Structure Club
Structure Club invites scientists to present one of their papers to a broad audience. Listeners will hear about cutting edge science from the scientists themselves.
Structure Club
Danielle Grotjahn and Benjamin Barad
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Danielle Grotjahn and Benjamin Barad present their paper, "Surface Morphometrics reveals local membrane thickness variation in organellar subcompartments"
https://rupress.org/jcb/article/225/3/e202505059/278755/Surface-Morphometrics-reveals-local-membrane
Welcome to Structure Club. I'm Scott Stagg. I'm a professor at Florida State University, and I use cryoelectron microscopy to determine the molecular mechanisms of membrane remodeling.
SPEAKER_04And I'm Ashun Francis, assistant professor at FSCU. My lab focuses on understanding the structural basis of virus source interactions. Together, Scott and I created Structure Club as a journal club podcast and YouTube channel where the papers are given by the authors themselves.
SPEAKER_02Today's speakers are Danielle Grotjan from Scripts Research and Benjamin Barad from Oregon Health and Sciences University.
SPEAKER_04Welcome, Danielle and Benjamin. Nice to have you here. Thank you.
SPEAKER_02So Danielle got her PhD in biophysics at Scripps Research with Gabe Lander. She completed a brief one-year postdoc with Grant Jensen before starting her lab as a Scripps Fellow in 2019. And she's been there ever since. Benjamin got his PhD in biophysics at UCSF with James Frasier and then did a pro a postdoc with Danielle at Scripps. And he finished in around 2023. And he just started his independent lab at OHSU in uh 2024. So they are going to be uh talking with us about their uh recent paper, Surface Morphometrics Reveals Local Membrane Thickness Variation in organelle subcompartments. And the manuscript just came out in the Journal of Cellular Biology in December of 2025. Danielle and Ben, the mic is yours. Thank you.
SPEAKER_00All right. Great. Thank you so much. So, yeah, I was mentioned before, uh, I started my lab in 2019, and I was really so fortunate as a young PI to recruit not one, but two all-stars right away. Um, Michaela Medina, who you'll hear more about, was my first graduate student. And then, of course, of course, Ben Broad, uh, the other lead author on this paper. And when it was just the three of us, we were just collecting a lot of tomography data focused on really looking at organelle ultrastructure and all these different stress conditions. And at first, honestly, the goal was just we need to get beautiful tomograms. Um and so uh once we had kind of mastered that aspect of the workflow, that's when we realized the real challenge in the fun was beginning. Uh, because what we realized was that there was uh a gap in the field. We didn't have great ways to extract out biologically meaningful quantitative information about membrane ultrastructure. And that was a real challenge for us because we were interested in looking at all these different conditions and how membrane remodeling might be coupled to those conditions and then connect that to function. And so most of the existing approaches at the time relied on manual measurements, which just weren't going to be at the scale that we needed to look across all the conditions we wanted to look at. So that realization really planted the seed for surface morphometrics, which uh Ben was the real um like developer of along with Michaela.
SPEAKER_01Yeah, so the the surface morphometrics pipeline, which this paper is sort of uh an expansion of. Uh we first wrote about in in JCB two years earlier. Um, and and sort of at the core of this toolbox was software that takes segmentation, which decides where membranes are, uh, and builds these really detailed models that we make out of triangle meshes uh of every different membrane within a tomogram, and then can make sort of this very classical suite of geometric measurements. So we can ask about the distance between the inner and outer membrane, uh, for instance, and you can see these blue splotches correspond to where the membrane enfolds into cristae on the inner membrane, so it's not opposed perfectly. Um, we can ask where membranes are curved, and we're showing that in red in the middle there. And you know, membrane curvature is one of these really interesting features that happens a lot more in cells than it does in a test tube, and that's because of both proteins and lipids that drive curvature. Um, and as well as asking questions about the relative orientations of different membranes with regards to each other and how that might impact things like membrane-membrane contact sites. So this toolbox was sort of like the beginning of the equivalent of an atomic model, but for membranes.
