schmit (00:40:39):
Okay, so the next property we might look at is structure. And we know in biology, that function comes from structure or structure encodes function. So if we want structure in a system in order to encode functionality, this is the better place to be. We want to have these strong interactions that are going to encode structural specificity in order to hold molecules in a position where they can perform some sort of specific task. And we don’t get this property emerging in either the vapor or liquid state, okay? Now you can have activity within a molecule here, but here, I’m talking about activity that’s emerging from the fact that we’re building a super molecular complex here.
Jeremy Schmit (00:41:25):
Okay, but now the third property I want to look at is dynamics. Over here, we have no interactions, there’s nothing to slow the molecules down. So these molecules are buzzing around fast. These molecules are still moving. And based on the viscoelastic properties, they may be moving fast, they may be moving slow, but they’re moving. But these molecules here are essentially frozen. Okay, so the strong interactions that give you structural specificity are going to freeze your dynamics.
Jeremy Schmit (00:42:02):
All right, so let’s look at this high concentration region over here, where we’re building a condensate here. And what we look and find here based on the red and blues is that we have this fundamental trade off here. Okay, if I want to have a liquid phase, this is great because I have dynamics. And dynamics are good because you need to have chemical reactions, and you need to have activity going on in there. But the price you pay for that is you get an amorphous blob, okay? There’s no structure showing up over here.
Jeremy Schmit (00:42:35):
Whereas if I want to have structure, if I want to have functional behaviors that emerge from assembling something, the price I pay for this is I don’t get dynamics. Okay, if my interactions are strong enough to encode structure, and I build something that’s three-dimensional, the result is a rock. Okay, and rocks are not alive.
Jeremy Schmit (00:42:57):
Okay, so this is this basic trade off we have here. And so now the question is, is how do these two systems that I showed you at the beginning, how do they bridge this gap? How do they reconcile this fundamental trade off? And the answer is, is that they combine both strong and weak interactions together. Okay, so what you do here is you use strong interactions here in order to build something that has structural specificity. Okay, so the strong interactions in this case come from these zipper-like assemblies. So it’s actually a cooperative effect of lots of weak interactions stacked together. And what this does is it builds your one-dimensional assembly here, this has the correlations in the binding sites. And this is the thing that has your ability to recruit specific molecules. And then what you do is you drive it together by a weak association, these weak associations are these basically monovalent cross-links at the interaction sites, okay?
Jeremy Schmit (00:43:53):
And so these weak associations, these are going to be dynamic, and it’s going to allow your system to flow, diffuse and allow biochemistry to occur. Okay, same basic idea happens in SPOP and DAXX, we have these strong SPOP-SPOP interactions that build your rods. And we have strong SPOP-DAXX interactions that put the bristles onto them. And then these are going to assemble by weaker interactions into something that has dynamics, okay? And that’s true here whether you’re in the liquid phase, and if you squint your eyes and turn your head to the side, you can convince yourself that something similar is happening in the gel phase too, and basically these interactions here, although they’re strong enough to sort of freeze the system here, they’re not so strong that it’s not reversible if you change the stoichiometry and you want to go back to the liquid phase. Okay, maybe I’m pushing the model too far there, but that’s where I’m at.
Jeremy Schmit (00:44:49):
Okay, so what would this look like for a generic condensate? Okay, so let’s put together a cartoon that is complete fiction, okay. So don’t take this too seriously. But this is basically you’re starting chalkboard sketch for most condensates out there. Okay, so we’ve got some ingredients that we’re going to start with here, we’ve got some scaffolds, and here I’ve got three of them, it could be different depending on your system. And now we got some clients. Okay, so what we want to do is we want to put these things together. And our goal here is to build something, a functional assembly here, okay? We want these molecules to come together in some way that has a defined structure that it’s going to allow it to have environmental sensitivity, or specific recognition, or allostery, or whatever, okay?
