Scientist Stories: Ron Germain, Imaging the Immune System

Are we good? Yeah. So there's John and Jennifer and Japan, and I'm Germain, but it's a G. I really want to thank the organizers for this opportunity. I've been to, I think, all of the meetings, and I always leave time in my schedule for this because I think it's just absolutely terrific. But there's something called IAS, Immunology Avoidance Syndrome, and it's probably characteristic of many of the people in this room. So I'm going to take a few slides to give you a little introduction to immunology just in case you don't think about the immune system very much, and that hopefully will explain why we do some of the work that we do. So we live in a dangerous world. We have bacteria or viruses or fungi or parasites. And there are an enormous diversity of these organisms that the body needs to respond to and protect us against. But not only is this diversity a challenge, but we have the challenge that the invasion and the desired location for infection can be anywhere in the body. And so you have to think about what kind of evolutionary strategy would emerge to deal with all of these essentially unknown potential problems in all these various places. And so the way the immune system has solved this is to combine a series of mobile elements with certain fixed tissues and therefore enable scanning and concentration of those elements and material from these organisms in a way that's highly efficient. And so we have primary sites of cell generation in the bone marrow and in what's called the thymus. We have the characteristic lymph nodes that are in the peripheral tissues and then specialized mucosal tissues that also contribute to host defense. And getting down to a more granular level, we've got cells that come out of the bone marrow. They may be myeloid or lymphoid. The myeloid cells can find inflamed endothelium, adhere, roll, transmigrate, and go into the tissue to mediate effector function. Lymphocytes circulate around and look for these secondary lymphoid tissues such as a lymph node illustrated here. Lymphocytes enter through specialized vessels, high endothelial venules. They segregate into subregions for reasons that immunologists still don't fully understand. But clearly there is a rationale evolutionarily to do this type of microanatomical segregation. And then these cells have this peculiar feature of having somatically derived by recombination, clonotypic-specific receptors. So each cell has a different specificity. For T lymphocytes, they're looking for signatures of infection based on small peptides bound to cell surface molecules called histocompatibility molecules. If there is a match here, these cells will undergo differentiation, expansion, and again, like the myeloid cells, look for inflamed endothelium to go out into the tissue. And so if you think about this in a larger sense, there are some key questions that we can address, which are what are the real cellular dynamics at all different levels? The rates of circulation and movement through these different tissues, but within the tissues, How fast do the cells move? How do they maintain adhesion long enough to exchange information while constantly moving and searching for other elements? We need to understand why there is this particular organization to lymphoid tissue because although all these cells are turning over rapidly within any given tissue as you scan through your repertoire, there is a certain structural organization, and I'll come to that during the talk. We obviously want to know molecular signals that control many of these events. And then what happens when the immune cells are out in parenchymal tissues? What do they actually do and how do they interact? And so about 15 years ago, a number of groups took a page out of the neurobiological world that had been using two-photon imaging for a while, mostly in brain slices, and started trying to apply this to the immune system with the notion that we could actually see how it behaved in vivo. Since it's an imaging audience, I'm not going to walk through this slide in great detail. It's a standard way of providing 4D datasets. And these are played back in movies that look like this, about 200, 300 times real speed. But although there's a little bit of noise, you see there's this black background. And one of the things I want to stress is we only see what we make fluorescent plus the free channel or two for second or third harmonics, as Peter Friedl may talk about. And yet they're in this environment. And so one of the things to keep in mind, you know, seeing is believing, but not seeing is not something that you should ignore. If you— if something is in the black, you should be careful about how you interpret your data, and that's not always true, unfortunately, for people doing these types of imaging studies. And so the immune system has 3 major components. It has barrier function, the skin and mucous membranes. If you breach those barriers, you have innate responses that occur in minutes to hours and may extend over days unless you get into certain chronic diseases. And then gradually you develop what we're more familiar with as part of the immune system, the adaptive responses. These are the ones that are entrained when you get vaccination or you're protected after an initial infectious bout. And these emerge over days to weeks and can last for years. And so let's look at a barrier breach, a sand fly biting the skin as a successful blood meal. It means it's gotten down to the dermal capillaries. And if we do this in the ear of a mouse, I'm going to orient you. There's