Scientist Stories: Michael Rape, Development Under Stress & New Ubiquitylation Enzymes

Okay, we're zooming past 50 people, so Misha, maybe we'll go ahead and get started.

Sounds good, Jennifer.

Okay, great. Well, hi everybody. It's an incredible pleasure to be able to introduce my Berkeley colleague, Misha Rappa, to you today. And I wanted to tell you a little bit about Misha's background. So, um, he's been a member of the Berkeley faculty since 2006. He's currently the Peter Hurth Chair of Cancer Biology and an investigator of the Howard Hughes Medical Institute, and before coming to Berkeley, he was a graduate student at the Max Planck Institute in Martinsried, followed by working as a postdoctoral scholar with Mark Kirschner at Harvard Medical School. He's had a long-standing interest in quality control, especially quality control of proteins and understanding cell fate determination, looking at post-translational modification with ubiquitin. Now, it's, you know, Misha to me is really just, you know, one of the reasons why we all stay at, you know, at a great research university like Berkeley because he's always done absolutely first-rate biochemistry and he studies fundamental mechanisms that lead to insights that in turn create new opportunities for therapeutics. And that'll be the topic of this discussion today. But, you know, he's— to give you a few highlights of what he's done over his career, he's worked on understanding how new ubiquitin chain types are involved in protein quality control in cells, including understanding and really discovering key ubiquitylation enzymes and substrates and mechanisms that are central to human development and disease. And most recently his lab actually discovered quality control of protein complex composition essential to nervous system development and maintenance. His work led to the first prospective development of these molecular glues that target E3 ligases that are, uh, greatly helped in opening up the ubiquitin system for drug discovery, and he founded a company called Nurix that is involved in developing some of these fundamental discoveries to, to figure out new therapeutic opportunities and approaches to drug discovery. So, Misha, it's a real pleasure. I look forward to your lecture, and I just want to remind everyone listening— you probably know this from other Zoom seminars— but it'd be great if you could type your questions into the Q&A box, and after the lecture we will have an opportunity for Misha to answer those questions. Misha, over to you.