Jeremy Schmit (00:45:37):
And so we’re going to build something here using strong interactions, that has some sort of functionality in it. Okay, now, the next step is we want to condense this thing into a high concentration state because high concentrations are good. And so what we want to do is we want to link these things together using weak associations between them. And in this case, I’m sort of cartooning that using these sort of blue brush-like molecules here.
Jeremy Schmit (00:46:05):
Okay, so I’m calling this the hierarchical model of condensate assembly here. And let’s go into some of the consequences of this model. Okay, so one consequence here is that condensates are going to have different characteristics at each length scale. Okay, so let’s see what that looks like. On the SUMO/SIM model, we see that on the short length scales, we get this sort of one-dimensional zipper-type morphology, whereas on longer length scales, we get something that looks like a random network.
Jeremy Schmit (00:46:34):
Okay, things get really interesting when we come over here and we look at the SPOP/DAXX thing because we can describe at least three different length scales that are relevant here. Okay, so on the short length scale, we get DAXX, that binds to SPOP here. And actually, if I look at this, this doesn’t look all that different than this over here. So I wonder if there’s some similar physics going on over there. Up here, we have this brush-like assembly, this is the thing that’s responsible for switching us back and forth between gels and liquids. And then inside here, we have this sort of amino acid level sticker-and-spacer topology here, that Rohit told us a lot about back in August.
Jeremy Schmit (00:47:21):
So what this means is that we have different properties that can emerge on each length scale that emerges here. Okay, so each one of these different length scales can have different encoded functionality.
Jeremy Schmit (00:47:37):
Also, the way we’re building this hierarchical structure is that we have to mix different energy scales together. Okay, we’ve got our strong interactions in here, and we’ve got our weak interactions at the cross-links there. Now, every time you introduce a new energy scale, you’re also introducing a new timescale because strong interactions take longer to break than weak interactions. And so what this means is that you’re going to see different behaviors, depending on whether you’re probing those strong interactions, or those weak interactions, properties change on length scales, properties change on timescales, properties change on energy scales.
Jeremy Schmit (00:48:15):
And so the bottom line is that in order to investigate these systems, we’re going to need experiments, simulations and theory that are able to probe these systems on multiple different scales and do so simultaneously.
Jeremy Schmit (00:48:30):
Okay, so let’s start to summarize where we’re at right now. So this is again, my cartoon for an in vivo condensate. And of course, this thing is very complicated. And we’ve been working as a community on this for several years now trying to disentangle what’s going on in there. And one of the most natural things you might do is when you’ve got something complicated here is try and figure out what are the most essential interactions for bringing this thing together?
Jeremy Schmit (00:48:58):
Well, notice what this model says. This model says that if I go in here and I disrupt the strong interactions in the middle, the whole thing falls apart, okay? Also, if I go in here, and I disrupt the weak interactions at the periphery of the strong complexes, the whole thing falls apart, okay? And so the analogy I want to make here is that this is a little like looking inside of a clock. Okay, if you peel the back off a clock, you see some big gears in here, you see some small gears here, you see some intermediate gears in the middle there.
Jeremy Schmit (00:49:30):
Which one of these is most important for the clock? Well, that’s a stupid question to ask because the functionality in the clock comes from the way that these gears fit together. You need the big ones and the small ones, in order to all join together in order to give that thing its properties. And this is exactly what’s going on in a condensate too. You need your strong interactions and you need your weak interactions and it’s the way that they work together and they fit together that allows functionality to emerge in these types of situations.
Jeremy Schmit (00:49:59):
Okay. So what have I shown you? I’ve shown you one system where we have this switching between dynamic and arrested kinetic behavior. And I’ve shown you another system where there is a specific client recruitment between molecules here.