SPEAKER_00Yeah, and we, as Ben mentioned, we really started off looking at bulk membrane geometry, but our long-term goal and vision was to always take this to more local changes in membrane structure, which is where tomography, I think, really uniquely shines. Um, so that led us to ask whether these surface models could be used to support more precise spatially resolved measurements. And we actually took inspiration from line scan intensity analysis, uh, both in fluorescence microscopy and from similar intensity profile that had already been used in uh cryo yem, where the distance between bilayer leaflet peaks is used to estimate membrane thickness. What hadn't been done yet was extending this kind of analysis in three dimensions in the cellular context. So we really wanted to ask in this current paper that we're discussing uh whether we could make this possible with our surface morphometrics tooling. And so what you can see here is that um we have in our data, um, these arrows are pointing to um density not just for two distinct membranes, but for the leaflets actually of the lipid bilayer. And this is like for a tomographer, kind of like your low-key indication that your data is pretty good when you can resolve uh this uh the leaflets. And when you um it's also important to take um acquire your data a relatively low to focus. And so uh what we could do now is um you'll see this uh surface mesh reconstructions go right down pretty much the center of the membrane in these cases. And so uh we developed approach to generate line scans um across the surface mesh on an individual triangle basis. But Ben's gonna get into why we have to sort of come up with uh new strategies because the data is noisy if you just look at individual line scans.
SPEAKER_01Yeah, so um, you know, this is you actually see things like this when you do fluorescence as well, but um, in any given voxel within the tomography data, you can see, even though you can sort of squint and see the lines, you can see there's lots of bright spots and dark spots within the data. And this is fundamentally unavoidable because of the signal to noise. We're only collecting a few electrons per tilt uh within a tomogram, and only maybe 120 electrons per square angstrom total across the entire tomogram. And so we're very dose limited when we collect this data. So, what we can do is for every triangle, and there's lots of triangles even in this um visual here, uh, you can uh make your measurement where you scan along the normal angle to the uh to the mesh and sort of measure the density. We just interpolate the density at the voxel every quarter nanometer. Um, those we can get the measurements. They are unsurprisingly kind of all over the place. And so the first thing we did was we just took the entire inner membrane, for instance, and we averaged it. Um, and that's what you see in the blue line on the right here. Uh, and the blue line is the average of the data across the entire inner membrane. You can see there's little dips, those actually correspond to the sort of white spots that you get from defocus when you have an underfocused image. Um, but then you also have the peaks that correspond to the two phospholipid head groups, and there's just a little bit of a dip in between the two peaks that correspond to the tails that have slightly weaker density. Um, and so because we can see these sort of two peaks clearly, we can fit a pair of Gaussians to the data. And we this is sort of a very simplistic treatment, but it was robust enough that we could do it with our noisy data. We can fit our two Gaussians, which correspond to the two head groups, and then we can measure the distance between those, and that's what's shown in the in the vertical uh sort of green lines that span shows the measured thickness of the membrane uh for this inner membrane.
SPEAKER_00Yeah, so of course, with this type of tool, um, what you're really interested in doing is comparing it across different conditions or features. And so one of the first um comparisons that we made was just to ask do different organeller membranes within the same cell have different um global membrane thicknesses? And uh what was pretty striking to us was that um this they did appear to have slightly different uh thickness measurements. And consistently we saw that the um outer mitochondria membrane shown here in OMM consistently has um thinner membrane relative to the other organeller membranes, such as the inner membrane within the same mitochondria or the endoplastic reticulum and vesicles. And we also looked at this um in yeast cells, which in the same uh field of view of a tomogram, you can actually get in a lot more um diversity in organelle membranes. And consistently we found um that this outer mitochondrial membrane has uh thinner. And so I think this really is already opening up, like was already exciting to us that this is um opening up some cool biology and some hypotheses that we can test. But something that um, you know, my lab were full of mitochondria. We love mitochondria. So for us, um, what was pretty cool was that our surface morphometrics 1.0, um, what we had shown with that is we could actually use our very precise calculations of outer to inner membrane distance to automatically subclassify the spatially and functionally distinct subcompartments of the inner mitochondria membrane. Um, we have names for these, like the inner boundary membrane shown in purple, the crista junction, which is sort of this transition period that tends to be highly curved, and then the crystal body where the components of oxidative phosphorylation machinery are generally housed. Um, and so this was a beautiful example of how we could use this geometric um calculation distance-based filtering to automatically classify these and then ask how thickness varies throughout those. And what we found was that um the crystal body tended to have um thicker membranes uh relative to the junctions in the inner boundary membrane. And so this again opens up a lot of new um hypothesis and models of what you can um predict might be dictating these variations.
SPEAKER_02Danielle, can I ask you all a question? Sure. So um uh how many points are represented in one of these violin uh plots?
SPEAKER_03Yeah, then you want to go over the three.