Well, thank you very much, Jennifer. Thank you for the invitation, and thank you very much for the very, very kind introduction. I told Jennifer before that today is the only day that I'm cool to my 10-year-old daughter, not because of any of the data that you'll see, but because Jennifer will introduce me. So again, congratulations on what you've accomplished, Jennifer, and thank you very much for inviting me to speak here. I will talk about two new projects in the lab that kind of changed a little bit the direction of what we went after. Both of them I think are very important in terms of therapeutic questions in that they open up new ubiquitylation enzymes for targeting by small molecules, but they I think also reveal interesting facets of regulation, particularly in sulfate specification and differentiation. So, our entry into mechanisms of differentiation as they apply to human development and disease often is this language that cells have evolved called ubiquitylation. This is a post-translational modification that, as any reasonable modification, is put in place by very specific enzymes called E3 ligases. And the two key stories around the talk today will feature two of these E3 ligases that confer substrate specificity onto this pathway. Now together with the help of other cofactors, these E3 ligases can add ubiquitin first to a lysine residue in substrates and then often to lysine residues in ubiquitin itself to build these ubiquitin modifications that are then read out by effector proteins that basically cause a biological output dependent on this signal. This is a tightly controlled but also reversible reaction where deubiquitylases can either completely clip off the ubiquitin or edit the ubiquitin signal. And the reason for why we are excited about this pathway and why I'm still working with it is that the human genome actually encodes for many, many enzymes that either write, read, or eradicate this modification, with especially these 600 E3 ligases being in the center of our interest. Now, another reason for why this is a good entry point into understanding very tight regulation in biological systems is that ubiquitylation is a very flavorful modification. It comes in different flavors that have different consequences. It was first discovered as a single addition of a single ubiquitin to a substrate that is called monoubiquitylation. And this typically changes protein interactions, but might also affect protein activity, localization, conformation, and so forth. Now, Alex Reshavsky in the 1980s actually discovered that you can modify not only a substrate lysine, but also lysines in ubiquitin itself to build these polymers polymeric structures or ubiquitin chains. The one that he discovered is the so-called lysine 48 chain where each ubiquitin is modified on lysine 48 with another ubiquitin. So this is a so-called homotypic chain. And work by Alex, by our lab, as well as others now showed that there are 8 potential homotypic chains that can induce different outputs as the single mono-ubiquitylation. And then a few years ago, a graduate student in my lab, Herman Meyer, recognized that you can not only modify ubiquitin with a single ubiquitin, but you can also add more than one ubiquitin molecules such a moiety to create a branch in the structure. These heterotypic chains or branched ubiquitin chains again have other functions and expand the power of this ubiquitin code, the area that it can regulate quite substantially. So, key to the signaling is then that dependent on what this effector molecule recognizes, whether it binds a single ubiquitin, a homotypic chain, or a branched ubiquitin chain,, you can induce different biological outcomes. The best known and the one that is really important for the talk today is protein degradation, usually by the proteasome if you think about K48-linked chains, but also by autophagy, which often depends on K63-linked ubiquitin chains. Ubiquitylation can also change protein interactions. That might be through mono-ubiquitylation or through non-proteolytic chains. This is key to, for example, regulation of the ribosome or the spliceosome.. But it can also, as I mentioned, have impacts in protein localization, with endocytosis being the best understood example. Now, the players that really impose this code on cells and therefore allow this regulation to occur are these E3 ligases, which bind substrates in a very specific manner. And these are really key enzymes and very important enzymes for regulation. Some of them are essential, like the anaphase-promoting complex, which has been a workhorse for my lab. Which I won't talk about today. If you delete the APC, you will not get past the first cell division. What is more interesting probably in the context of the talk today is that mutations in E3 ligases cause disease on a quite significant scale. Indeed, if you take all human cancers and you look what are the most frequently mutated tumor suppressors, 3 out of the top 7 most frequently mutated tumor suppressors are actually E3 ubiquitin ligases, with one of them, this protein KIP1, being in the center of the talk today. This is— KEAP1 is a tumor suppressor whose mutation is the most frequent cause for lung adenocarcinoma. So, if you have such important and often mutated enzymes, of course, you want to target them with small molecules. And for a very long time, that was believed to be not possible. Ubiquitylation enzymes have been thought of as undruggable, but that really changed over the last few years in a very dramatic way. Most of you might have heard the idea of induced protein degradation really spearheaded by Ray Deshaies and Craig Crews, but to which we also contributed by being able to find prospectively small molecules that not only bind these E3 ligases but sandwich between the E3 ligase and the substrate to act as glues or molecular glues to induce degradation of, in this case, an oncogenic protein. So we now know that ubiquitylation is important. Very much able to very precisely control cellular regulation, and we can control it with small molecules, making it accessible to drug discovery, meaning that we are at a very exciting stage in this field where we have to understand which E3 ligases are the best to target and why would you want to target them. And the two stories I want to discuss today are two examples of E3 ligases that we have recently discovered and that we now understand in atomic detail that both, I think, are very attractive targets for this induced degradation therapeutic modality, but also reveal very interesting facets of how ubiquitylation can control signal transduction. The first is what Jennifer already mentioned, our discovery of quality control of complex composition, meaning that cells can detect if a complex that looks just fine on the outside has the wrong subunits. And the second will be our very recent observation of a stress pathway called the reductive stress response, which is basically a rheostat monitoring mitochondrial activity in cells. The most important slide probably in this talk, or one of the two most important slides, is who's responsible for it. Well, I get the pleasure to talk about it. The first project, quality control of complex composition, is really the work of an outstanding, amazing graduate student in my lab, Elijah Mina, who discovered it. And then went all the way to describing the structural basis for this pathway. Elijah just left the lab to start his postdoc in Steve Elledge's lab at Harvard Medical School. His work was founded on a pathway that we had discovered over the last few years based mostly on work on graduate student Lin Yangying and Colleen McGourdy, as well as a postdoc, Achim Boerner, where we found that there are several ubiquitylation enzymes or E3 ligases called Cullin3-KPTPT8 or Cullin3-KALCH12 which are critical for craniofacial development in human cells. They control the differentiation of human embryonic stem cells first into neural crest cells, which are multipotent progenitor cells that among many cell types also produce so-called chondrocytes, which then in a ubiquitin-dependent manner secrete a collagen matrix, which is basically the blueprint for craniofacial development. The enzyme that caught our attention specifically was this one here, COLN3-KPTBDD8, which accounts for a differentiation of human embryonic stem cells into neural crest cells. It is an enzyme that is a little bit more complex. It has a Cullin-3 scaffold, then an active site which is centered around the ring finger protein called RBX1, and the cofactor or E2 enzyme which actually carries the ubiquitin to be transferred. And it then uses as a substrate adapter a protein called KBD8. And for these Cullin-3 ligases, it's really important to note that Cells contain up to 120 different flavors of them that are distinguished by the type of adapter they are using to recruit a specific protein for ubiquitylation. For this enzyme, the key protein is— are two proteins, one called TCOF1 and one called NOLC1, which upon binding through multiple recognition motifs, they get mono-ubiquitylated. This mono-ubiquitylation again changes interactions, in this case really allowing them to build a platform onto which the cell can build ribosomes. And the interesting thing that we found was that the ribosomes that are produced in this TEACOF and NOLC1-dependent manner prefer to produce proteins that are needed for neural crest specification while they are less efficient in producing proteins used for telencephalon differentiation, a competing pathway. So ubiquitin, in a sense, helps ribosomes to specify the outcome of the protein synthesis pathway. And KPTB8 was interesting for us. This is really the specificity factor because it really defines this pathway and it's often affected in disease. For example, mutations in the substrate which would impair binding to KPTB8 account for a disease called Treacher Collins syndrome. About 95%, actually, of the disease cases are caused by problems in recognition of the TCOF1 protein by KPTB8. So if you don't recognize this, you get less neural crest cells, and the lack of neural crest cells causes this treatable syndrome. Conversely, sequencing studies have shown that mutations in the promoter of KPTBDA, which lead to overexpression or prolonged expression during development, cause melanoma. They are one of the main drivers of this disease. Melanoma is a cancer that is caused by aberrant neural crest proliferation, and so too much of this pathway causes you too much you know, to many neural crest cells, and that gives you cancer. So this suggested that you just have to get it right during development. You have to have KBD8 at the right place and at the right time with the right activity. And that to us suggested there must be regulatory mechanisms that control how