Genomically encoded oscillatory waves: Using frequnecy modulation to get streaming single cell data. # Introduction using frequency modulation of streaming single cell data. So I believe that's what we're going to hear about today. Please welcome Scott Corral. All right. Okay, well, hopefully we'll get to hear that exciting talk on spatial RNA and protein omics shortly. First, I just want to say thanks for the opportunity to speak at this meeting. I am not by any means a single cell proteomics person. I am a single cell. I have a lot of interest in single cell biology. Whoops, sorry, I'm just trying to get this working. And so I've really enjoyed listening to all the talks and learning about kind of the scope of the data that all of you are now able to generate and collect at a single cell level using this method. It's mind boggling to me. What I'm going to tell you a little bit about today is imaging-based tools that can't really necessarily capture the kinds of scope of data that you guys are able to capture using proteomics, but it does give us an advantage to look at individual cells over time as they perform various tasks. And I'll tell you about some of the imaging technologies that my group has been developing that we think can enable and improve our ability to make internal state measurements inside cells using these sort of psychedelic waves that you're seeing here. # Motivation Our motivation really is that cell biology, so the activity of cells is animated by dynamics, proteins activities being organized in space and time. So when a cell is crawling around like this, proteins are being, the activities of proteins are being organized and states, internal states inside these systems are changing over time. That's what's powering the ability of these cells to perform tasks like interact with their environment and process information. The activity of individual cells is of course integrated over much larger length scales in organisms where all of the individual cells working together powered by those internal protein dynamics organize complex processes like development. This is a video taken by my former colleague, Jan Huskin of a zebrafish vasculature development. And so within my group, our kind of overarching goal is we really wanna be able to understand how the gears of cells the internal protein machines and inner workings of cells turn in real time as biological processes unfold. Really there kind of two themes to this approach that we interested in in terms of single cell biology. What is an aspect of measurement? What are the internal sort of hidden states or activities that are driving cell behaviors we see and how are they changing over time? And secondly, we're also interested in perturbation, which is how does the spatiotemporal organization of molecules inside the cell impact behavior and what kinds of behaviors can cells realize by simply reorganizing molecules and components in different ways. # Measuring protein dynamics What I'll tell you about today are some approaches that my group has been taking in which we've been developing genetically encodable tools and technologies that generate synthetic protein dynamics inside cells, particularly oscillatory dynamics, and how those kinds of synthetic protein dynamics can actually enable a lot of progress on both of these fronts. I'm going to try to prioritize maybe the measurement aspect of this, but I saw this morning that I had like half an hour to talk, so maybe I'll spend a little bit more time on perturbation as well, depending on how things go. So to give you some context, why or how might the ability to create synthetic protein dynamics assist us in our ability to make single-cell measurements? So in my field, where we use a lot of imaging to query what's going on in cells, usually the way in which we follow dynamic cell states is through fluorescence microscopy, taking images like this. And the way we do that is we use cameras to collect the intensity of fluorophores at some moment in time. And that intensity provides sort of the direct readout of what we're interested in. Now, some issues with this approach, even though it's very quantitative in many respects, is that these units are arbitrary and they're susceptible to a lot of artifacts and unwanted effects, like photobleaching, illumination artifacts, or instrument to instrument variation. And furthermore, these signals lack metadata, which connects the source of the photons that we're measuring, that is individual cells, to their contribution to the fluorescence measurement. So there's not really necessarily an effective way to separate out or multiplex, sort of the way I observe many of you guys doing with TMT or other kinds of tags. You can't sort of stack data from the same cells in a single fluorescence channel. This method of acquiring data, having objects transmit information to us, kind of contrast with most other forms of information transmission or measurement that we make either in our daily lives. I really enjoyed the discussion about optical traps and things like that where you can acquire a lot of powerful information through studying oscillatory dynamics. # Background and patterning Usually when wireless signals are transmitting information to all the observers on the Zoom call, what's happening is that the signals of interest are being used to modulate the temporal structure of an oscillating carrier signal. And this carrier provides a very powerful information rich handle that actually