Jeremy Schmit (00:50:14):
And I’m 100% sure that we are only just barely looking at the very tip of the iceberg here. There is just a huge number of different functional properties that can emerge from these types of systems. And it’s going to take a lot of system specific analysis in there to figure out what’s going on. Now, the way we figured out what was going on up here is that we were able lucky enough to get data collections, where we saw properties in there that just could not be explained by simple sticker in space, or Flory–Huggins type properties there. And we are very definitely interested in continuing this type of work. So if you have a condensate out there that’s got interesting behavior that you suspect might be functionally relevant, or maybe it isn’t. Maybe you don’t know if it’s functionally relevant. We’d love to throw our modeling techniques at it and see if we can disentangle what sort of hierarchical structures are emerging in there.
Jeremy Schmit (00:51:13):
And so with that, I’ll throw up my acknowledgment, slide. Kamal, was responsible for all the wonderful SUMO/SIM work that I showed you. Nelson has done some work in trying to look at what happens as you sort of morph this from one type of condensate to another, I didn’t show you those results today. And with that, I’d be happy to take any questions.
Mark Murcko (00:51:37):
That’s great. Thanks, Jeremy. And I’ll just note in passing that one of your collaborators listed here, Jill Bouchard is of course, the very same Jill Bouchard, who introduced you at the very beginning.
Jeremy Schmit (00:51:49):
Mark Murcko (00:51:49):
So a lot of questions have come in, some of which I think have actually been answered along the way, in some cases, actually, by other comments put in the chat window, which is great. So there were a number of questions towards the beginning about the mutations in SPOP and DAXX, and the effects that those have, I think those were mostly answered. And so I’d say we can skip those.
Mark Murcko (00:52:14):
There was an interesting question about what we know about the expression levels and the concentration of SPOP and DAXX in cells, and what effect that may have on the functions of those proteins in the cells. That may be a question you want to address what we’re not.
Jeremy Schmit (00:52:33):
That’s a question I want to address by asking Jill. She’s going to know that better than I do.
Jill Bouchard (00:52:39):
What are you doing put me on the spot for? All right, let’s repeat the question to make sure I got it right please.
Mark Murcko (00:52:50):
Expression levels of SPOP and DAXX in cells.
Jill Bouchard (00:52:54):
I’m probably going to have to call on Tanja a little bit too, here. SPOP’s usually pretty low, I want to say it’s like 500 nanomolar, is what Mass Spec was showing. One of the tricky things with the system is that we expect these expression levels actually to change upon transcriptional, or signaling cues. And so we couldn’t really reproduce that kind of behavior in cells, which is why we had to go to an overexpression model. So what we think is that they are cyclic around the need for ubiquitination, which is, that’s what they’re doing. But they are typically pretty low, and so these condensates definitely help concentrate them. And the cancer mutations additionally, and their expression levels, all are consistent with everything that Jeremy said.
Mark Murcko (00:53:46):
There was a question from Charlotte, who asked about whether the client length in the network, the SIM/SUMO network, what are the effects of client length? So for example, can the length of the client exceed the binding sites that are available?
Jeremy Schmit (00:54:04):
Right. At this point, everything we’ve seen is consistent with this sort of one-dimensional, you know. And so we have these gap defects in the model here where you can open up a free binding site. Well, the mirror image of that defect is sort of an overhang defect, where you have two molecules that are both competing for the same binding site. So one of them has to be dangling free in solution. And we ignored that at the beginning just because it’s better to start with something simple and add complexity to it, than to get overwhelmed with complexity at the, from the very beginning.
Jeremy Schmit (00:54:49):
And we’ve never seen anything that indicates that those sorts of overhang defects are present. What we have seen is that when you go to very high scaffold stoichiometric asymmetries–okay, so you got a huge excess of one scaffold over the other–eventually you can get to the point where you have scaffolds that come in sort of form T configurations where they form sort of monovalent binding at the cross-links. But we haven’t seen anything like that with the scaffolds or with the clients yet.
Mark Murcko (00:55:23):
Okay, well, that’s actually good. And that ties into another question that came in from Tanja, where she was asking about how the zippering in the SUMO/SIM system is encoded in the protein architecture, and how that relates to the spacing of the binding sites, how does that relate to the stiffness and the overall physical properties?