this complex actually is formed. And that was really Elijah's project with which he started. His work soon allowed us to realize that the pathway that I'm going to describe is not only focused on KPTB8, but on all proteins that share this characteristic BTB domain. So, this BTB domain of KPTB8 is used to both bind to Collin-3, but also very importantly to dimerize. And this homodimerization of BTB domains is essential for their activity. There are about 100 or so BTB dimers that act as E3 ligase subunits, but there are also about 100 BTB transcription factors. Including Bcl-6 or Bak1, which are very important for development as well as for disease. And just as with these E3 ligases, the BTB domain in these transcription factors controls homodimerization, and it is actually homodimerization that is essential for activity. And so, one of the features which is really unique about the system is that they have a fairly small dimerization interface, but there are so many of these BTB proteins. So, how can you control that you form only the active homodimers and prevent formation of inactive heterodimers. And this is basically what Elijah discovered in searching for regulators of KPTB8 initially, and then most BTB proteins. He discovered another enzyme called SCF-FBXL17. And I'll very briefly describe this only with 1 or 2 experiments because that has been published now a long time ago. FBXL17, again, is the substrate adapter, which binds specifically BTB proteins. It is connected to a related E3 ligase to what you've just seen, only that you now have a collagen I backbone and again, a catalytic site here. In this case, ubiquitylation will produce a ubiquitin chain that leads to BTB degradation. And the striking thing that Elijah recognized very early on was that this enzyme doesn't recognize a single BTB substrate. So, it's not a typical enzyme-substrate relationship where an E3 ligase recognizes very few substrates. But it can bind almost all of these 200 BTB proteins. So, it seems to be specified for a domain rather than a protein. And so, that of course had to be regulated very, very tightly, and we actually had to look for quite a long time to figure out how this was. But this is one experiment that illustrates the specificity of the system and really underscores a really intriguing way of regulation. As I said, BTB proteins are active as homodimers. They are inactive as heterodimers. In mass spectrometry experiments, we found that because of the similarity of this BTB dimerization interface, they can actually form heterodimers quite frequently in cells. That is more pronounced if you would overexpress a BTB protein, which can often happen in disease backgrounds, for example. To see whether there might be specificity in recognizing homo versus heterodimers, Elijah built a module, a cassette that allows him to very quickly form homodimers or heterodimers of defined composition, basically by fusing them in a structure-informed way. And then he can ask whether these different complexes bind distinctly— differently to the specificity factor of this E3 ligase FBXL17. And what you see here is on the end is always the homodimer, and in the middle are the two orientations of a heterodimer that you can build under the same conditions. And quite strikingly, as you can see in the binding reaction here on the right, the homodimers are not recognized at all, whereas the heterodimers are very efficiently detected. So, this work, with many other experiments, as you can imagine, allowed us to describe a new activity in cells, which we called at that time dimerization quality control, and which we now describe as quality control of protein complex composition. So, FPXL17 is able to distinguish active homodimers, which it leaves intact, from inactive heterodimers, which would interrupt signaling. It only binds the heterodimers, and then it degrades the two subunits of the heterodimer. And by degrading the heterodimer, it makes sure that the cell only contains the active homodimers. And what was striking, and I will not go too much into detail about this today, is that if you lose FBXL17, you would really accumulate specifically the heterodimers, and that would disrupt the formation of both the peripheral nervous system that includes the neural crest cells that caused, you know, Arijit to start looking into this, as well as the central nervous system. This activity, while it's very important for nervous system development and for disease, as you can imagine, however, also raised a really intriguing biochemical question. And that is, how can a single enzyme, you know, first of all distinguish homo- from heterodimers? From the outside, these complexes look basically identical. Sometimes they, you know, contain the same subunits, only that you have 2 copies of a subunit here and 1 copy of a subunit there. So how can it make this decision? And then how can it remain specific because the cell contains about 20,000 inactive heterodimers, yet the single enzyme needs to be able to discover all of them. So these are questions of specificity that we hadn't seen before, and we try to understand this. Now, while we are genetics and we can do some biochemistry, at this point we realized that we are at the end of our rope and we needed to learn new technologies. And so very much to the credit of Elijah, learned crystallography. He didn't do this by himself. John Currie taught him, so he got probably the best teacher you can imagine. And he collaborated with Basil Gerber in Evan Ogalus' lab for cryo-EM analyses as well. To start with that, I want to show you the cryo-EM structure, our initial entry that he did with Basil in Eva's lab. In this case, we purified the whole E3 ligase, together with a mutant form of KEAP1, this tumor suppressor BTB protein, we had found that we can set very specific point mutations at the dimer interface that basically mimic heterodimerization. And so mutant dimers are just a much easier way to produce a substrate E3 complex for structural analysis. We would mix these complexes, we would then stabilize them through crosslinks, and then we produce an 8 angstrom cryo-EM structure, which includes now a full-length substrate bound to almost full-length E3 ligase, which we hadn't had before. And this is how the structure looks like. In blue is the BTB protein, the substrate, and quite nicely, as expected for an enzyme that recognizes the BTB domain and exerts quality control on the BTB domain, it encircles the BTB domain of KEAP1. This is a so-called cal repeat domain where the substrate of KEAP1 would be bound. This is the adaptor, SCYP1, and then we have the rest of the E3 ligase with the active site here. And this structure already gave us a few interesting highlights. One I want to mention is that substrate recognition occurs in one domain, the BTB domain, yet catalysis would occur here in this other domain. So, the enzyme doesn't bind and ubiquitylate the same domain, it does so in distinct domains. That's important for all those of you who think about induced protein degradation as a therapeutic modality. It's sometimes not as important to have lysine residues very close by where the substrate binds, because these enzymes often are very flexible and can modify residues very much in the distance from the specificity module. But of course, the most striking feature of this structure was that even though we had a dimeric KEAP1 substrate here, and even though the affinity for dimerization here is in the high picomolar, low nanomolar range, in our structure, we only found a single molecule of this KEAP1 protein. And that raised the question right away that what this quality control machinery would test is somehow related to stability or complex formation or dimerization. As hoped for, for dimerization quality control. The problem here was that an 8 angstrom cryo-EM structure, while it's really neat and nice and can give you ideas about the whole complex, doesn't provide you with enough molecular detail to set mutations and test your models. And so we had to come up with ways to investigate this by crystallography as well. So Elijah was able to do this by purifying just the FBXL17 module,, and then only the BTB domain that targets the central recognition site. He purifies them separate from another. Again, we have a picomolar BTB dimer. He mixes them and then he gets the complex, he forms the complex and then purifies it and was able to solve a crystal structure with about 3 angstrom resolution. This is the crystal structure and from the side view, and then you can also look at this from the top. And I think what you can see right away is that, again, we only have one BTB domain in this complex. So how dimerization quality control must work is through testing in some way or another complex formation dimerization. Another feature which was really nice was that the leucine-rich repeats of this FBXL17 protein really snuggle around the BTB domain, it really follows the shape of the BTB domain. And so this with many— and then it encloses it with a C-terminal helix at the end of the last leucine-rich repeat. With many experiments, we could show that really having the appropriate shape rather than specific degron sequences was key for substrate recognition in this case. So one way how you can recognize 20,000 different combinations is they all look alike, even though they might have very different sequences. The shape is pretty much conserved. And so this is the first way of how specificity is established in the system. The structure showed us right away why it binds a monomeric BTB and not a dimer, because if you look at BTB dimer shown at the right here, the green is one subunit, the turquoise here the other one, FBXL17 has the C-terminal helix or a pincher helix that really fits very, very nicely in this groove, which is usually used for dimer formation. And so you can either form a dimer or you can bind to the E3 ligase, but not both at the same time. Under-scoring, while it only binds the monomer. This, however, was not— if you would lose the C-terminal helix, the protein could actually bind perfectly well to a dimer, suggesting that what might happen is that you initially bind to the dimer and then use this helix to pry apart the two different subunits. And this is something that Elijah was able to test quite nicely. He could make BTB dimers where the two different subunits are labeled. With different fluorophores so that we could get fluorescence energy transfer. The donor fluorescence is quenched by fluorescence energy transfer in the dimer, yet if we would pry apart the dimer, then this quenching effect would be lost and you would get an increase in the donor fluorescence if this happens. And so Elijah did this in the presence of FBXL17, and indeed we see very nicely that the donor fluorescence