allows many, many kinds of data streams to be sent or transmitted on the same medium along different unique channels based on the underlying frequencies that we assign to different centers in the population. A general interest of my group has been how can we create biochemical analogs of these kinds of carrier signals that we can potentially use to improve our ability to make single cell imaging measurements or perturbations. The approach that my group has been taking has been to identify simple systems that can generate a rich space of spatiotemporal dynamics that can be implemented at the protein level. And we drew inspiration from a class of self-organizing systems called reaction diffusion systems. These are models for how interacting chemical species can self-organize themselves through dissipative processes into either spatial or temporal patterns. The idea here is that a certain combination of reactivity and nonlinear reaction terms in the presence of diffusion can sort of spontaneously give rise to self-organizing phenomena. This is something that was conceptualized by Alan Turing a long time ago to explain patterning in animals and has since been sort of an object of study for physicists and for some time. So to find protein-based reaction diffusion systems that we can deploy for synthetic applications inside cells, we turned actually to a class of bacterial patterning systems that are found in nature that are unique to bacteria and are not found in eukaryotes. And what these bacterial pattern systems do is that they use the energy of NTP hydrolysis to organize a reaction diffusion process that can position molecular cargos in the cell So basically there a wide family of these proteins and they all work by basically undergoing a nucleotide dependent interaction with the surface that can be displaced through the activity of an activator protein that will stimulate hydrolysis. So something assembles in a nucleotide dependent manner, it recruits something that can trigger ATP hydrolysis and then those molecules fall off. This process is used by bacteria to actually generate very stunning oscillations inside themselves that they use to perform various kinds of tasks. # Oscillatory waves So if you were to look in an E. coli, which many of us just use for making plasmid DNAs or mini preps, to prepare for cell division, they instantiate these waves like you're seeing here inside themselves. What they're actually doing is they're using this oscillation that they've created to perform a computation, which is to locate their midpoint because this wave will actually have a nodal structure where the node will appear in the middle of the cell, and that will be the location where the cell divides. So they actually just use this wave phenomena to compute a geometric property. There's many examples of these systems all throughout biology. Some of them are used to segregate plasmids in these bugs or space complex organelles like carboxythosm. So there's an enormous space of these systems just sitting out there being used by bugs to create interesting dynamics. When I was a graduate student, an interesting paper was published by Petra Schwil's group reconstituting and showing that basically these two proteins plus membranes were sufficient to generate self-organizing phenomena. So if you just take an ATPase and activator from this Mindy E family of patterning systems and you throw them on a supported lipid bilayer, they'll spontaneously begin to organize into these wave structures you see. So these two proteins are sufficient to generate these kinds of phenomena. Given the minimal requirements to produce in vitro self-organization, my student Rohith wondered whether we could potentially exploit that simplicity and compactness of the system to generate synthetic protein dynamics in eukaryotes where these proteins don't exist. What he did was clone and deliver these components into eukaryotic cells. This was actually a rotation project that he started. By the end of his rotation, he had shown that we could, when these proteins go into cells, you can generate traveling waves inside these human cells like you see here. You could generate very fast standing oscillations, waves inside these human cells like you see here. You could generate very fast standing oscillations, and you can even generate stationary patterns, the kinds of things that Alan Turin was imagining. Here's this. My mom really likes this because it's sort of a leopard printed organization of these components here. And he saw all of these things actually from a single transfection of cells. It was pretty cool. And I'll explain what gives rise to the different phenomena we see in a little bit. But beyond the beauty of this, I'll explain many applications. But before I do, I just want to go on to say that since his initial discovery, we've been able to generate these kinds of synthetic protein dynamics in all kinds of eukaryotic systems. In fact, we haven't yet had one that hasn't worked. So I had a student that wanted to do high throughput screening in yeast. It's very straightforward. You put them in the yeast, off they go. We were working with a where we can put these into primary neurons where now the waves propagate all through the neurites up and down the dendrites. It's very fun to look at. But we've also started to put this into more multicellular context. So we can put these components into tumor organoids where now each individual cell within that structure is beginning to