SPEAKER_01Yeah, I don't remember the exact numbers. I think it's about 15 mitochondria for each bar um that we have in the paper the exact numbers. But each one of those mitochondria, we're taking the average of typically around um between 100,000 and 1 million square nanometers of membrane. So that one point is very robust. Um, you know, we do have variants within the the tomograms as well, and and so far we haven't gotten there, but that's actually as we transition, I think it's a very convenient transition to the next thing we tried to do, which was to ask sort of instead of taking all of your triangles for a given mitochondrion into one of those data points, uh, can we instead of doing this global averaging, just sort of look in your local neighborhood? Um, and to, you know, to Danielle's started hinting at hypotheses, but a big one for us is was protein content was a major driver of these thickness changes. You know, the outer mitochondrial membrane is much thinner and also conveniently known to be much lower protein density than the inner membrane. And within the inner membrane, the cristae body is like the most dense with proteins of any like any biological membrane. It's it's sort of just you know electron transport chain components with membranes stitching them together more than anything else. Um, so we we took this and uh instead of just doing global averages, we we put together an algorithm that for every triangle looks at all of the nearby triangles and it weights them such that the further away they are, the less they contribute to the average. But it averages, you know, maybe a couple hundred triangles. Um and it it produces a plot like what's on the right, where instead of having just a single color for each um uh segmentation, you can now see the variance. And we start to see actually quite strong variance where you have sort of hot spots that are thinner and hot spots that are thicker within any given uh membrane. And you can see the instead of single peaks, now in the bottom left we have a histogram that shows this sort of distribution, and you can see the outer mitochondrial membrane has its own distribution that's very clearly left-shifted from the others.
SPEAKER_02So this is interesting. So the you know, the uh do you think that explains why on your um your violin plots they have a they're they're more elongated than I would have thought. Given the vast number of uh uh points that you have, I would have thought it would have been like, you know, a very clean distribution, but it's sort of shapely, right?
SPEAKER_01It is it is sort of shapely. Um there's there's two answers to that. One is that I think that it is there's real biological variance within within our data, and we're not capturing every mitochondrion in the same state that you capture mitochondria that are undergoing different transitions that might have you know tubular or lamellar cristae, and we think that that drives some of the variants. Um the other is that you know, we're sort of on the bleeding edge of signal to noise here. Um and so you know, our pixel size is about in our raw data is about two angstroms or 0.2 nanometers. So the standard deviation here is actually about a pixel. Um so you know, these have some shape, but I would note that we're their their full spread is maybe half of a nanometer, which is still uh we think fairly precise.
SPEAKER_00Yeah, I think I think paying attention to the y-axis scale to here is important because it does like uh you know, they are we're within their variants of you know less than one nanometer.
SPEAKER_01So some combination of pushing as far as we can in the signal to noise and you know the the data itself being variable.
SPEAKER_04So Benjamin, so uh what you guys are saying is so it comes down to the quality of the tomogram, right? The cleaner it is, the more you know signal to noise, the better the segmentation, absolutely and the more precise the quantitation.
SPEAKER_00All right, great. Exactly. It's just like building you know a model into a cryo M map, right? You're more confident in the areas that you can resolve better and less confident. It's the same with this tomographic data.
SPEAKER_01Yeah. All right. Um and uh so you know, we can do these local measurements. We can also ask questions like how do these thickness measurements, now that we're doing it locally, correspond to geometric features in the data like um curvature. Uh and this one was one where you know the biophysicist in me, the the sort of person that loves these in vitro, you know, bending vesicles, you know, we know from years of those studies that the thicker a membrane gets, it gets much stiffer. And so more bent membranes should be very hard to uh to thicken. Thicker membranes should be harder to bend. Um, and instead we see sort of the opposite that if you look at your top 1%, the absolute most curved uh membranes that you see, and those correspond, of course, to crisp tips in these inner membranes. Um, we see a really statistically significant increase in thickness that that we saw uh across our IMM data in particular. Um and and this was surprising, uh, but also sort of suggested that either we might have problems with sort of data quality, as as you guys sort of hinted at, or there might be a really specific uh protein effect.
SPEAKER_00Um yeah, just to add on to that, like um, you know, this is when we started thinking, you know, we needed to ask questions are are our measurements reliable or are they being skewed or can what can we trust them? Um so to test that, we um we paired up uh with labs that had already developed approaches, um, very nice approaches to look at thickness um measurements of um of liposomes by uh cryo EM. And so uh we had this very clean benchmark of a protein-free in vitro vesicle system, and then we uh measured membrane thickness in two ways from the same sample on the same EM grid. So on the same grid, we collected untilted uh 2D projections at high dose crow EM and then tilted um like your standard tilt series collection that we're using for our lamella data, but um with this in vitro sample. And then we analyzed both with the established method on the 2D and then our thickness measurement. And um, what was reassuring to us um was that both approaches essentially gave the same answer. Um, thickness values fell within the expected sort of three to four nanometer range. And um, we're really just I think it's within 2% difference.
SPEAKER_01The mean is with the median is two percent off, yeah.