increases, showing you monomerization. And then very importantly, if we lose the C-terminal helix, we do not see this monomerization happening. Moreover, if we use a trap— so, this could of course also happen spontaneously, and so to test for this you give the system a large excess of basically something to soak up any monomers, which are these mutant forms of KEAP1. If we do this, we also do not see this, showing you that at least there is an active contribution of the C-terminal helix to pry apart the complex and basically probe for the stability of it. So the second issue of how quality control of protein complex composition can occur is by testing the stability of the complexes. And then the final thing in this part is, of course, if you have a reaction like this, you still have to be able to distinguish a homodimer where you don't want to pry apart the complex from a heterodimer. And our structures were able to reveal this as well. When you look at the structure of the dimer, the entry into the system that Elijah also solved, we found that there was a very interesting sequence element, an amino terminal beta strand that pairs with a beta strand of the other subunit to form an intermolecular beta sheet. Now, if you look at the structure of the subunit bound to FBXL17, this beta sheet is lifted and is basically formed into this position, to now form an intramolecular beta sheet. This is a nice example of interdomain 3-dimensional domain swapping. And because this is, of course, something that is really unique to dimer formation, we asked whether this might play a role in defining whether you can distinguish homo versus heterodimers. When you actually look at structures of heterodimers, this is a model structure of two different BTB dimers, and we looked for clashes. We indeed found that most clashes occur in this domain-swapped beta sheet that we have down here. And so we asked if we would replace just these clashing residues, where we both use the beta sheet as well as a region that we call the hinge region, which is used to place this beta sheet in the right direction, could we basically prevent recognition by dimerization quality control? So the idea is you have a heterodimer, and then you swap these two clashing residues by the sequences that would normally be there in a homodimer.. And we asked whether it's still recognized by the quality control machinery. So just as before, the homodimer is not recognized by the quality control machinery, doesn't bind to it. The heterodimer is very nicely recognized. Now we introduced all these swaps, and this here is the basic swap region here. And you can see if everything is a heterodimer, but only this like amino terminal beta sheet looks like a homodimer, the system does not recognize the complex anymore. So, this became really exciting when we then looked at conservation of the structure. There are about 200 BTB proteins. This is a 5 with a 5 amino acid amino terminal beta sheet. In all of these 200 residues, these 5 amino acids are distinct, suggesting that this is a molecular barcode for a very highly, very rapidly evolving molecular barcode for dimerization. If it's fitting, if it's the same, if it's a homodimer, it can bind into this very tight position. Yet if it's a heterodimer, you would have clashes, it would be replaced, and the enzyme— the dimerization quality control machinery could detect it. So the final thing of getting specificity in place is complementarity that's defined by this unique molecular barcode. And so to summarize this mechanism, quality control of complex composition in this case works by the homodimers having this closed conformation, which is resistant to recognition. And very rarely would show this open conformation. The heterodimers, however, would be mostly in this open conformation, allowing FbxL17 to bind and use the C-terminal helix to pry apart the complex to ubiquitylate the subunit and degrade it. And of course, a monomeric subunit can then be caught by another molecule of FbxL17 to ubiquitylate and degrade it. So, this work had some really intriguing features. One of them I mentioned, that ubiquitylation sites don't have to be in the same domain where you recognize. The protein, so substrate specificity and catalytic domain— catalytic activity can really affect different target domains. That's really important for building molecules for induced protein degradation. You don't have to worry so much about, you know, lysine residues being close by. On the other hand, it's probably very difficult to predict what would be a good target. The specificity is also determined very differently. It's not a sequence as we normally have in the ubiquitin pathway, it's not a degron in this case. Its aspects of the target like domain shape and complementarity that can be shared between many different substrates to then probe the stability in the complex formation per se. And the other thing that we are really excited about is this is a concept that we now just described for the first time for the BTB domain, but there are many modular domains in our genome like lucein zippers, coiled coils, SCAND domains, CRAB domains, and so forth. We have found this happening, quality control of complex composition, also for these cases. So we are now looking very much into other examples of these pathways and whether we can discover new principles of how complex composition in human cells can be controlled. I think this adds a really intriguing layer of specificity to signal transduction networks, and we're very excited about this. Now, this was one example where we focused really on the E3 ligase and the biochemical mechanism, a more biological project now is the second example, which highlights, I think, the breadth and the depth of this ubiquitin-dependent signaling and how much we can learn from this. Also something with very high relevance for drug discovery, and that is the reductive stress response. This is work that was really spearheaded in the lab by a phenomenal postdoc, Andrew Manford, who came with the idea of studying a particular developmental system that I'll introduce in a second, and also included a Gilliam Fellow and a fantastic graduate student in the lab, Fernando Rodriguez-Paris. Fernando is now on his way out. He has graduated as well, and he will join Vishwa Dixit's lab at Genentech. So he stays in the Bay Area. Now, the question that we addressed here really comes from our wish to identify more enzymes for drug discovery. And we believe that stress or quality control enzymes are particularly good examples about this. So stress basically is an adverse condition where the cell experiences some difficult situation, for example, lack of nutrients, high temperature, you know, chemical denaturants and so forth. And then stress responses basically have two arms. There's the detection arm, which can sense the stress, and then there is an effector arm, which can help the cell cope with the stress and induce a counterreaction. Now we know, particularly from work in protein misfolding stress, that problems in stress responses can cause diseases. And often these are age-related diseases, suggesting that cells age faster if they cannot cope with stress, such as cancer or neurodegeneration. But more and more, it has been observed that if you knock out stress response elements, stress response components, you don't necessarily have an aging phenotype. What you get is a problem in early development and tissue formation, suggesting that the signaling pathways that allow cells to cope with stress are already important early in development. And that's an area that we like to look at very, very closely. Now, this is really highlighted by a very famous stress response called the oxidative stress response. This is a stress response that reacts to too many reactive oxygen species in the cell. They are mainly produced by mitochondria as byproducts of the electron transport chain. These reactive oxygen species can oxidize proteins, lipids, DNA, and cause all sorts of problems. If you have them, if too many reactive oxygen species arise, the sensor arm here is an E3 ligase called KALN3-KEAP1. That's the same KEAP1 that we just looked at in the structure. And that leads to inhibition of this ligase. If this ligase is inhibited, it stabilizes a transcription factor called NRF2, which then, as the effector arm, sets in place antioxidant signaling pathways. Now, if we have mutations, we get these age-related diseases like lung cancer, renal cancer, as well as neurodegeneration. But if you would knock out KEAP1, for example, you also have a developmental phenotype, suggesting again there is this dichotomy of a function in tissue homeostasis, as opposed also to an important role early in development. And so we wanted to understand what is the role of stress responses in development. Rather than looking at this one example of a stress response, Andrew decided to look for new stress responses in a developmental pathway that is really prone to stress, and that's myogenesis or muscle formation. This is a pathway that's quite well understood from its core components. Starting with stem cells all the way to the multinucleated myotubes. We know core regulators and transcription factors which are responsible for that. We also know that this is very much dependent on stress. So for example, if you go into the gym— or not these days— if you go running these days, you impose exercise-mediated stress, energy stress basically, onto the pathway, and that induces— that can induce myogenesis. Conversely, if you are fasting or if you have you know, diseases in order that impact your ability to turn metabolic energy into real energy, then muscle cells are among the first ones to degenerate because this pathway is blocked. So clearly this can both be activated and inhibited by stress. And so Andrew built an image-based genetic screen trying to find new regulators of myogenesis with the hope that he can find new stress response regulators. Because ubiquitin is so important in stress signaling, we focused this on the E3 ligases that I showed you before. And indeed, we got some E3 ligases which surprisingly made myogenesis better, but we also got some which really strongly inhibited myogenesis. And the strongest inhibitor was this E3 ligase, KEAP1, the regulator of oxidative stress, which if you didn't have it, you didn't get any of those myocytes in your cell culture. So it was absolutely essential. Now, as I said, KEAP1 controls the oxidative stress response. And so if you have oxidative stress, you typically get temporary inhibition of KEAP1, which leads to a peak in Nrf2 activity to produce proteins such as enzymes for glutathione production that can scavenge these reactive oxygen species. However, if you take out KEAP1 genetically, this will be a