produce these kinds of oscillations. And recently with a collaborator in Claire Richardson's lab, our student Nikita has put these into worms. So there's a C. elegans now where we have epithelial cells that have been labeled with these proteins. And to the best of our ability, it seems like cells are very tolerant of these waves running through them because they have no endogenous interaction partners in the cells we put them in. So that's all very visually interesting. We can put them in the nucleus too. I forgot to put that slide up there. But that's all very visually interesting. But the real power that we believe that these systems have isn't for making my mom feel excited about my research for the first time in my career. But it's actually has to do with the temporal structure of the fluorescent signals that we're able to observe from these proteins. Rohit actually had an NMR background. He noticed when we were looking at these videos that if you look at the time trajectory of any individual pixel over time, you'll get this really, really nice oscillation. So this isn't sort of like a bursting pulsatile pseudo oscillation, this is a very consistent one. As an NMR person, he wanted nothing more than to apply mathematical techniques for analyzing oscillating signals like FFTs, Fourier transforms to these so you can take pixel level data and map it from the time domain into the frequency domain. What happens when we do that is that you get basically a very crisp solid peak that arises from these oscillations within that cell. If you apply that technique and cast it over all the pixels in the image, individual cells like this little pair of recently divided cells up here will light up at a very specific frequency slice within the transform image time series. What this means actually is that if we have different cells in the population oscillating at different frequencies, then they will appear on different slices in this when we forward to transform these images. Rohit developed a technique based on color coding these cells by frequency to help you visualize the output of that, where now many different cells in this population are being imaged in one channel, but the signals are being mapped and colored by the frequencies of the cells that are there. This has a variety of interesting applications, which is that we can actually sort of see through cells that are overlapping with one another. So fluorescent signals from overlapping cells contain multiple frequencies, which we can deconvolute in the frequency space using Fourier methods. This gives us a very interesting sort of approach in which instead of the photons that we collect being agnostic to the source of where they came from, The cell's own biology is sort of coupled into the fluorescent signals that we measure, giving us this unique cell-specific fluorescent signal that we can track and analyze. This is something where now if we think about some of those neurites, what we've been able to do is do things that are sort of like brainbows type of imaging, where individual neurites that are highly overlapping or complex can be mapped into this frequency space to enable things like neurite tracing in very complex overlapping environments. So with some amount of time that I have, and I really don't know how long I should go, I'll tell you a little bit about the goals we have of sort of turning cells into streamers using these sort of synthetic protein wave approaches. # Platform So I think I'll briefly talk a little bit about the platform that we developed to be able to control the kinds of oscillations that we generate, because in order to use that sort of technology, we really have to understand how to control the frequency and amplitudes of the waves that we can generate. And then I'll And I'll tell you how we've used that knowledge to design circuits. So these are biochemical engineered versions of these components that enable us to stream interesting kinds of internal state data about the cell using either AM or FM modulation approaches And then if I have any time left I also tell you about how we been sort of using this approach in reverse, sort of inspired by my son's recent interest in RC cars, where we can perturb or target cellular processes and with this inside the cell with these oscillations to actually explore different kinds of dynamic constraints or make new kinds of dynamic measurements inside cells. And if you're interested, we had a paper about this last year, as well as a few new preprints that you can read to learn more about this technology. Okay, so to control the oscillations that we observe, there's really two kinds of levels that are at play here. There's a genetic level in terms of how we deliver these components, what the expression levels of those components is going to be is going to have an impact on what kinds of phenomena we see. And then there's a protein level effect. These are kind of more biochemical effects in which depending on what components we use or mutants or what species we derive from them, various reaction constants or rate constants within the reaction cycle powering this change and that changes how expression levels map to frequencies. So Rohit developed this pipeline in which he collected thousands and thousands of single cell oscillating data used that FFT approach to measure the frequency and amplitude of a wave as a function of the concentration of the ATPase or activator in the cell. And he identified some really nice trends. I