SPEAKER_00Yeah, so we felt um pretty confident after this that um that our surface-based thickness measurements in cryo T are consistent with other approaches um of these in vitro. And because these vesicles don't contain proteins, um, we could also use them as sort of a control. Um, and Ben's gonna talk about that in the next slide.
SPEAKER_01Yeah, so um these vesicles are made, or these liposomes are made so that they're supposed to be fairly consistent uh in radius, but we were fortunate to still get like a somewhat of a range. And the nice thing about a vesicle is uh, you know, the radius exactly informs the curvature. So the smaller the radius, the higher the curvature. And so we could look within these and go vesicle by vesicle and ask are the narrowest vesicles or the small, the lowest diameter vesicles considerably thicker than the highest diameter vesicles? And what we saw was that there was essentially no effect and certainly no statistically significant effect, uh, which really strongly suggested to us that what we're seeing is probably being driven by either very specialized lipids that mitochondria have like cardiolipin that we know are curvature-inducing, or by specific proteins that are curvature-inducing, such as the mycos complex or uh ATB synthase dimers.
SPEAKER_00Yeah, so both the combination of um our localized measurements that we can take, as well as sort of our confidence that our measurements are relatively accurate based on state of the art, and this idea that something about the native context of having proteins there, not just we're in vitro where you only have lipids, really started to get us thinking about let's look at um the regions, these sort of hot spots of um local changes in membrane thickness and see what's there essentially. And so what you're looking at is where we have these patches where we've identified the patches of the thinnest membranes and then the thickest membranes uh in blue. And what we found is that when we go back to the location of these patches in our tomographic data, they often correspond to regions where we can see large um, we can see signal for large macromolecular complexes, um, some of which we've felt like we could identify based on their shape, like for example, the mushroom shape of ATV synthase, um, others that maybe um resemble, for example, respirosome structures, but then some that we um maybe are not yet uh confident in identification, but have at least shown that there is, um they're associated with a patch of change in membrane thickness. And so this got us thinking, you know, maybe we can extend out um our line scan analysis where we've just been focusing on this lipid bilayer, but what if we scan this out and try to capture the signature, if you will, of um of the these macromolecules that we're seeing here. And so this is a gallery just showing all the examples of things that we're we know what they are and maybe are yet to be discovered. And what we started to see, and I hope you can appreciate that this green line that's showing here. So we sorry, we see the signal for these two peaks for the inner membrane very nicely. Um we're picking up signal there. Um, but then all of these different, what we're calling these signatures. That were captured as you extend away from the membrane itself. And so I'm going to let Ben take over and talk to us about one of these, which is the ATP synthase.
SPEAKER_01So it conveniently we have one protein that we're really good at finding lots of. And so, you know, we're finding these hot spots and we seem to be able to sort of go through them, and they seem to almost always have these proteins, but we don't know what they are a lot of the time or we can speculate. But ATP synthase we know how to work with. And so we had a data set where we could have a bunch of ATP synthase particle picks. And we could do these scans for each ATP synthase particle pick. And there's an example shown on the left here. But what we can do is we can do that over a few hundred ATP synthases because we can use more traditional template matching approaches to grab ATP synthase. And we did that, and what we see is, you know, everything's a little bit more smoothed out from this averaging. And there's a bunch of features we think that are driving that. But what we can see really clearly is that the ATP synthase patches have both an increase in density sort of right in the center of that lollipop and a slight decrease on the shoulders of it that we think correspond to that uh defocus effect in the state.
SPEAKER_02I'm a little lost. So so I get for like the these the these plots at one one line, one point, you know, you're you're you're going along, but and that works for a membrane. But but now we have a an area, a volume. And so but now but you're but it's one D.
SPEAKER_01Yeah, so that's a great question. Here's what we do we take the the position in the star file, so just the the sort of centroid position of this lollipop. We find the nearest triangle in the membrane, and so we're doing all of our scans still on the membrane. We then define a patch that's corresponding sort of to like this shadow that the the ATP synthase casts. It's about 12 nanometers out in every direction, and we define that patch. So we grab all of those triangles, and then we do it for every ATP synthase and grab all of those triangles. Um, and then we do those line scans and average that set of triangles together. So it's still looking centered on the membrane, and the the normal is defined by the membrane. We're just using the ATP synthase locations to decide which bits of membrane to focus on. But yeah, that's a that's a great question. Um and we think a lot about sort of this interplay between where proteins are and where membrane is. So it can, we you know, sometimes that shorthand can get skipped. So thank you for clarifying. Um yeah.