Jeremy Schmit (00:55:45):
Yeah. Well, the one-dimensional architecture comes from your linker length, okay? And so the question that… A molecule comes down and forms a bond right here, and now its neighboring binding site wants to find a binding partner somewhere. Now it can find the cheap date, it can find the person that is going to say yes, but it’s not exciting. It’s right next door, okay? But this costs you a lot of conformational entropy because you’re aligning these two polymers together. What you’d rather do is you’d rather search three-dimensional space; this allows you to keep your conformational entropy and find another binding partner.
Jeremy Schmit (00:56:32):
Well, depending on how long your linker is, if you have a short linker, that neighboring binding site looks mighty attractive, you’re not losing a lot of conformational entropy. But if you’ve got a long linker, that’s a huge conformational entropy cost, but you’ve got a much bigger region of space that you get to search in order to find a binding partner. So that gets encoded in the linker length. And if I recall, there was a second half to that question too that. I don’t think I got to. I’m not remembering it.
Mark Murcko (00:57:02):
The idea of the match spacing of the binding site?
Jeremy Schmit (00:57:06):
Yeah. So we’ve looked at the transition going from a random network to a one-dimensional filament. This is the work that that Nelson did that I alluded to there at the end. So we have models for that. In fact Nelson has also looked at asymmetric linker links, as well.
Jeremy Schmit (00:57:24):
And so if anybody had data that we could … The problem is we don’t have any data to compare those theories to and so-
Mark Murcko (00:57:32):
Jeremy Schmit (00:57:33):
They’re waiting for us to have some experiments that we can do some first order sanity checks on.
Mark Murcko (00:57:41):
Exactly. So maybe one last question that came in from Bede. It’s an interesting question about can dynamic and arrested states coexist, so that you could have a dynamic liquid, which contains smaller gels that are more structured. So you can think about the idea that there’s really more of a heterogeneous nature perhaps, to these phases.
Jeremy Schmit (00:58:07):
Yeah. Okay. So the answer to that question is yes. And it’s a very sort of nuanced yes, because you can have this happening in many different ways. So in Jill’s micrographs, if we looked at those, you’d see some of these droplets that were very liquid-like at one point, but then they’d have sort of spiky stuff jetting out at the other end. And so these are probably sort of systems that hadn’t quite finished equilibrating, deciding which phase they wanted to go into.
Jeremy Schmit (00:58:40):
But I think you might be asking a different question. And that is, can you have this sort of nested hierarchy of length scales? And in fact, that’s very much what I sort of have in mind, right here is that inside of a droplet here you have something that’s relatively solid and immobile, whereas these assemblies can slide past each other on longer length scales relatively easily. Or you could even go a step further and say, “Okay, what if I had a network that looked like this, and then like, liquid stuff moving around inside of it?” All this stuff is fair game. And this is all different variations on this sort of hierarchical model, that I think is going on in these.
Mark Murcko (00:59:28):
Right. No, that makes good sense. So we are out of time, and in fact, we already have started to lose some participants, but there still are a number of really interesting questions coming in on the chat. So Jeremy, I wonder if you wouldn’t mind afterwards, maybe you could take a whack at providing some answers to some of these additional questions.
Jeremy Schmit (00:59:47):
Yeah, I’d love to hang out as long as people have questions.
Mark Murcko (00:59:51):
Yeah, that’s great.
Jill Bouchard (00:59:53):
I’ll also make sure to post those on the site with the video as well for anybody wanting to get it.
Jeremy Schmit (00:59:59):
Mark Murcko (00:59:59):
I think what we’ll do is we’ll call it closed now, but we can come back to the questions later, we can put them on the website.
Jeremy Schmit (01:00:07):
Mark Murcko (01:00:07):
Because we always start to lose people after the hour. Well, this has been great. Just the sheer number of questions that are still coming in. And clearly you’ve got everybody thinking.