guess they came, does this appear? No, okay, sort of. He identified some really nice trends. The main thing being that the ratio of the ATPase and the activator was what was setting the frequency that we observed. And that sort of makes sense. The activator controls the rate of ATP hydrolysis. So more and more of that activator will cause these waves to go faster and faster. Whereas the ATPase itself controlled the amplitude of the wave. And this also sort of makes sense because in order to produce a wave, the ATPase assembling in the first place is rate limiting in this process. And so he has sort of this little programming language that tells us how these concentrations map to the waveform that we're measuring. And this structure actually revealed some interesting things that's kind of nice. So that tells you that a wide range of frequencies and amplitudes are available to a cell, but if we generate a single copy clonal population, the frequency within those cells will actually tend to be extremely robust because even though cells may vary in their overall total bulk expression levels the fact that the frequency is a ratio metric property tends to lead to the system only being kind of buffered against all that extrinsic noise. And that means that when you follow a colonel population, like over here on the right, for more than 24 hours of continuous imaging, these cells maintain this frequency over time. And that's very helpful because it means that we can potentially use frequency modulation to encode information in these systems effectively. And yeah, we can make different clones and stuff. # Bandwidth Now I will add, and I won't spend too much time on this at this point, I don't think, but I'll add too that although this suggests that we have a wide range of frequencies and amplitudes available to us, any specific capability for using these components is constrained by this phase portrait as well. What I mean is if you wanted to have a red cell here, you can't do it because the protein's biochemistry doesn't support giving you that behavior. So working with an undergraduate in our group, we actually had Rohith and this undergraduate Tommy cloned a variety of different evolutionarily related components that varied in their sequence composition. And they tested these and built face portraits for those components and found that they could actually find a number of species where the expression levels could be the same, but the oscillation dynamics could be very different. So here's kind of a panel of slow, medium, fast, and extremely fast all being realized at the same expression levels. And they went on to show that you could kind of mix and match these different components together to create a very large bandwidth of frequencies available at a very modest frequency range. And an interesting thing that will come up later before I just move on past kind of the core part here, is that they also found that you could compose multiple activating components together into one system, where the ratio of those components will control the frequency that we observed. And this actually suggested to Rohith a strategy for building frequency modulation circuits that I'll describe a little later. So this main point I just want you to take away at this point is that this combination of expression level control with a lot of protein components at our disposal, gives us an excellent ability to control the kinds of oscillations that we can generate. # Unique frequency per cell Okay, so now how do we go from the ability to kind of give a cell a unique frequency to encoding to control the kinds of oscillations that we can generate, to control the kinds of oscillations that we can generate. Okay, so now how do we go from the ability to kind of give a cell a unique frequency to encoding data in that structure that we can recover? And again, I want you to be kind of thinking that we want to emulate or sort of try to replicate this paradigm of something inside the cell is changing that we wanna measure and we're gonna use that to manipulate the structure of these waves over time. # Digital signal processing techniques So I'll describe, I guess before I do, I just wanna mention that we've developed a variety of tools to analyze these data based on digital signal processing techniques. So when you have a time varying, we have a non-stationary signal over time, a Fourier transform is actually not the correct way to analyze these signals. So we've developed a lot of ways to apply either kind of FIR filters, Hilbert transforms, or wavelet tools to the signals that are generated by here. I won't go too much into it, but thanks to Zuckerberg's team of PyTorch nerds, we can accelerate these processes because these are essentially large tensor type structures and do this kind of image processing very efficiently now. So armed with that toolbox of computational approaches, I'll describe kind of two ways in which we can encode data. # Amplitude modulation One of these is an amplitude modulation strategy in which the idea is to simply have a second fluorescent protein inside the cell that can conditionally interact with the waves that are going on inside the cell. The idea here is that in the absence of any interaction or any signal promoting the interaction of these components, this blue protein is just gonna be sitting in the cytoplasm doing nothing. However, if a signal promotes or favors the formation of an interaction between two proteins inside the cell, as