SPEAKER_00Yeah, so um so this was exciting to us. I think kind of hints at maybe where we can go in the future in terms of thinking of using this analysis to um to look for patches of membrane based on different features, thickness, geometrical features, or these density line scans. I think one thing um that's also very exciting to us is um using these thickness measurements to ask what is the local membrane microenvironment surrounding different macromolecules. And so this is something that we just recently um pre-printed, um, just a little taste, um, where we um actually extended this further. So, as Ben was talking about before, we um picked all of the um prohibitin complexes, which are a really fascinating um molecule inside the mitochondria membrane, inner membrane, and we identify the nearest triangle in those picks and then isolate this patch. And what we found is that um the patches associated with prohibitins consistently have thinner membrane. And whether they're inducing this membrane thinning or specifically localizing or enriched at thinner parts of the membrane is something that we can't um tease out yet. But uh I think this is just an example of maybe thinking forward of kind of the dream, I think, of a lot of people of doing this local membrane analysis, right, where you're connecting the localization of a protein um with its local environment. And so this is something we're we're very excited about.
SPEAKER_01Yeah, we had a big, big team work on this paper. So Danielle and I were co-corresponding on this, and it sort of started up just as I was leaving the lab. We started the thickness measurements actually on a different story, and then this really began with the local measurements. Um, but the this project was really led by Michaela Medina and Addy Chang, who did all of the analysis on the mammalian cells versus the yeast, respectively. And Addy also contributed all of that patch analysis that we showed at the end. Um they were supported by Hamid Romani and Dan Fuentes, who had done a lot of the uh work on the ATP synthesis that we then were able to draw in and do all of this contextual analysis. Um, and then the the remaining four people who actually added came in between preprint and final publication. Uh, Mark Frank is a postdoc in my lab that really undertook this collaborative study of the in vitro vesicles. That was something we added during peer review. Um, basically, we got introduced to Neil Watson and Fred Heberl after our preprint came out, and this became this very fortuitous collaboration. Uh, and Zidane is uh the student in Fred's lab that did all of the 2D analysis. Uh yeah. So I think that's the the full story of this paper. We didn't hit every figure. Some of them were, you know, negative results, like we didn't actually see a thickness effect with the ATP synthase, which we were expecting. Um but uh we were really sort of excited about this technique as a way to to build forward in the future. And Danielle already mentioned some of the directions that were excited to take it. Um but yeah, I think that's the that's the story of this paper. Results like we didn't actually see a thickness effect with the ATP synthase, which we were expecting. Um but uh we were really sort of excited about this technique as a way to to build forward in the future, and Danielle already mentioned some of the the directions that were excited to take it. Um but yeah, I think that's the that's the story of this paper.
SPEAKER_04That was awesome, guys. Really, it's like I was thinking about questions that you guys are answering into the next slide and stuff, and the future was very, very clear. So this unbiased mapping of local constant, that was just that changes the game, right? Uh, you know, now you can actually tell patches where things are, and then you can look at features. And if the features have some similarities, and then you can count, you can use it like a single particle measurement. I mean, like I think there's a lot of features.
SPEAKER_01That's where we're trying to go. And um, you know, uh both the the prohibiting story that Danielle hinted at at the end and some of Addy's separate work have sort of built in that direction. But we're really excited to continue to formalize this and think a lot about the math as we start getting into these sort of new new ways of measuring things.
SPEAKER_04So how does you know, just one uh one uh one uh additional question let me ask you this. So if you do all these newer denoising uh algorithms, that will drastically improve this this type of analysis.
SPEAKER_02Or will it make it worse?
SPEAKER_04Or does it make it worse? Yeah, okay, okay, okay, okay. Or it'll make it worse.
SPEAKER_01Have you guys tried that?
unknownYeah.
SPEAKER_01Um is this do you mind if I jump into it?
SPEAKER_00Go for go for it, yeah, yeah.
SPEAKER_01So this is actually one of the first things we did with uh Fred and Neil was we took their data because we knew exactly what the thickness was for their data. Um and what we found is that some denoising has essentially it helps a little bit. And even just like deconvolution or treatment to try to, you know, do like a very simple filtering seems to improve it. But when we did something like ISONet, that's like this missing wedge correction, we actually systematically reduced the thickness by 10%. And it seems to not change the variance. We're just sort of, we think we're blurring our two peaks together. And so, you know, when your bilayers blur, you sort of, it's like in NMR when you start to have mixed peaks, they kind of move together, and that's what's happening.
SPEAKER_04All right, so the raw data is the best for this. Uh these employees.
SPEAKER_01Raw data, maybe a little bit of decon, but like just make sure your data is really nice.
SPEAKER_04Okay, got it.
SPEAKER_02Okay, so um uh you've probably I'll bet the the your reviewers ask you this too. Why no Golgi?