Jeremy Schmit (01:00:20):
Oh, my goodness. They’re like, coming in at the, by the…
Mark Murcko (01:00:25):
Jill Bouchard (01:00:25):
Jeremy Schmit (01:00:27):
Yeah, thank you for coming. I hope you enjoyed it.
Mark Murcko (01:00:30):
Very much so. Thank you, Jeremy. And thanks to everybody for attending. Okay, see you all again soon I hope. Thanks. Bye, bye.
Jeremy Schmit (01:00:45):
Okay, so where’s my chat window here? Okay. Great talk. Can you say anything about the viscoelastic properties of the assembly?
Jeremy Schmit (01:01:14):
So is David still here? Yes, he is. Okay. Well, what I can say about the viscoelastic properties of the assembly, is that I would expect them to change, to be widely varying. I think what this hierarchical picture says is that if you say, we’re going to do a microrheology experiment, you could get very different results depending on how far apart your probes are. If your probes are separated by distance that’s characteristic of the strong interactions, you would get one type of viscoelastic behavior, whereas if they’re separated on a length scale characterized by the weaker interactions, you’d get a very different viscoelastic behavior. So I think that’s what I would say is that the prediction here is that you’re going to have heterogeneous viscoelastic properties, and they’re just going to be different.
Jeremy Schmit (01:02:19):
Did I answer your question? Oh, yeah. He’s up. Unmuted.
David Zwicker (01:02:30):
I’m unmuted. Yeah, thank you.
Jeremy Schmit (01:02:42):
Okay, good. Next question. A couple questions. What are the typical binding strength of a single SUMO/SIM pair? Can you speculate/comment on if a 1D zipper model might work at concentrations close to balanced stoichiometry. For screened electrostatic interactions between the energy stickers, in this case is around 1kT.
Jeremy Schmit (01:03:03):
Yeah, so the binding energies that we got for SUMO/SIM, they are on the order of kT, they’re two to three kT, it’s something relatively weak. And of course, it depends on what units you do them in. But I think that it was something on the order. Oh, wait, no. It’s like 2.3 kT, something in that neighborhood.
Jeremy Schmit (01:03:33):
If a 1D zipper model might work at concentrations close to balanced stoichiometry for screened electrostatic interactions. So are you talking about sort of polyampholytes here? Just long charged polymers?
Jeremy Schmit (01:03:56):
Yes. Okay. Well, actually, we’re working on that right now, about what a long charged electrostatic region might look like. And so we haven’t tried using anything like a SUMO/SIM type of model on that yet, although the latest thing that I cooked up a day or two ago is starting to have some hints of that flavor. So I’m going to punt on that question right now. You’re going to have to ask me about that in a couple months.
Pradeep Natarajan (01:04:40):
Okay. Gotcha. Thanks. So I was just interested in … So I’ve been studying systems of like RNA and positively charged tracks and IDRs and trying to understand like the phase behaviors, so I guess-
Jeremy Schmit (01:04:51):
Pradeep Natarajan (01:04:51):
I was just wondering if a picture like this might be useful?
Jeremy Schmit (01:04:54):
And that’s very similar because we’re playing around with models of almost exactly that right now.
Pradeep Natarajan (01:04:59):
Jeremy Schmit (01:05:00):
So if you’ve got data there that you’d be willing to share. That’s the kind of thing that would be applicable for the types of calculations we’re doing right now.
Pradeep Natarajan (01:05:12):
Mm-hmm (affirmative). Gotcha. I actually I’m a theorist as well. I’ve just been playing around with simulations and models as well. So it’ll be nice to chat with you, sometime offline.
Jeremy Schmit (01:05:24):
Okay. Good. Yeah I’d love to. Yeah, for sure. Send me an email or let me know. And I’d love to chat.
Pradeep Natarajan (01:05:32):
Yeah, sure. Thank you so much.
Jeremy Schmit (01:05:34):
Okay, the next question is from Karen Lasker, and I don’t see her online anymore, unless she’s using somebody else’s name… Okay, Carlos. No, he just disappeared as well.