much more persistent stabilization of Nrf2. What this does, it gives you these enzymes for glutathione production or protein reduction on a much longer timescale, and that leads to persistence scavenging of reactive oxygen species. Now, it has been shown— it's not fully understood why— that reactive oxygen species are actually physiological signaling molecules. They have important roles in signal transduction. And so if you completely scavenge this, you actually impose another stress onto the cell called reductive stress. Now we ask whether it's actually this reductive stress that is important for the blockade in myogenesis. And you can test this quite nicely when you look at this pathway. If it is, you would assume that this block in KEAP1 depletion in myogenesis is due to persistent stabilization of Nrf2. If you lose KEAP1 as well as Nrf2, you should rescue this. And that was very much what we saw. We even went one step further. If the key targets of Nrf2— it's a transcription factor with hundreds of targets— the key targets are the ones that impose this reductive stress. If you would co-deplete KEAP1 with those enzymes, you might also rescue this. And in fact, we saw this as well. So, this suggested to us that it is indeed reductive stress that prevents myogenesis. And in fact, we could impose reductive stress with chemicals rather than with depletion of KEAP1, and that also prevented myogenesis. Now, oxidative stress, as I said, is very well defined. It's too many reactive oxygen species. Reductive stress is less well defined. It's basically the lack of reactive oxygen species. And that's problematic because they have signaling functions, for example, in growth factor signaling, that would then be impeded. It's actually a quite frequent situation, even though it's not so well understood. It's quite frequent. You can get it by problems in metabolism or in energy metabolism, particularly in glycolysis. But you can also have it by mutations in various proteins. Often that's mutations in KEAP1, which are among the top 3 tumor suppressors in all human cancers, but also mutations such as in this laminin A protein, which lead to protein aggregation and basically sequester KEAP1 away from Nrf2. If you have reductive stress aside from lung cancer or renal cancer, you can also develop cardiomyopathy or diabetes. So it's something that cells basically have to be able to deal with. You know, while it was very well understood how oxidative stress is sensed, even though this is really important for signaling, how reductive stress was sensed or alleviated was completely unknown. Whether there is a reductive stress response, how it would look like, was not known. And so we figured we have this, like, really interesting in vitro differentiation system, and that should allow us to find components of the reductive stress response. Basically, we hypothesized that it's not just the lack of of reactive oxygen species, which blocks myogenesis. We thought that this lack of reactive oxygen species would be sensed by a stress pathway, which would then prevent untimely differentiation. And if that's the case, you would impose reductive stress, you would not get myotube formation. If you would then co-deplete a component of the reductive stress response together with KIP-1, you might rescue myogenesis in this in vitro model. This is a simple genetic modifier screen, and it had a simple read out a single E3 ligase in our first iteration called FAM1B, which we got very excited about because we had seen it before. It was the single most striking activator of myogenesis. So if the oxidative stress response of the depletion inhibits myogenesis, if this is a reductive stress response component, it had the complete opposite effect, as you kind of would have naively predicted. Just to show you the striking phenotypes that we get here, Again, as I showed before, if you lose KEAP1 and you impose reductive stress, you do not get any myotubes. If you lose FEM1B, and so you lose the ability to deal with it in the proper way, you get actually more myotubes. If you co-deplete both of them, both of these phenotypes are perfectly rescued. So they are really opposing antagonistic activities in cells. And so this suggested to us two things. First of all, it suggested actually that if there is a reductive stress response, response, it's tightly connected to the oxidative stress response. They are not acting in isolation, they're talking to each other. And while you might have a Cullin-based E3 ligase for the oxidative stress response, I'll show you in a second FEM1B is a subunit of another Cullin-based E3 ligase that seems to be part of the reductive stress response. Of course, here we know the substrate to show that this is indeed a stress response and that it's direct and not some indirect by key signaling enzyme, we had to like figure out what is the substrate of this particular enzyme and how is it regulated in response to the lack of reactive oxygen species. And so the ligase here, FEM1B, this is our screen head, is again a substrate adapter as we had seen it over and over before. It now binds to a third type of Collin-based ligase now with a Collin2 backbone. It has the same catalytic center that we had seen before. And as for the Collin1 ligases, this enzyme machinery typically assembles polymeric ubiquitin chains that lead to substrate degradation. Now, if you want to find the substrate, you can do so by IP mass spec experiments, but that might be difficult because it's very transient interactions and it induces degradation. So, in order to catch it, we had to use a little trick. We introduced a substrate-trapping mutation that prevents incorporation of FAM1B into the full complex so that it can still bind its substrate but not ubiquitylate and degrade it. And we then looked at this part of the mass spectrometry plot that we get, which are the proteins that are enriched in the substrate trap as opposed to the wild-type complex. And there we find two very interesting proteins, FNP1 and foliculin, which form a complex and have been known to have interesting functions for muscle biology. Mutation of FNP1 changes how muscle cells can differentiate in animals. Just to show you the data that, you know, it turned out that FNP1 was the key substrate of this machinery. If we induce our E3 ligase, FNP1 goes away, it's degraded. This does not happen if we have the substrate trap, which is shown here on the right, as well as a point mutation, which took us 1.5 years to get, a single cysteine to serine mutation in FEM1B, which did not induce degradation either. When you looked at binding, they had different functions though. Wild type FNP1 was found in a complex with FEM1B. If we had the cysteine to serine mutation, it was not found. So, this is a substrate binding deficient mutant. Whereas the substrate trap bound better. This was the, you know, catalytic inactive mutant basically. And we know that this is the essential substrate because we can do the same genetic rescue experiments that I showed before. If we have FEM1B depletions, we get more myocytes in these differentiation reactions. If this is due to accumulation of a particular substrate such as FANIP1, if you deplete FANIP1 together with FEM1B, you should get a rescue of this phenotype. And again, we see this for FNP1, but we didn't see it for any of the other binding partners that we had found. So, we were able to like get this substrate for this, which is a metabolic regulator, and I'll come to what it does in a second, FNP1. But again, key to a stress response is that it happens in response to an adverse condition. In the case of the oxidative stress response, it is oxidation of cysteine residues in this KEAP1 subunit which inactivates it. So we asked, is there a way of how FEM1B and its recognition of NIP1 is regulated by the lack of reactive oxygen species? That would be key if that is a reductive stress response. And so in order to answer this question, we had to find the little triangle here, the sequence of NIP1 that is regulated by FEM1B, and then we had to learn whether this is affected in any way by the absence of reactive oxygen species. And so we took the 1,100-residue protein, NIP1, we cut it into bits and pieces. It's a little bit more difficult than I say, because if it doesn't really bind to its partner folliculin, it can be degraded non-specifically. So you have to test a lot of different combinations. But through the work of a fabulous undergrad in the lab, Karen Shi, at the time, we were able to get two mutants where we saw that both binding to FEM1B as well as degradation in the cell were inhibited. And we then asked whether these two mutants, whether they comprise a so-called degron, a sequence in FNP1 that would be recognized by FEM1B. And for this, we developed a fluorescence-based assay that allows us to test this quite nicely. So we produce a green fluorescent protein fused to this degron sequence. And on the same plasmid, we have a red fluorescent protein driven by an internal ribosome entry site. And so the ratio of GFP to mCherry that we can measure by FACS is then really dependent upon whether this degron here imposes degradation of GFP, which would lower the ratio, or whether it would stabilize it. And so if we do this in green, the green curve of the GFP to mCherry ratio gives you the normal distribution of green versus red. But if you then impose— activate the ligase, you turn it on, you see that the green curve now shifts to the red, to the left. And this tells you that we have degradation of the GFP. So this is indeed a degron that seems to be sufficient to impose FEM1B-dependent degradation. We know it's specific because both the substrate-binding deficient mutant of FEM1B or the catalytic site mutant didn't really impose this degradation. We also know that there is some degradation happening in cells under normal conditions because we can use Jennifer's CRISPR strategy to delete FEM1B, and very, very nicely we see stabilization because the curve moves to the right. And the ratios here also tell us that under normal conditions, the normal conditions we use to culture cells on plastic and in our tissue culture rooms, about 80% of FEBPNIP1 is stabilized and only 20% are degraded under homeostatic conditions, which could be shifted by deletion of the ligase. Now, we have the degron. Now, does this tell us something about regulation? And indeed it does, because you can look at the structure of the sequence of this degron, and you can look at what residues are invariant and are not conserved. And it's really basically 3 residues that are invariant across all species that have FEBPNIP1 1, and these are cysteine residues. And of course, cysteine is an amino acid that is redox sensitive, so it can either be oxidized or it can be reduced. And so we