SPEAKER_00Oh um well, I have to say, in our we're biased in the data that we collect in my lab to focus on mitochondria. And actually, you know, mitochondria and Golgi, we don't often capture them next to one another. Oh, sure. Yeah. Um, so I do have to say in that regard, um, and then I I can't remember, we did have a data set of yeast, but again, we were targeting um mitochondria. But uh there um there are data sets out there that we can apply this to, like the clamidomonas data set. Um, yeah.
SPEAKER_02So I think Well, and it's known that the the membrane gets thicker as you traverse from the cis to the transgology. And so I thought you know that would be like, ah, that's a perfect positive control, right?
SPEAKER_00Yeah.
SPEAKER_01Do you want to should we do this? Should we uh even have to publish it?
SPEAKER_02I don't want to see it. This the same kind of things I think what you will find uh in the Golgi as you are like with the IMM, because you know interesting things are happening at the tips of the Golgi and less in the middle, and um, you know, the there's different proteins decorating, you know, at each place. So I mean I would be really interested to see that.
SPEAKER_01So yeah, it's we we yeah, I mean, we hope that this is a you know, we called this a toolbox. Um, this is something we have users around the world that have been able to use it, I will say. I do my best to make sure it's a software that's really easy to use uh to varying degrees. We the the sort of the thickness measurements a little bit more complex because you have to have both the segmentation and the tomogram, and they have to be lined up perfectly. Um, but yeah, I mean, this is something we think is gonna be applicable to zillions of different questions.
SPEAKER_02And well, and that's the thing about these, you know, these lamella, these tomograms, they're just so overwhelming with signal. Like this, you know, how do you even begin to tease apart unless you're like looking for a ribosome or something like that? Right. So, so having tools, I think what it's gonna be is like lots of tools like this, where you can begin to quantify, characterize different regions of the cell. And uh, you know, you know, having a a series of these is what's gonna allow us to really begin to take these things apart and get at all the richness of data that's that's present.
SPEAKER_00Totally. And I'll just add to that um that you know, I think oftentimes the field has been sort of dominated by this structure-centric perspective, right? Like tomography was touted as, you know, it's the technique that's gonna give us a structural understanding of the cell. Uh, but the reality is, is if you go from structure first, right? You ribosomes are easy to identify, maybe filaments, membrane-associated things too, right? Um, but I think Ben and I are really thinking about flipping the script a bit and starting from the context first, um, because you can see where we're going, right? We start with geometry that can answer a lot of tons of questions that many labs care about, right? And then you can go towards, you know, bilayer biophysics, um, but then now kind of hinting towards can we find actually proteins or microenvironments? Maybe we aren't even thinking of individual proteins, we're thinking of microenvironments that have a function. And so I think that that's sort of where like Ben and I are really aligned in how can these tools in in complement with many other tools kind of move tomography away from just being beautiful, complex uh three-dimensional tomograms to actually useful um data that we can extract out meaningful biological information.
SPEAKER_04Wow.
SPEAKER_02Awesome.
SPEAKER_04That's great.
SPEAKER_02Uh so uh I think with that, let's uh move on to the um the interview part of the podcast. So it was really enjoyed the science part, but um, Ash, you want to go?
SPEAKER_04Uh yeah, so I mean I think you guys uh addressed a lot of this was just so exciting. Okay, it was about compliments, congratulations to all the authors. Yeah. Um I wanted to uh ask you so so computationally it must be very intensive, right? And uh what could improve your the ease of using such a technology, right? What what do you think would help with that?
SPEAKER_00Evo man.
SPEAKER_01Yeah, I'll I'll take this. So I will say it's computationally very intensive. When I was a postdoc, this was much more of a pain to do than it is now because segmentation has gotten much better. And so we start with segmentation. And so I spent my first like year in Danielle's lab figuring out how to segment these well. Um, and now it's like so easy to do membrane segmentation compared to what it was even like two and a half years ago. Um so the the sort of the big advances on the horizon we're thinking a lot about GPU acceleration. You know, you think about triangles, this is like what GPUs were made for. And right now we do everything in CPU world in Python because that's what I know how to do. But um I think there's a lot of opportunity speed up. Yeah, I mean, we're literally like we could write this all as shaders and be through it in like uh real time.
SPEAKER_02So um was there a a breakthrough moment for y'all? Like where you know, a lot of times, you know, you're doing a you're working on something and it's like, oh, this feels like it's going somewhere, and then it's like, ah, we're on we're on the trail now. Did you have a moment like that with you with this paper?