many signaling processes do, this blue protein will now bind or coassemble with the wave components. Now that blue protein will begin to co-oscillate with the wave that we've generated. What this means is that the blue protein now is a second signal that we've generated in the cell, whose amplitude is gonna be modulated by the level of protein-protein interaction that is occurring between those two components. We kind of like this because it's a very biochemical, we're reading and gonna read out the actual biochemical interaction between these two components directly And so I gonna present an example that we done this now for many different systems where we took a sort of model kinase signaling pathway protein kinase A which is activated by cyclic AMP through G protein coupled receptors. And so there's a pair of proteins that interact in a phospho specific way. And so one of those, the FHA domain will be fused to our wave generating components. And then the substrate for pKa will be fused to a second protein. So if kinase signaling is occurring, these proteins will interact and we should see co-oscillation. So I'll show you an example here of how what this data can look like in practice. So the top is a carrier signal. So that's an oscillation that's always running in these cells at all time. And the bottom is our, what we call the data line, which is the blue protein whose co-oscillation we're going to be looking for. And at a certain point in time indicated by this graph, we add a GPCR agonist called isoprenolene. And when we do, you'll see the blue protein goes from being non-oscillating now to co-oscillating very strongly with these components. So we're seeing that interaction between those proteins visualized for us by these wave structures. Now, what's really nice is there's a lot of tools from electrical engineering we can just borrow from to quantify this co-oscillatory behavior. So every pixel in that image that we collect provides a readout for the interaction that we're interested. So we have a carrier signal on top, and below we have this data line that upon the addition of isoprenolene begins to immediately co-oscillate. I realize for this audience maybe some time scales would be helpful. We're talking about that whole trace there is about three minutes of video. So you see it almost instantly upon the addition of this compound. And then there's typical techniques of using power ratio in between these signals to quantify at a pixel level everything that's going on. This turns every cell in the field of view here into a display where when we add isoprenyline, cells are going to light up in this blue channel as their signaling activity goes up. And then we can analyze using a variety of cell segmentation things, the different kinds of response profiles as we see. What's also nice about this is there's a lot of ways to calibrate these signals using, for example, soluble or cell permeable cyclic nucleotide analogs so that you can go from the amplitudes that we measure now to the true internal concentrations of the CMP that we think are going on inside these cells The system works also can be multiplexed at the level of different signaling pathways by simply putting more reader and substrate interactions onto the same protein. So we've done this for PKA and PKC simultaneously, which has been pretty cool. # Frequency modulation But I do wanna mention, so maybe some of you are thinking like AM radio, isn't that like where, isn't that why our country's in the toilet? Can we move it to maybe the FM world? And for those of you who aren't familiar, one reason why FM sort of gives us a much nicer, higher quality signal is that it's very, it shows what's called excellent noise rejection. That is, intensity fluctuations in an AM modulated signal can creep into the data and corrupt it. But if you encode information in a dynamic property alone, on this frequency itself, then it doesn't matter if the signal goes up or down in amplitude, all the information is encoded in the dynamics directly. So Rohit decided to develop what he calls these GOFM systems for frequency modulated data encoding, where the idea is to exploit the fact that we can have more than one activator in this system and still produce functional oscillations. His strategy is very simple. He'll always have the expression of one ATPase activator pair, but he'll couple the activity of a second activator molecule in that system to a biological process of interest. So this could be transcription, proteasomal degradation, any number of different things. To figure out whether this would work or how to best make this work, he did a lot of simulations. I won't really go into it here, but he basically identified kind of the composition that you would like to have in order to get a very strong amplification of small changes in protein levels delivered into a change in frequency. He also developed a way to calibrate these circuits based on measuring the frequency of all possible combinations of these components all at the same time, and built a machine learning model to invert this mapping. I'll show you just one proof of principle that we've implemented on this. So he just wanted to show for example that if you take an existing transcriptional reporter circuit in this case a docs inducible component here you can basically have docs drive expression of the faster activator component and then see the frequency modulation over time. So in the absence of any docs, you can follow cells over