Jeremy Schmit (01:06:01):
How clustered would charge need to be to get to this point? Unit form polyelectrolyte coacervation was pretty well modeled with existing theory. I guess that’s a good point. In the system that we’re looking at right now, we have reasons to think that we’re not seeing coacervation so much as we’re seeing one-to-one complexes. And so this is, again something that’s sort of showing up due to mixed length scales. So if you just had one length scale, then the whole system just going to collapse, and you’re going to get coacervation.
Jeremy Schmit (01:06:48):
What seems to be showing up in the system we’re studying right now, is you have these strong interactions that are essentially opening up the network and preventing these things so we’re getting more complexes like this than we are like a coacervate, where everything is sort of mixed together. So I agree with you that if you didn’t stop anything, then probably, existing theories will worked just fine. But once you get into sort of these hierarchical type assemblies, I think that all bets are off.
Erik Martin (01:07:25):
Jeremy, I just want to chime in, quickly to follow up on that because the way you’re describing this sounds really interesting to me. And I was wondering if you could elaborate a bit on sort of where this discrepancy in length scales has to get to where the kind of general theories on coacervation no longer apply? So is there some kind of crossover point where you need to start content–like what sort of topology are we talking about where–
Jeremy Schmit (01:07:56):
Well, yeah. Okay. So if I’m just doing a standard coacervate, there’s sort of two length scales that emerge. One, would be sort of the radius of gyration of your polymers. And two would be, that would emerge from the concentration. So that length scale, basically is the separation between center of masses of the two molecules, okay?
Jeremy Schmit (01:08:33):
So on a coacervate, you expect that the radius of gyration length scale is going to be much longer than the molecular spacing. And so what that means is that each molecule is going to have many, many interaction partners.
Erik Martin (01:08:47):
Jeremy Schmit (01:08:48):
And so the system that we’re looking at right now, you have a scaffold here that limits the extent to which these molecules can condense. And so what that does is it forces the mean intermolecular separation to be on the order of your radius of gyration of your interacting molecules. And so now what this means is that the number of interacting partners is on the order of one.
Erik Martin (01:09:17):
Right. Interesting. I’m guessing this is not published yet, so you can’t say anything more specific about what type of architecture would result in this?
Jeremy Schmit (01:09:35):
Shoot me an email. On a one-on-one conversation, I’m happy to spill all the beans.
Erik Martin (01:09:43):
No doubt, understandable.
Jeremy Schmit (01:09:44):
Yeah. I’ll be a little bit more cautious when I’m being recorded.
Erik Martin (01:09:51):
Jeremy Schmit (01:09:53):
Yeah. Is there anything else there? I don’t see any more questions from people that are here.
Jill Bouchard (01:10:07):
Yes, and we’ll send those to you, and then you can type them up and at least put them on with the video.
Jeremy Schmit (01:10:14):
Okay. All right. Well, thanks a lot for having me. And thank you all for sticking around, and for the great questions, and this was a pleasure.
Jill Bouchard (01:10:27):
It’s been great. Thank you so much, Jeremy.
Jeremy Schmit (01:10:28):
Sure. Thank you.
Jill Bouchard (01:10:29):
See ya’ll later.
Question from Caitlyn Cardetti: Is there any salt in that mixture?
Response from Jill Bouchard: Yes, about 150 mM NaCl.
Question from Charlotte Fare: Are there mutations to SPOP or DAXX that affect SPOP:DAXX binding that are associated with disease?
Question from Xiao Yan: Can the complex phase behavior of SPOP-DAXX explain the disease mutants?
Jeremy’s Response: See comments by Sarah and Tanja below…
Response from Sarah Clatterbuck Soper: Most DAXX mutations in cancer are truncations, in sarcoma missense mutations are often associated with the ATRX binding domain or the H3.3 binding domain.