SPEAKER_00Yeah, I can say that the moment the times before the breakup moment, which was when Ben finally when we figured out how to make these line scan measurements, um, and then you know, we figured that out, and then we had plots of just the noisiest Christmas tree, you know, like um like uh noisiest line scans you can imagine. And I I looked at that, I thought to myself, oh no, like how are we ever going to do this? Like, how can I mean it just it reminds you that the the signal to noise in the raw tomographic data is you know, is low. Um and so yeah, I think for me, like uh, you know, Ben, I think had a few sleepless nights trying to come up with how how could we do this, but then you know, I saw the the wrapped in a bow product. Um, so I don't really know what the under the hood uh anxiety. I just knew that we had to fix that part of it.
SPEAKER_01Okay. Yeah, getting the local averaging was was, I think my last um lab meeting in the Grochon lab was showing non-working local averaging. And then it was like a few months later that like we got it all dialed in.
SPEAKER_02Um but that I'm sure that was what made the difference, right?
SPEAKER_01Like I mean, suddenly we went from sort of you know, the global figure is beautiful, but isn't like it's it tells you something, but suddenly when we start seeing these patches on the surface was when we knew we had something.
SPEAKER_04So so you guys clearly you know mentioned, you know, I can I can't see the future of this, right? This is so much you can do with this, and you know, you're going in the direction of uh, you know, this local environment and how they could look. Could you could you elaborate on, you know, you know, you know, biological perspective, right? How those things matter.
SPEAKER_00Yeah, I mean, I think part of the vision and dream is, you know, right now a lot of what we study when we're doing tomographic analysis is we have a hypothesis of some feature that might change, some membrane-membrane distance, some curvature, some protein localization, right? Um, but the dream is to look at different conditions and then let the data tell you what are the patterns that are emerging. Um, and I think especially for, you know, in this era, because we want to connect this to function, right? Eventually we want to be able to connect structural changes we observe to function. And some of those are more clear to do, like easier to do now than others, but eventually incorporating in um, for example, fluorescent readouts of of organellar function, incorporating that data in the measurements that we're looking at, and then allowing some sort of um, you know, unbiased clustering analysis to tell us what is is changing and what's functionally relevant. I think that's um another future area that I am really excited about.
SPEAKER_04That is so cool.
SPEAKER_02Okay, so now fun questions. So both of you, these will be questions for both of you. Okay, so um in your career, uh, do you have like an image or a piece of data or a plot or something that's sort of like burn into your memory, like, ah, this is some this is important. Like, so I've you know, we're really on to something. Do you have like examples of that for each of you?
SPEAKER_00Um, yeah, so this one's easy. Um, because yeah, again, this cryoet data, this low signal cryoet data has kind of plagued me throughout my PhD and now in my lab. Um, but what I worked on as a graduate student was the structure of um uh a motor protein complex called dynine dinactin on the microtubules. And um, what was frustrating about this was in the raw tomographic data, I could like very nicely, I mean, it was still you know noisy, but I could make out these complexes. But then when you throw them into Reliant or a subtomogram averaging software, it was just blobology or it was just latching onto the microtubule. And um, and so, you know, in the end, um what ended up, I I'll spare everyone the long story of this, but um I ended up staring at a lot of this data and I thought I could see something in there, but it was so noisy that I was like, no one's gonna believe me. Um, but then when I finally figured out a way to get the structure, we made this surprising discovery that everyone thought there'd be one homo dimer of dyne bound to its cofactor dinactin. And we actually saw two of them there. And this is something that I I thought I saw in the raw data, but it wasn't until we had this structure. And that, yeah, that will always go down, like for me, as like your classic Eureka moment in my career.
SPEAKER_02Ben, how about you?
SPEAKER_01Yeah, for me, I think like the number one is as a grad student, also, we we're always, I feel like, most formed during our grad school years. But like I was I was doing um this big collaboration where we were trying to do time-resolved temperature jumps in in proteins to see how get them away from equilibrium and try to observe them by wide-angle X-ray scattering. And so this is a synchrotron experiment. We'd apply for time, we'd get to go out there a few times a year. So you're sort of like doing these very intensive, like 72-hour data collections. And I remember on like our third trip, so it'd been a long time, we were finally able to get our temperature jumps working really well, and we did um some very basic uh analysis and and were able to see that we had a change in the middle of the night. And this is like, you know, it's uh it's wax data, so you we don't really immediately know what that change is, but we saw a shift and we saw that that shift was both temperature defined and time resolved. And that was like, you know, one of those moments where we knew we we like had it. Uh right, yeah.
SPEAKER_02Yes, yeah, yeah, I love that.
SPEAKER_04So, you know, yeah, you know, this is for our junior, uh junior investigated listeners, right? And we want to tell them your story and stuff. So, you know, could you both tell us uh individually, you know, how did you know you wanted to do science? Like, you know, just going back a little further, right? Uh just to you know to give some sense of uh reference point for our junior listeners.