wanted to know whether these invariant cysteine residues are key for recognition. And so we use this assay that Andrew built. In solid curves, we have the normal GFP to RFP ratio. In the dashed curves, when we induce the ligase, the wild type, you see degradation— oops, sorry. We see degradation. If we mutate this cysteine residue, we get almost complete stabilization of the protein. If we mutate these two cysteines at the same time, we also get almost complete stabilization. If we mutate all three of them, we actually have complete stabilization. This shows you that these cysteine residues not only are conserved, but they are essential. And very hot off the press, we, we really have just gotten this resolved. We now understand how this looks like on a molecular level. We solved the crystal structure of FEM1B bound to FNP1. And it actually shows you a different type of protein-protein interactions that, you know, explains the redox sensitivity. Rather than having amino acid side chains of one partner binding other amino acid side chains of the second partner, in this case, it's these three cysteine residues that we found in FNP1 together with the cysteine residue that we found before in FEM1B, chelating two zinc atoms. And the two zinc ions here really act as molecular glue that tether the substrate, in this case, to the E3 ligase. And so one thing is, of course, this reaction is zinc-dependent. We can show this very nicely. But on the other level, you can only chelate zinc if you have a reduced cysteine, not if you have an oxidized cysteine, showing you why only the reduced FNP1 would be recognized and not an oxidized form. So this gave us the triangle, and it immediately raised the possibility that the oxidation state of NIP1 would control recognition by this E3 ligase. So we can test this by taking a peptide that's just the triangle, basically, we label it with a fluorophore, we titrate it into reactions that contain FAM1B, and so we look to binding of the peptide to the E3 ligase to fluorescence polarization. And in green is the curve as you would do it as a good biochemist with a reducing agent present, and you see very nice binding with a pretty high affinity for an E3 ligase of about 19 nanomolars. On the blue curve, we for once forget that we are good biochemists and we leave out the reducing agent, and it takes about 30 seconds to completely lose binding. So, these are extremely reactive cysteine residues that like to form intramolecular disulfide bonds. If we now take this peptide with the intramolecular disulfide bond and we add back a reducing agent, we get perfect binding to the E3 ligase. Again. So, reversible oxidation inhibits binding, and in fact, we could show that in cells about 80% of the NIP1 protein is oxidized at any given stage, and so is prevented from binding the E3 ligase. Of course, reductive stress removes the agent which imposes this oxidation, and so we asked, does that mean that reductive stress can induce degradation of NIP1? And indeed it does. We can impose reductive stress by various means, and When we do so, we see that this shift— this curve that I showed you many times before now shifts to the left, because the cysteine residues are reduced, the protein is recognized by the E3 ligase, and it's now degraded. If we have oxidative stress, the remaining 20% that has not been oxidized yet is now also oxidized, and that leads to stabilization of the newborn protein. And so this now reveals the regulation of the reductive stress response. In this case, it's oxidation not of the ligase, of the substrate that recognizes or gates recognition by the E3 ubiquitin ligands. And so the last part that we had to figure out is what is the effector arm of this complex. In this case, it's a transcriptional arm where you use enzymes that would scavenge reactive oxygen species. For the reductive stress response, NIP1 is not a transcription factor. It's a metabolic regulator. It had really nice activities in controlling mTOR, as shown by my colleague here in Berkeley, Roberto Sancho, But we found pretty quickly that mTORC signaling is not important in this pathway. So, it was a different activity. And it turned out that the activity was linked to mitochondrial activation. And with this, just to remind you that mitochondria that we see here in electron microscopy pictures are the major source of reactive oxygen species in cells. This is the textbook picture as you always see them, and that's what we saw in our transmission electron microscopy. If you deplete SNIP1, you get actually larger mitochondria. They are very well organized in their cristae, but it's not a drastic phenotype. The drastic phenotype comes into place when you deplete the ligase, and now you would stabilize the SNIP1 protein. Suddenly, all of your mitochondria turn pitch black. That's a phenotype that was first described in the 1960s as matrix condensation, and it suggests that these mitochondria have run out of metabolites for the TCA cycle, and so for oxidative phosphorylation. In fact, the second most obvious phenotype is this onion swirl mitochondrion, which is an overproduction of cristae as a response to a lower activity of their electron transport chain. And the third frequent observation we made is the induction of autophagy, mitophagy, which often happens when you lose the membrane potential of mitochondria, which in fact is the case under these conditions. And all of these phenotypes were really only caused by stabilization of the single substrate FNP1 because we can rescue them by co-depleting FNP1. We can show that this is really happening on the level of mitochondrial reactive oxygen species. You can visualize them in cells with fluorophores. If you deplete the ligase, so you stabilize this protein FNP1, cells do not have active mitochondria. They do not produce any reactive oxygen species. But that's only really dependent upon NIP1 accumulation. If you would get rid of NIP1, you would get actually a little bit more of them. And then we finally wanted to know how this actually happens and realized that I should come to an end. So quickly, the mechanism, we don't fully understand it, but by metabolomics, we found that when you deplete FAM1B, you stabilize NIP1, which is the second one here. You basically lose metabolites in cells that are transporters of other metabolites into mitochondria. Mitochondria. If you get rid of NIP1, these metabolites are upregulated, allowing the mitochondrion to import other cell— other metabolites into cells. So we believe that degradation of NIP1 basically is a mechanism by which the cell can regulate metabolite import to bring especially glutamate into the mitochondrion so that the TCA cycle can run highly active to produce the redox equivalents that you need in order to run oxidative phosphorylation and thereby produce reactive oxygen species. So to put everything together in a model now, the reductive stress response, not only does this exist, but it's a really cute mechanism. If you have normal mitochondria that produce ROS, you will have oxidation of these cysteine residues. They form these intramolecular disulfide bonds, and that leads to stabilization. When FNP1 is stable, it prevents metabolite import into mitochondria. If you have reductive stress, meaning your electron transport chain is inactive, you run out of reactive oxygen species, these disulfide bonds will be reduced, giving you the reduced form of the substrate, which can now be recognized by the E3 ligase and can be degraded. The degradation of NIP1 therefore releases this brake to metabolite import, allows cells to shuttle metabolites into mitochondria to jumpstart the TCA cycle, and thereby to jumpstart production of both ATP, of course, but also reactive oxygen species. So in this case, the reductive stress response really is a mitochondrial rheostat. If you have reductive stress, it senses ROS depletion that leads to degradation of this gatekeeper of the mitochondria, and in turn, it activates reactive oxygen species production that directly counteracts the stress that you have. But it also works in the other way. If you have hyperactive mitochondria that produce oxidative stress, they have too many ROS that are produced by them. This leads to further stabilization of NIP1, shutdown of mitochondrial metabolite import, and so it removes sort of the source for this oxidative stress. So it also allows the cells to balance the levels of reactive oxygen species. What was also striking actually was the fact that now we have 3 pathways in the cell that basically control its oxidative state, the oxidative stress response, that I mentioned before, the hypoxic stress response that was discovered by Bill Calin and others, and now also the reductive stress response. And when you look at this, it's really striking how they have co-evolved. They all look very much the same. They're all dependent upon coloring ligases that control degradation reaction. They directly counteract the source of the stress, but they're also then regulated by the oxidative state of the cell. Quite nicely, showing you how ubiquitin is really a very powerful mechanism to control stress signaling and development. And so with this, there are also many, many things that come about that we are interested. Of course, we want to know where and how does ubiquitin degradation regulate mitochondria. What's the molecular basis for this? We're trying to understand how this affects metabolism in development and disease. For this, we just generated FEM1B knockout mice, and we want to know whether they rescue, for example, the keep FEM1-driven tumors, as you would expect from the genetic interactions we see in our tissue culture cells. But we are also very much interested in the effects of aging onto aging or other diseases, because what we basically do by knocking out FEM1B is caloric restriction by genetic means. And then finally, we like to actually turn this discovery into direct E3 ligase activators, small molecule compounds that would be very useful in the clinic. We have an E3 ligase that has a catalytic cysteine residue where it normally binds substrates. That's quite unique, and I think offers many opportunities for drug discovery. And so, with this I would like to close. First of all, the two key players here really, Elijah was the driver of the dimerization quality control pathway, Andrew together with Fernando for the reductive stress response system. None of this would have been possible without amazing help from colleagues at Berkeley and beyond, including John and Eva for the structures, and others for developmental biology questions. And then, of course, the funding that made this possible. So thank you very much for your attention. And if you have questions, you know, I'd be happy to take them if you put them in the Q&A.