a blood vessel here, although it doesn't fill with tracer in this particular movie. This is the site of the bite and the damage, these green objects. Are genetically GFP-labeled neutrophils, a type of white blood cell. These red objects are Leishmania parasites. And when we look at this, there's a lot of biology in one image here. So if you're interested in directional control of cells in vivo, no neutrophil comes out of the wrong side of the blood vessel, which means over one cell diameter, directionality is absolute. Those of you interested in integrins, these long tails are the remnants of cells sticking to the endothelial cells that have to dehisce in order to move. We get very directed motion to the center here, but this underestimates the complexity of how the system behaves. If we turn this all on the side and now we're here in the dermis with our free channel for the collagen, you see these cells moving up into apparently empty space, but that's the epidermis. We just don't have any signal there. And the question is, what are these neutrophils doing in these particular regions? And the answer is they're filling in the gaps where the sandfly bit.. And this now reestablishes barrier function, prevents pathogens from gaining access down into the disturbed dermal capillary bed and then going systemic. So clearly there's very complicated behavior of these cells in 4 dimensions, time and also in space, and yet our paradigm in the field is this very simple carbohydrate-based rolling and then chemokine-driven adhesion through integrins and transmigration. The question that we asked a few years ago is what happens in this very complex extracellular compartment with non-cellular matrix components, many different cell types, what are called damage-associated molecular patterns from dead or dying cells, and of course material from the pathogens. This is work that we published, and I'm only going to highlight some key features. Tim Lammmermann did the studies I'm talking about. He's now a group leader at the Max Planck in Freiburg here. He developed the sterile injury model. We used the pulsed two-photon laser to create a scaled size of damage. I'll come back to the size a little later. Directly in the dermal region. This way we don't put a needle through the skin and transmit what are called commensal PAMPs, things that are from the bacteria that live on the skin. So we know it's a sterile lesion. And then we can image. Here we've inverted the colors. Neutrophils are red and these green objects are what are called monocytes. We get this very characteristic swarming behavior around the damage., and then gradually these monocytes form a shell around the neutrophils. If we deplete the neutrophils, the monocytes don't show up, so we know that's a secondary reaction. And what Tim did through a series of knockout studies was to show that— and I'll give you some illustrations of this in a moment— that the first few scouting neutrophils touch the damaged area and some of them die. That initiates a cascade that leads to the production of a lipid mediator called leukotriene B4. And then there's a peculiar property here, that leukotriene induces other neutrophils to make LTB4 themselves. Instead of a diffusive gradient for guiding the cells, we get signal relay over space. And so from hundreds of microns away, the cells turn, very rapidly move to the damaged area. Interestingly, for those of you who think about integrins in the context of mesenchymal cells with focal adhesions, these cells migrate in dense tissues in an integrin-independent manner. But forming the swarm is dependent on MAC-1 and LFA-1 for immigrants. And what this is doing is essentially segregating the damage or the infected material away from the healthy tissue on the outside. And then gradually, of course, you have to stop this inflammatory reaction. And at least temporally, the timing of the accumulation of those monocytes matches when this stops. And Tim is working in his lab to see if these make a series of mediators that are known to interfere with the leukotriene that drives this whole response. So let's talk about this shutdown because from a biomedical standpoint, this is important. If you do have to have active regulation to stop the inflammation and you imbalance it, then you're going to get a lot of neutrophilic tissue damage. The problem, as many of you know, is you have an animal that's anesthetized for several hours, can get cold, hypoxic, hypoalimented, and we put a lot of photon energy into this imaging field. And so not moving does not mean anything other than maybe you have a bad prep. So how can we evaluate whether there's actually, you know, physiological control at this level? And so what we've done is to take a page out of some of the things that Jennifer did and use photoactivatable GFP to mark the cells that are already in a swarm. And then we've created new damage. If it's an effete preparation, then nothing is going to happen. Basically, the cells are frozen because the the system is essentially dead. But if this is a dynamic balance between pro- and anti-inflammatory signals, everything is going to reactivate based on the damage now changing the balance toward inflammation. And this is the one slide that truly is seeing is believing. You do not need to do quantification because as soon as we do that other damage, all the cells go to where the new damage is, including the peripheral ones that are already in this first swarm. So we know this is a dynamically balancing system, and we need to work out now what are those actual mechanisms that control that, because