SPEAKER_00Yeah, so I'm a 90s kid, so I was obsessed with the movie Honey, I shrunk the kids. Um that don't know that movie, just like wacky dad who is kind of like a scientist and then develops this machine and accidentally shrinks his kids. And it was such a cool movie because you got to see this totally new perspective on the world, right? Like the kids are as tiny as like a pollen grain, and there's this scene where the sun is like planted in a flower and then gets you know scooped up by a bee and is flying around. And so that to me was kind of like my first sort of realization that there's a hidden world beyond what we can just immediately see. And then, you know, as I got my first microscope kit as a nerdy kid that was also opening up to this hidden world. And then I've just sort of followed this passion to see things um and to come up with new ways to see things and new ways to, yeah, quantify or or analyze um images, I think has been a theme for me that um I don't plan on giving up. Anytime soon.
SPEAKER_01How about you, Ben? Yeah, I didn't want to be a scientist for a long time. I thought I would, I was, I wanted to do math, which is like still sort of, you know, it's still basically being a scientist. But I um like I started as a math major in college. It was like I was sure it was what I wanted to do. Um, and I ended up just loving my science classes. And then I worked in a lab uh looking at sort of how proteins break down this sugar lignin. And I did it originally just as a as a not a sugar, poly uh polysacchar polymer, but biopolymer. Um and I did it as a summer project, but then like really became obsessed with sort of the complexity of how biology is able to make and break down these like pretty um sort of magical materials. Uh and that was really the beginning of the end for me of thinking about how proteins work in these big complex environments.
SPEAKER_02Yeah, you were doomed.
SPEAKER_04Got the hoax in.
SPEAKER_00We can't run away from math though. Math is, yeah, you know.
SPEAKER_01That's like that's that's what I that's what I work with now. That's what gets pays the bills.
SPEAKER_02Okay, last question. So um, who are your scientific heroes and why? Danielle, you want to go first?
SPEAKER_00Oh gosh. Sure. I mean, this question is I I feel is daunting because I feel like I've been inspired by so many who have come before me. Um so one example is Bridget Kerriger was um at Scripps when I first started. And actually, one of the reasons why I came to Scripps because of the amazing cryoum infrastructure that she and Clint Potter and the group had set up. Um, so I, you know, I think she's definitely up there. Someone who I haven't um like had a lot of direct mentorship with, but has had a lasting impact on sort of how I think about setting my up my lab is Sandy Schmidt. Um the two pieces of like sort of advice that um she presented to our junior faculty group was um number one, your lab doesn't exist until its first paper. And so for me, that was very much like, okay, gotta, you know, focus on that. And then this concept of like, what is your minimal publishing unit? Um, and so I think that that's something that I think about a lot. Um, and I I kind of try to train my um group members to think about. And so um, yeah, I could go on and on, but I think those two were kind of really, you know, um transformative, and and especially how I'm thinking about managing my lab and um sort of establishing myself as an independent researcher.
SPEAKER_02I quote Sandy all the time. She uh there was a uh there was an ASCB that she that I went to and she was giving some talk about, I don't know, career or whatever. Yep and it was like the exact same things you just said. And the quote was like base hits, not home runs. I don't know if she is the exact same with you, but I was like, Yep, it was like boom, oh yes, of course. That's what you know, base hits, not home runs. Because the home runs will come, but you gotta have those base hits.
SPEAKER_00Absolutely, yeah.
SPEAKER_02All right, Ben, your turn.
SPEAKER_01Uh my answer is a little closer to home, so to speak. But I my both my my mother and my grandfather uh were career scientists. Um, and so that probably explains why I didn't want to be a scientist in some ways and somehow ended up getting dragged back in. Um, but you know, I grew up around people that were always asking why things are the way they are, um, and and being sort of challenged to do that in my life, sort of in all in all ways. Uh and I think I've really, you know, my research doesn't look anything like what theirs, what theirs did or does, but um, you know, I think sort of approach and philosophy, I really feel inspired by them every day. Nice.
SPEAKER_03Awesome.
SPEAKER_02All right. Well, uh, Daniel and Benjamin, uh, thank you so much for taking the time and sharing your science with us today. Uh, we I really enjoyed hearing about it all and chatting with you. Absolutely. And so that does it for this episode. We hope you all join us again for the next episode.
SPEAKER_01Did I hear something? Sorry, I was just saying thank you for having us.
SPEAKER_02Of course, yeah.
SPEAKER_01All right. Well, so we managed to not have over talk till the end.
SPEAKER_02That's pretty good. Okay, so anyway, that does it for this episode of uh structure club, and we hope you all will all join us for the next episode.