time and you'll see that they just kind of crawl around and they stay at the same frequency even when they divide. So you'll see them kind of looks pretty fun, but they stay blue basically in this video that I'm showing you over here. And then in contrast, if we add docs to this cells, this clonal population over here will begin to get faster and faster and faster over time as the frequency increases. He was able to quantify a variety of nice FM responses for cells with different sort of backgrounds and then use his model to infer the transcriptional states at play. These kind of AM and FM strategies for data encoding that we have, enable us to combine the power of the sort of frequency barcoding method I showed you, with other strategies for introducing, to map data onto those structures to recover. I just wanna thank some of the pros and cons of these different strategies. You know, the AM strategy tells you about two proteins that are in direct interaction. So we're really measuring that interaction as it's occurring and we're just seeing it sort of displayed for us on the system. So these are very straightforward to design. These FM strategies that I presented as well, seem very actually well suited for things like transcriptional state measurement or things where the levels are changing perhaps more slowly over time. An advantage of these FM encoding strategies is that these measurements we make really are true and independent of any sort of intensity effects. So if you measure at different magnifications, if you measure in different instruments, that frequency is a true value that is not arbitrary and that gives us a lot of ability to collect wide field or zoomed in data that can be integrated very nicely. # Ongoing work We have a lot of collaborations with people on campus. If there's something interesting you would like to measure with these, you can just reach out. We've been looking at effects of chemotherapeutics on PKA signaling in pain-sensing neurons. We have our own funding to look at oncogenic signaling in tumor organizing. And we have another migration-oriented project with Anne Hudenlacher's lab. So I think rather than talk about the last thing, I'm just going to park it here and jump. If anyone's interested, I can show you some cool things that we can do. And I will leave Mark Zuckerberg alone for a moment. And I hope at this point you've seen, while we can't, certainly can't make the same measurement scopes that you have, the idea is perhaps by integrating the kinds of large data sets that can be made from single cell proteomics, we can find sort of like Nikolai was talking about with Katie, a subset of particularly interesting measurements that we might wanna make dynamically over time. And so I think there's actually a lot of opportunity for synergy between these different approaches. And I'd love to talk more with all of you about these. So thank you for your time. And thanks for the people that gave us the money to do this kinds of research. # Q&A Questions for Professor Kuo. Q: Nice talk. Good to see this min-CED proteins after a long time. So my PhD work on the bacterial cell division. I've worked a bit on this. My first question is that, so this min-CED, though it has ATP-ADP binding site and it works, but mostly in the bacteria it works along with the FTHGD or cell division proteins. That's where it functions. The question is that when you express this protein in the eukaryote or in other cell lines, does it fold exactly the same the way it is in the bacteria? A: So are you asking if I were to include FTSZ what happens or are you just asking what? Because you express the min-C, D proteins in the eukaryote. We only do DE, no C, but- Q: Yeah, so how is the 3D structure? Because bacteria is a prokaryotic and it's eukaryote, and there is no cell organelle in the bacteria, whereas it's fully organelles in the eukaryote. How we get localized to its specific thing in the eukaryotes, how different it is from the prokaryote to eukaryote? A: Gotcha. So I think the question was about kind of the differences in the 3D structure of eukaryotic cells versus bacteria and the different kinds of lipids. How does it choose which membranes to run on and things like that. Yeah so when we started this in bacteria there's no endoplasmic reticulum. And so these waves propagate on the exterior of the cell. And so I actually wasn't sure whether this approach would work when we tried it in the eukaryotes, because oftentimes to make, you know, we thought maybe these waves would run around the edge of the cell and be hard to see. And instead we kind of saw this beautiful space filling wave occupying the cell. And so it turns out, and I didn't show any data for this on this, but they're using the endoplasmic reticulum to help support the wave propagation. And what that, it seems like the membrane composition of endoplasmic reticulum is very compatible with most of these mean and family proteins. And it actually works very conveniently for us because it causes the entire cell to light up in this way. Q: What is the rationale? What is the specialty of this min protein to select? Because there are a lot of proteins which have same kind of ATP, ADP kind of mechanism. So what is the unique about this protein to select? A: Oh, why did we choose the mins rather than like the par proteins? I think mostly because the mins were probably maybe the most well-characterized in vitro. So I feel like the in vitro reconstitution of two proteins on