you can easily imagine where that imbalance— if it's too rapid, you don't get enough of the neutrophil effect, and if it's too slow, they can damage the tissues. I'll show you in the next few slides. So that's an older model, and more recently Stefan Uterhart has developed a new model. And the reason we want to look at this is that in the last few years in immunology, there's been a lot of attention paid to the fact that macrophages, a very classical innate immune cell, do not all derive from circulating monocytes as was previously believed, but many come from embryological sources like the yolk sac or the fetal liver. And they form what are called fixed tissue macrophages. And the perineal cavity has a beautiful array of these. I'll show you in a moment. You can get classical macrophages from monocytes here. You can also bring in neutrophils. So we can look at all these myeloid cells in one time. In this compartment. And so it's a beautifully organized region. You've got the muscle wall from the perineal abdominal wall. You've got serosal cells on the surface and a beautiful array of both collagen and elastin, as well as these very organized vessels that run parallel to the muscle fibers that you'll see as this scrolls back. And so in that environment, if we do an en face imaging, we also can see that these red objects, which are the macrophages, are embedded in this elastin net. And it raises interesting questions as to whether they have mechanosensors so that disruption in this elastin bed may signal those cells that there's something adverse going on in the tissue. If we look at this now back in the 2P mode, when I say fixed in the tissue, they do move processes around, but you see they don't migrate as I was showing you for those lymphocytes where there's neutrophils. Earlier on. Even if we add bacteria here in red, we get a massive neutrophil infiltration, but until this field fills up, you can see the underlying macrophages do not migrate. The neutrophils do what they should do. They come along and vacuum up the bacteria. But you also see that there's not all the swarming that I showed you previously. If we look carefully, though, when this neutrophil takes up this bacterium, it dies, and as soon as it dies, you get a swarm. And we can look at that in more depth, so in higher resolution. This is the damaged area. This neutrophil will die right now, pops. And now all of these neutrophils rapidly swarm. So we know that that death is the initiation of the swarming behavior and involves contact with the damaged region. And that's critical for the next few— slides I'll go through. So just to reiterate, you can get some swarms from spontaneous death. We make a little bit of laser damage here. Everybody goes to the damage. So while chemotractants bring neutrophils into the tissues, it's the damage in the tissues that directs their movement and behavior at that point. It's always the damage that dominates their migratory behavior. If you infect the lymph node, you get this very dynamic clustering here. As you get little foci of growth and then control, and then another focus explodes somewhere else. The problem with neutrophils is they're very useful. You don't want to be without your neutrophils, but this is the collagen in the dermis, and you see that where you get the swarm, you get a hole. So you're getting tissue damage as a consequence of having this inflammatory response. We put this into the perineal model. Here's the damaged area. Here come the neutrophils. You can see they digest the damaged material over time, but if you look down here, you see they also make holes in the collagen matrix. So we asked the question, do you always have damage associated with cell death because cell death causes the neutrophils to come in? And everything we had done to this point was damage at the level of 3, 4, 5 cells at a time. But other studies we had done suggested that mechanical forces could cause individual cells to die. And in serosal surfaces, pleural surfaces where there's friction and pressure within organs, you might imagine this would happen episodically on an ongoing basis. And if every time that happened, you got this kind of neutrophilic response that made a hole in the tissue, Over time, that would be extremely bad. So we figured there must be another mechanism for dealing with that. And so here is a lesion now the size of an individual cell. And we immediately within a few seconds to minutes get this sensing response from the nearby macrophages. And over a few minutes, they send out processes without moving their cell bodies that basically cloak or encase the damaged area. So what happens when that process takes place? So remember, when a cell dies, that's going to attract other neutrophils. Here we brought some neutrophils in first with a chemoattractant and then we made the damage. So there hasn't been a chance to get that cloaking response when the neutrophils come into the system. That one touches, dies, and And now you get this very directional migration and the beginning of a swarm. But in this case, we've turned the tables around. We've made the lesion and now this neutrophil has shown up a few minutes later. And under these conditions, none of them managed to touch the damage. And there's no swarming. And there's no tissue damage. And the reason for that is that those macrophages have completely cloaked the damaged region. When they do that, they very rapidly degrade it. So if we zoom in here, you can see the large phagolysosomes that are being formed and they are clearing this out very quickly without an inflammatory response. And