Response from Tanja Mittag: SPOP cancer mutations weaken interactions with DAXX and result in widening the concentration regime of the gel phase. It is difficult to recruit enough DAXX to make the brushes that result in the liquid behavior.
Question from William Chen: In the thermodynamic curve fit is the density of SPOP DAXX interaction sites a free parameter? And does it come out about right after fit.
Jeremy’s Response: The free parameter is the ratio of the site density in the gel phase to the density in the vapor phase. Values in the range 2-3 work well, which is exactly what you would expect going from a 1D assembly (a single rod) to a 3D assembly (a bundle of rods) given that the DAXX linkers are too short for next-nearest neighbor interactions.
Question from Giulio Chiesa: Are the expression levels/concentration of SPOP and DAXX known in cells? If so, do you if their relative stoichiometries lay in the “gas”, “solid” or “liquid” phase? Or they tend to change with cell function, visiting different states depending on the cell state?
Jeremy’s Response: I referred this question to Jill, I don’t know!
Response from Jill Bouchard: SPOP’s usually pretty low; I think it’s on the order of 500 nanomolar by Mass Spec analysis. One of the tricky things with the system is that we expect these expression levels to change upon transcriptional, or signaling cues. And we couldn’t really reproduce that kind of behavior in cells, which is why we had to go to an overexpression model. We think the substrate levels increase/decrease upon external cues. And when a substrate reaches a high enough concentration where it needs to be ubiquitinated (and degraded to pull the levels back down), SPOP then phase separates with it, re-distributing SPOP to various membraneless organelles. Since their typical concentrations are pretty low, the condensates definitely help concentrate SPOP and its substrates so they can find each other in cells. Additionally, the cancer mutations that affect substrate binding reduce SPOP’s ability to phase separate with them, which Jeremy’s model reproduces nicely.
Response from Tanja Mittag: The expression levels particularly for SPOP are quite low, certainly lower than the protein concentrations we need in in vitro experiments to reach the two phase regime with DAXX. We consider the possibility that there are additional multivalent interactions that mediate phase separation. E.g. DAXX typically localizes to PML bodies and is known to be SUMOylated. As seen in Jeremy’s other system, multivalent SUMO/SIM interactions can also drive phase separation (and these seem to be the interactions mediating assemblies of PML bodies). There could be an interesting coupling between the SPOP/DAXX and SUMO/SIM systems in cells. Either way, we know that SPOP is found in all sorts of liquid nuclear bodies in cells (and the particular localization is probably dependent on levels of different substrates) and is never diffuse in the nucleus pointing to the importance of its phase behavior for its localization.
Question from Erik Martin: Hi Jeremy, I’m curious if the defect density (and I suppose this might apply to SPOP/DAXX as well) depends on the on/off rates of the sticker motifs versus diffusion rate of nascent clusters? I’m thinking about the recent simulation work from Ranganathan and Shakhnovich
Jeremy’s Response: The defect density should an equilibrium property that does not depend on kinetic factors like the diffusion rate. This is going to be extremely challenging for simulations to handle, precisely because there are lots of energy and length scales. That will make it difficult to adequately sample the long timescale behavior, which will lead to artifacts like defect densities that depend on kinetic factors.
Question from Nitisha G.: Hello. Why can the blue filaments not bind at the gap sites on increasing the concentration ? Is it due to steric hinderance?
Jeremy’s Response: The scaffolds can bind in the gaps, but the affinity is so low that this is negligible in most cases. The concentration of soluble scaffold is misleading in this system because the large majority of it is in the form of dimers, which have no available binding sites. When free scaffold starts accumulating at concentration comparable to the client concentration, we start seeing gap binding. In our preprint we apply a simple correction to account for this, which fixes the droplet stoichiometry. This is also way the theory is systematically below the experimental partition coefficients when the poly-SUMO concentration gets above 70-80.
Question from Charlotte Fare: is there a client length in the SIM/SUMO network where the ability to partition into the system goes down? that is, can the length of the client exceed the available defect binding sites? or do you plateau?