a membrane producing this gave us a lot of confidence that that would work. Whereas, and we've started to look into some of these, but some of the other bacterial ones use DNA as the scaffold for the supporting structure to oscillate on. And they'll have additional sort of factors that they require, like specific DNA sequences that they need to support the wave propagation in those systems. So this was kind of like the most streamlined set of components. And I guess I didn't mention this, but the combined system is only like 30 kilodaltons, so like 22 kilodaltons for the D and 8 kilodaltons for the E. So it's very compact, easy to deliver into cells. So that's another component, very small. Q: Can we use this property to isolate single cells? A: Ah, so what I want, the one thing that we can't do is make this measurement on the cell sorter. I actually feel like there's a lot of really interesting things that could be done if we were able to like hold a cell long enough to make this measurement and then sort it I had a colleague Phil Romero that made droplet microfluidic devices We had this plan actually to kind of like measure it several times as it went down the channel but then he left the university so now I have to find a new guy that knows how to do microfluidics for that. But that would be extremely cool if we could sort based on that property. Q: Yes. Great talk. Love the waves, especially the tiger waves. There are a lot of natural waves in biology have you tried to modulate them and sort of try to cancel out and- A: Yeah okay really cool idea so um i'm glad you appreciate that because i've also had like colleagues that are like there's no what's a wave like um what happens when you're in a very biochemistry RNA oriented department i actually have a colleague bill bement who studies natural roactin waves in Xenopus. So Xenopus, when it starts as like a big cell, before it gets going, it creates these cortical actin waves on its surface. They look actually almost identical to the ones I'm showing you here, but they're made up of propagating gradients of ro and actin. And we've actually been working with him to disrupt basically the endogenous roactin wave biology with the waves that we create. These are things that are actually quite easy to do because the proteins are so amenable to engineering that we can put things on them like an actin nanobody or to interact with them and manipulate those structures. So it's ongoing, but it's definitely something we're thinking of. Nikolai. Oh, there's someone in the back. Good question. Q: Thank you for the talk. Really cool that you can measure protein-protein interactions in real time directly. Is there a way to couple this system with protein-RNA interactions or with ribonucleoproteins like CRISPR systems? A: Yeah. Okay. Fantastic thing. So that's something my student Casey actually has worked on. So we can use, I believe it's called Cas13b. There's a few of these Cas13 proteins that basically the guide RNA will base pair with an RNA. She's shown actually the hardest part of that system is visualizing the RNA component to convince yourself that it's working. But yes, you can basically have Cas13B interact with RNA. So like an idea we have, or that she had, that I thought was neat, was like, maybe you can measure endogenous RNA levels inside the cell in some way But the challenge the major challenge remains visual like having a way to actually confirm that like what that make the RNA itself fluorescent And there's all these techniques people use like putting MS2 hair pins into the RNA. But a lot of them don't work very well. So it's kind of like kind of hard, but we have shown that if you transfect a fluorescent RNA in just to cheat to make sure that you can work, then you can definitely drive an interaction that way, which is pretty cool. And that creates some interesting opportunities also for localizing RNA in the cell as well. Biological processes happen in time. So I'm deeply appreciative when people invent technologies to take time into account, and they think we don't do enough, that generally biomedical research. Q: It's really fantastic work. I have a couple of different questions. The first one is if you can give us a rough sense of the order of magnitude abundance of these proteins in terms of copy numbers that you're expressing. And what fraction of the ATP plus is being consumed by that system? Is that a significant fraction of the cells' energy or totally negligible? A: Yeah, great questions. and some quite important for when you're thinking about making measurements and whether your tool is perturbing the system. So we've done a little bit of what the name of that instrument is. It's called a seahorse. It measures metabolic activity in cells. As far as we can tell, the ATP budget associated, even with very high expression levels of these components, so those of you that like the leopard print, you need to express them quite highly to produce those kinds of phenotypes. There's almost no detectable change in the energy budget to produce these waves. And I think actually this is because sometimes the hidden cost of life is mostly in terms of making nucleotides for DNA replication process or making your transcription. Towards the other point in terms of the copy number, again, it would depend a little bit on what we're trying to achieve with that. One thing that we quite like, and I have a lot of dudes in my lab that always want to push the expression levels to the highest, but