so we now have a different view of inflammation and resolution, which is the classical model. There are really two types of responses to tissue damage. If it's small enough, episodic of the type I talked about, you get cloaking that prevents the neutrophilic damage and you get rapid healing of that very, very small lesion with no scarring. If you're in this other context, now you get this neutrophilic response and they're not shown here. This now attracts those monocytes. They go through a phase first of inflammation and then gradually they shut this down, but you wound up without having to remodel the entire tissue. So it's really important to understand the scaling of how this works. And although I've only shown you the peritoneal cavity, those fixed tissue macrophages exist in all the serosal tissues, the interfascial region and muscles everywhere in the body. And so all these places that have frictional components have these cells living there. And we imagine that this process of cloaking and prevention of inflammation is generalized and critical. And we're now looking to see whether if we impede this by knocking out the relevant signaling molecules, we get the development of chronic tissue damage, and then to look at the molecular processes to then try to link it to GWAS studies and other studies that might suggest this has a role in chronic human disease. So let's move over to adaptive responses, the kind that people are a little bit more familiar with. This is a classical picture in immunology. You get the cells that take up the antigen or the pathogen and present it to the lymphocytes in the system. And those lymphocytes can be of the CD4 flavor or the CD8 flavor. And they can work with B cells to make antibodies that are what protects you after most vaccinations. And the problem with this cartoon is numerically it makes no sense. Because early in an infection, there may be 1 in 100 to 10,000 of the of presenting cells in the lymph node that actually have the relevant antigen on them. And there's 1 in a million of these cells that have the right receptor, 1 in a million of those cells, and 1 in a million of those cells. And so the challenge for the immune system is to get all of these cells together in time and space. And so the way that's been done evolutionarily is to do what we do in the lab when we increase the concentration of reactants to drive things in the direction we want them to go. As I told you at the very beginning, you don't put lymphocytes en masse in all your tissues. You have them circulating around and entering into these lymph nodes, and you concentrate that along with antigenic material or pathogen material that's drained through the lymph from the tissue into those same sites. So we first put two reactants together in these small organs. But then within those organs, there are subregions, for example, what we call the T cell zone. This has its own stromal structure, and we've shown— and I'll come back to this later— that the red objects, the lymphocytes, move on that stromal network as they search for their cognate ligand. And the interesting thing is that the cells that are presenting that ligand to them, the dendritic cells, live on that same fiber network. So we've now reduced the volume another level and increased the concentration to make these reactions more efficient. And then there's a bias of how the cells move based on chemokines once any cell does find its appropriate target. So we have this combination of physical but also chemical cues that take a very inherently inefficient system, you know, 1 in a million cells that has to find 1 in a million other cells in a platform of 1 in 10,000 other cells, and actually makes it work. And it's not a pinball machine the way the early 2P movies were interpreted of just sort of wandering around. And getting things to go. And so this 2P imaging has been really useful. We and others have looked in lymph nodes as I've shown you. We've watched these cells move on this fiber network. We've been able to look at the development of humoral immune responses in what are called germinal centers, look at cells sampling bacteria from the commensals in the gut, look at things like the kidney, look at myeloid cells forming granulomas in response to mycobacteria, and look at the way lymphocytes that are also in those same granulomas in the liver, with some difficulty imaged in the lung. I've shown you these innate responses, but we've also looked at adaptive responses in the ear, skin, and we can look at— in the bone marrow. So it's a very, very useful method, but there's a problem. In the mouse, if this is the lymph node we look at, we only image this area when we do 2P live imaging because the speed of the cells requires us iterating the volume collection every 30 seconds to a minute, and we can't do enough Z-steps to go much deeper. It's not an optical limitation. And so not only are there things in the black that you don't make fluorescent, but all of this is not in your imaging field, and that's a problem. And we said we needed to look at that by doing sections. The problem is histochemistry was limited at the time we started doing this to 2 to 3 colors. But those of you who know anything about immunology or flow cytometry know that There are 300 or 400 what are called CD antigens that are defined now, and you can look at any given subset using 3 to 5 different markers. And so typically in an immunology lab, people are using 9 to 15 different antibodies to do an experiment on flow cytometry, and we're doing 2 or 3 colors in histochemistry, and there's no way to match up what you