Jeremy’s Response: By comparing the scaffold partitioning to the client partitioning, we do expect that there is an interesting dependence on the client length. It seems that at some point clients will start to act like scaffolds and the non-monotonic behavior will disappear. We have started to do some calculations to understand what is going on.
Question from Tanja Mittag: Hi Jeremy, could you comment on how the zippering in the SUMO/SIM system is encoded in the protein architecture. Does this have to do with matched spacing of binding sites or stiffness?
Jeremy’s Response: We’ve looked into this a bit. The issue is whether the system would rather form local, zipper-like bonds or form a disordered network. The zipper architecture has a high probability of satisfying bonds, but it costs a lot of entropy because the polymers have to align. The long the spacers are, the more conformational entropy it costs. However, longer spacers can also search more volume to find a binding partner. So longer spacers will disfavor zippers.
Question from Bede Portz: Can dynamic and arrested states co-exist, such that a dynamic liquid can contain smaller, more structured, gels? Or, is one phase so cooperative that once it is formed, all components will transition into that state, like an amyloid?
Jeremy’s Response: Yes, our hierarchical model predicts that dynamic and arrested states are very likely to co-exist on different length scales.
Question from David Zwicker: Great talk! I like the hierarchical picture – can you say anything about the viscoelastic properties of the assembly?
Jeremy’s Response: This is similar to Bede’s question. There will be different length scales that characterize the stronger and weaker parts of the assembly. Since each energy scale brings a different time scale, you will see different viscoelastic properties on each length scale.
Question from Pradeep Natarajan: Great talk Jeremy! I have a couple of questions: 1. What are the typical binding strengths of a single SUMO-SIM pair? 2. Could you maybe speculate/comment on if a 1d zipper model like this might work at concentrations close to balanced stoichiometry, for screened electrostatic interactions between stickers? (The energy scale of interactions between stickers in this case is around 1 kT.)
Jeremy’s Response: I believe we determined the SUMO-SIM interaction to be about 2.3 kT. It is funny that you mentioned this for electrostatic interactions, because in the last day or two we just started thinking about something similar. I had not yet made the connection, though. I’d be interested in hearing more about your electrostatic interactions.
Question from Keren Lasker: Did you simulate the system when the clients are available during the scaffold assembly? Would clients interrupt the formation of the condensate?
Jeremy’s Response: Because we were using analytic theory, not simulations, we were able to compute the long-time behavior when both clients and scaffolds are present. It would be possible to eventually reach a large enough concentration of clients that they would disrupt assembly, but we don’t see that here (it would be easy to calculate).
Question from Carlos Castaneda: Jeremy, great talk — If you could make a list of the different types of experimental data you would need to model these systems, what would you need? Binding affinities, partition coefficients, etc.?
Jeremy’s Response: Thanks for asking! The most important thing is that we need systematic measurements as a function of thermodynamic variables. Examples of good variables are concentration, pH, salt, osmolytes (urea, TMAO), etc. Importantly, we need more than one variable (or multiple observables) to construct a model with confidence. There are many, many different models that can fit a single experimental curve, so we can’t have any confidence that our model is correct unless it can explain multiple curves. Beyond those requirements, we are very flexible in the kinds of data we can work with. In the SPOP/DAXX system the modeling was done using the phase diagram (the independent variables were the SPOP and DAXX concentrations). In the SUMO/SIM system we worked primarily with the scaffold and client concentrations (as a function of scaffold stoichiometry). In my work on amyloids and self-assembly, most of the experiments are done as a function of solution conditions.
Question from Erik Martin: How clustered would charge need to be to get to this point? Generally I thought that uniform polyelectrolyte coacervation was pretty well modeled with existing theory?
Jeremy’s Response: We’ve been starting to look at electrostatic association in condensates. In the system we are looking at now, strong interactions keep the spacing between polyelectrolyte IDRs large enough that coacervate models don’t seem to work. This is a work in progress, stay tuned.