Thank you.

Amazing.

Wow. That's really incredible work, Nisha. Very exciting and interesting. And I have tons of questions. Well, while people are thinking, so I guess I had one kind of more detailed one and then one more sort of big picture. So the more detailed one is I thought that, you know, the beautiful mechanistic work that you have all done to understand these, you know, these BTB complexes, it made me wonder, and I think you and others have done work on this already, but I'm curious if you could comment on the ability to manipulate those complexes using peptides or even small molecules, because I mean, you know, some of your own work has shown that you can, you can, you know, actually do a remarkable amount of manipulation with peptides.

What are your thoughts about that? Yeah, we, with peptides, we haven't tried it there because you need the dimerization interface is a pretty long helix, which should be able to compete to some extent. We haven't tried that. That's a very good idea. But what actually happened quite random— not randomly, through very nice research done at Boehringer Ingelheim, they found a small molecule for Bcl-6, a BTP transcription factor that can bind to the structure in an allosteric site, and that induces degradation of this machinery. So I think you can either bind at the interface to sort of a little bit destabilize it so that it now can look like a substrate for this machinery, or you can even have very nice allosteric effects. I think that's really important because there are oncogenes like BCL-6 where you might actually harness this machinery that we discovered for drug discovery. So we're very excited about these possibilities.