see in the tissue with what people see directly. And so several years ago, Mike Gerner in the laboratory, who's now at University of Washington, developed what we call histocytometry. We stain with up to 14 different antibodies in different colors at one time, tile image that area, deconvolve to get a bit better Z resolution, compensate to deal with channel spillover. You can do this in spectral imaging mode. It just makes bigger files and takes longer. And then using software tools, we can instantiate individual cells based on surface membrane staining, but then we can collect all the fluorescence information from those cells. Well, cells with fluorescence information are the datasets you use in flow cytometry, so we now can take this and put the data directly into flow cytometry software tools and gate as if we were doing flow cytometry, but all the objects obviously have associated XYZ location parameters, and so we can take anything that we gate and quantify this way and localize it in the original tissue section. And so we've done this, and I'll show you some other examples with humans in a moment. I'm not going to show you these data, but we've also shown that anti-phosphoprotein and anti-cytokine antibodies work as well. And so you can phenotype the cells, look at their signaling state with the anti-phospho antibodies, look at their function based on their cytokine production, and then localize them in the tissue all at one time and multiplex parameters. So for those who are at all familiar with flow cytometry, I'm going to make it simple. This is staining on one axis with an antibody that looks at B lymphocytes, and this is staining on the other axis with an antibody that looks at T cells. We can see there's a population here, and we can gate that and call that blue. There's a population here, we can gate that and call that red, and just have the software print all of the events in these two gates back in those colors, and what we get is a printout that looks just like a primary lymph node with peripheral B cells and central T cells, although none of that information was obviously put into the software package. And so we've done that and identified all of the different dendritic cells that exist in a mouse draining lymph node, and what's interesting is, in contrast to the notion that they were sort of randomly dispersed around, they all have very precise regions where they tend to live, and that now imposes additional constraints in our thinking about how the immune system works in that lymph node context because there's going to be differential access to antigen. If it's coming in through a lymphatic and percolating around on the outside, as I'll show you in a moment, these dendritic cells in the middle are going to get much less access to that antigen. If these are specialized for presenting antigen in a particular way, then they're going to talk to a subset of T cells differently than the ones that are in the middle, and we need to start taking account of that in the immune system. And so some of them, dendritic cells that we characterized in that study, actually are embedded in what's called the floor of the lymphatic sinus. So this is the capsule of the lymph node. Lymph comes in and it percolates in this region here. And these cells live both where they have access to the lymph flow here, but also where they're touching into the T cell zone here. If we go back to the 2P, we can see that live. These are particles coming from a vaccine draining into the lymph node in the lymph. This is the capsule. And here are those dendritic cells with processes that can directly capture that material. So we go back now to thinking about what I told you here of how everybody does their job and have to find each other, but also now this finer-grained structural organization. We have to think about how the cells are moving around and how long it takes for them to find their relevant targets. And so I want to back up a little bit and talk just a little bit more about how cells migrate around in the lymph node. There's this fibroblastic reticular network here that I've shown you previously. If we perfuse fix the system, Static imaging is slow. We get cells that obviously are not little round balls like you think of as lymphocytes, but these are migratory cells with uropods, and they're beautifully positioned along those fibroblastic reticular cell fibers. But this doesn't tell us they're really migrating along them. They could just be bumping into them as they're trying to move through space within that region of the lymph node. So to get at that, We've created an animal that is pan-green and then transferred in bone marrow that's in a different color, so we can now look at the stromal elements in green and the lymphocytes in another color. And when we do that, what we see is, as I told you, these cells are moving along the fibers. Now that appears to be a random walk, and I told you you can bias it with chemokines, but the question is, do they walk on all of that network that extends throughout the T cell zone or do different subsets of lymphocytes take up particular positions that begin to match the dendritic cells that I told you are specialized for talking to them and alerting them to infection? And so this is work that Antonio has done in the lab. Very quickly, to reprise what I said, these red cells are the more peripheral dendritic cells and they're specialized for presenting antigen to CD4 T cells. These green cells in here are the ones that are more specialized for talking to CD8 T cells. And that's just reprised over here and shown here as a