my student Casey has actually chosen a more different tactic of being like, how low can I go to make these measurements? You do not require very high expression levels. So many of these when we've done, I guess maybe proteomics might be a nice way to do this. Expression levels are like well below housekeeping genes will produce nice waves. So it's actually not really about like you need a lot. it's like you need a ratio that's balanced and then they will organize pretty well. So this can be well below gap DH levels and things like that. On the other hand, the higher the amplitude of the wave, the more straightforward it is to encode more data or to give yourself more dynamic range. So sometimes there's kind of a trade-off between like how low do you go versus like how much bandwidth you get for encoding that data. Q: My second question, I have more, but I'll allow myself only one to ask. My second question is you articulated very clearly one potential benefit in application about increasing the flex within the spectral domain being able to image many more cells what other applications can this have? It seems such a cool phenomenon that they haven thought much about that you have considered for a much longer time so share with us some additional- A: Right totally so i kind of presented most of this from the aspect of this measurement idea like like basically replacing better instrumentation with wetware sort of things to take existing fluorescent tools and make them oscillating. But like another direction, and this was actually kind of the original motivator for this is to use the tunability of the system. So the fact that we can genetically encode a space or rather a particular genetic setup produces a specific waveform and we can generate a wide range of different dynamics and amplitudes. One thing that we've been experimenting with is connecting those different kinds of waveforms to endogenous proteins in the cell to see what types of dynamic or organizing activities have phenotypic consequences for cells. Actually some of this was motivated I don't know if Galeed is here, but she had a, I think, I don't know if he was a student, but he was a postdoc in our lab, Jared, who developed this approach of using optogenetics to use light to reveal these kinds of like specific dynamic constraints on signaling pathways where it's like to turn on RAS, it needs to be on for at least some amount of time. Otherwise you don't get any signal out of it, right? And I remember being blown away actually, as he's the one that taught me all some of these Fourier transform stuff, at sort of these hidden time scales or dynamics at play that are important for creating kind of switch-like behaviors in these signaling pathways. But for him, he had to do all of that light stuff, like basically like one experiment at a time, right? So it'd be like, I'm gonna flash these cells for 10 seconds on, 10 seconds off. And then it would do like 20 on, 20 on, and try all these different patterns sort of manually using light. So one idea that we're quite actually interested in is can we sort of take that concept and package it into a library to screen cells all at once for those kinds of dynamic phenotypes. That one space that we looking at and we have some funding for it or maybe we did I don't know. I haven't checked my email this morning. But I think that'll be really interesting because there's all these timescales or even steady states that are maintained at the same level, the lifetime of that state is often invisible to us. and that lifetime is sometimes important for whether or not signals pass on. Third area that we've also been experimenting with is on a more engineering side, where we have a lot of collaborators where it's like, can we use these waves to localize things like in a T cell during, or an engineered T cell during killing or things like that, or to impose new spatial temporal organization on cells for therapeutic applications. I think that's kind of like cool, though it's a little bit maybe like too futuristic, I may be at this point. But mostly we're kind of just letting the phenomena and what observables and measurements we make sort of help us see how something that's cool has utility, I think. Thank you. Let's thank Professor Coyle one more time. # Summary Scott Kuo describes how his lab repurposed a minimal two-protein bacterial “Min” reaction-diffusion system to create genetically encodable, ATP-driven oscillatory waves inside living eukaryotic cells. These waves serve as intracellular “carrier signals” whose frequency and amplitude can be predictably tuned by adjusting expression levels or using engineered variants. By coupling cellular analytes (kinase activity, transcription, etc.) to modulators of the wave properties, the same fluorescent channel can broadcast multiple independent data streams. Amplitude-modulation (AM) schemes report real-time protein–protein interactions, whereas frequency-modulation (FM) schemes provide intensity-independent read-outs of slower processes such as transcription. The resulting “cellular streaming” approach enables multiplexed, long-term tracking of single cells, deconvolution of overlapping cells in frequency space, and large-scale screening for dynamic phenotypes. The technology has been deployed in yeast, neurons, tumor organoids, and C. elegans with no apparent toxicity, and the group is now exploring therapeutic spatial-patterning applications and integration with single-cell proteomics data sets.