Yeah, yeah, yeah. And also, and then I guess my other question is sort of thinking about the the significance of these pathways. And you guys have shown that it's really important for different kinds of cell differentiation and especially like early development. But on the other end, it also appears that, you know, you might be able to manipulate mitochondrial function that could be interesting in aging or age-related, you know, degeneration of tissues and things. Is that a realistic sort of way to think about it?

No, absolutely. I mean, the beauty— there are two beauties. One is caloric restriction has gotten a lot of attention in terms of aging. But it really shut down mitochondria specifically, where mitochondria have many functions other than ATP production. And you don't want to shut them down necessarily in aging. Whereas ROS generation, that's something you want to shut down. The beauty of this system is we can now selectively downregulate ROS production while having the cells still produce amino acids, lipids, heme, and so forth that they would need. So I think that's a realistic, really, really nice idea. That's why we have very heavily searched for inhibitors of FEM1B, which would, you know, allow us to really tune mitochondrial activity with small molecules. And through the Drug Discovery Center here at Berkeley, we found some really exciting possible compounds. So I think we can exactly test this. This would be a a drug given and not like not eating, which is probably also better in terms of lifestyle. But the most exciting thing really is in terms of neurodegenerative diseases, because for the first time, we're now able to manipulate mitochondrial output rather than just throw away a mitochondrion that's not working all that well anymore. And so in a project that we're trying to start now with Baris Bingöl's lab at Genentech, we're actually directly testing whether we can tune mitochondrial activity through this pathway and change the output of both ALS disease models and Parkinson's disease models, which both are pathologies dependent on mitochondrial dysregulation. So that's exactly the next stage of this project, which is very, very exciting.

Yeah, indeed. Melanie Ott is asking, she says, beautiful talk. Any direct regulatory role of oxidation substrates or others in the other two oxidative stress responses or only in reductive stress?

Can you repeat this?

I don't know whether I got the question. Yeah, she's wondering if there's any direct regulatory role of oxidation in the other two oxidative stress responses that you showed or only in reductive stress.

So both of the other pathways are controlled by oxidation, and I think they crosstalk very, very nicely. In the term of the oxidative stress response, if you have reductive stress, you know, that clearly will not oxidize KEAP1. But for the hypoxic stress response, what is really exciting is that the enzymes that cut— so this is regulated by the transcription factor HIF-1 alpha— is oxidized directly through an enzyme that needs alpha-ketoglutarate as a a cofactor for the oxidation step. It's a proline oxidation. And alpha-ketoglutarate is a really key TCA cycle component. And the way how we can impose reductive stress is remove alpha-ketoglutarate from cells. If we add more alpha-ketoglutarate to cells, we impose oxidative stress. So actually, I think reductive stress will have direct impacts on the hypoxic stress response. And all three of them are very tightly controlled with each other. We see this also on transcriptional profiles. So it's really that the levels of reactive oxygen species in cells are controlled by 3 stress responses that act at the same time and balance themselves out in very intriguing ways. So by changing one, you will affect the other pathways as well.

We have a question from Adi asking, you mentioned about flex— I guess flux of other metabolites.

Is there a direct interaction with a particular transporter? Yeah, this is a super, super, super question. The reason for why I think super, super, super is also I didn't really tell you everything about it, but our structure suggests that the ligase has a metabolite binding site. And there might be direct control. And there is a very obvious system that we like very much. This is the malate-aspartate shuttle. The reason for that is that that defines the ratio of NAD+ to NADH, which is as important for defining the reductive or oxidative potential of a cell as just reactive oxygen species. So yes, there will be— we're like trying to figure out all the details of this as we speak.

And then we have a question from Lorenzo from the Flor Lab at UCSF asking, is it common to have zinc-dependent interactions for other proteins exhibiting redox-dependent degradation?

So that's, you know, I'm not an authority on this. This is the, I think, third structure that might have is now solved, but it's the first time that we've seen a zinc-dependent protein-protein interaction where the zinc really is at the interface of two proteins and really defining the interaction. So I asked as many other crystallographers I can see, and they haven't seen that before. Zinc can induce conformations for domains like the ring finger that then allows protein interactions to occur. But at the interface, that really hasn't been seen. And that it's two zinc atoms at the same time, that's another feature that's really interesting, which might indicate that maybe this is also some zinc sensor of some sort that, that we are in, you know, that might be intriguing. A little bit— it's a little bit more complicated. I mean, I had to rush through it. We can strip away the metal and put back metals, of course, to really ask what is the key Zinc is the best, but copper actually works also to some extent. And copper is interesting because that's very important for mitochondrial function as well. So there is definitely, I think, biology that is in the metal biology realm that impacts that signaling system as well.

Yeah, yeah, wow, very, very interesting. OK, any other questions? I'm just checking. I don't see anything else in the chat box right now. If not, Misha, I want to thank you. It's really always a pleasure to hear about your beautiful work, and we're delighted to be able to catch up with you even by Zoom. Really, really incredibly interesting talk and lots of implications. Very complicated, and you've done such a beautiful job of dissecting it biochemically. I'm really delighted.

And impressed. Thanks so much. Thank you, Jennifer. Thanks all. Thank you. Thank you.