density plot because if we now look at CD4 and CD8 T cells, even though for decades they were imagined to be randomly intermixed in the T-zone of a lymph node, we get a matching profile of concentration. The CD4 cells are more peripheral and the CD8 cells are more central. So they're now in a position to match their respective dendritic cells. We understand— that's just quantification of it. We understand what positions them. There's an oxysterol receptor called EBI2. If we eliminate it, this biased distribution with respect to the middle of the lymph node disappears. It slows down the rate of activation because now the cells have to migrate over longer distances to find the relevant presenting cells. And that small difference in activation almost completely eliminates the ability to protect animals against malaria by a vaccine that elicits CD8 T cell responses because they need the cooperation of these CD4 T cells for that protection. So this is something that can be very strongly evolutionarily selected. And again, it argues that we have to keep looking at deeper and deeper layers. Just as you look at finer grain and go to super resolution, as we'll hear in a moment, to look at things within cells, what we're seeing here at a more macro scale is that what was described grossly by histology many years ago has many layers, and each of them contributes to making, again, this inherently inefficient process much more efficient. So this just summarizes what I've said, and I don't want to go through all the text. We applied this to human tissues, and Andrea has led in doing this. So here we're looking in a human lymph node. These are what are called germinal centers. They're the site of somatic hypermutation to form high-affinity antibodies that you need for protection. And we can get beautiful staining of the CD4 cells that help drive that B-cell response of nuclear factors like transcription factor BCL-6. And we can use that now to answer questions that are hard to address in human biology. Typically, a lot of immunology in humans is done with peripheral blood cells in a plastic dish. But that doesn't really tell you a whole lot about how they're actually working in vivo. By applying this more multiplex and quantifiable approach, we can begin to ask questions that at least give us clues to how cells may act.. So this is a very complicated diagram, but the only thing you need to keep in mind is that there are these green cells that are the go cells. These are the ones that help make an antibody response. But in the last few years, a red or stop equivalent type of T cell has been described, the TFR. And the way most people think this works, although the diagram is sort of agnostic to this, is it works in what I just showed you pictures of, the germinal center, because that's really the response shut down if you just do an in vivo experiment. But are the cells actually in the germinal center? And so here we're doing a multiplex staining. We're only showing you a few colors, and I'm even going to take some of them away. And so here's the germinal center. The red spots are those stop cells. And what you can see is, except for right here on the rim, there are extremely few of these cells in the germinal center. They're really around the edge. We can do this in higher resolution, zoom in, look at it carefully, see cells with both surface staining and nuclear staining so we know that we're really looking at the right cells. And do that in different compartments, quantify it by the kind of histocytometry I've told you about. And it's very clear here that there are virtually no significant number of these regulatory cells in the germinal center. So the model that existed in the literature based on teasing up lymph nodes in the mouse, even teasing up lymph nodes from humans or doing flow cytometry, loses the positional effect that tells you where these cells are operating, which is really on the border outside the germinal center. And that has real implications for mechanistic understanding. So I want to end to get to technology here. And the problem is I showed you the little box previously in this image for the 2P. But if we do a section like I just told you about, we're undersampling 3D space tremendously. And so many people in this room have helped pioneer methods for clearing. This is an older method with cytokeratin staining. And we can get these really beautiful images in 3D, for example, of lung here where we go from alveolar epithelial cells and large conducting airways over half of a mouse lung. The problem in our hands was that whenever we used any of the published methods, we either had to give up looking at fluorescent proteins if we tried to stain, or we could look at fluorescent proteins and not stain, and most of the staining needed to be amplified with secondary antibodies, so we couldn't do high multiplex, which as I told you is critical for our work. And so Michael got together with Weiji, and we wanted to develop a clearing approach that would do all of these different things at one time. And the answer is we have such a method. It's a single-step reagent, very cheap. It gives very good optical clarity. This is actually a non-decalcified bone. We can actually image that directly into the marrow. So now we can do this with direct fluorescentated or conjugated antibodies in multiplex. This is 5. We can do it in 8 or 10. So this is a millimeter-sized object. Image with no sectioning at all. These larger fat objects are the arborized high endothelial venules that the lymphocytes enter through. The smaller ones were the capillaries. You saw the peripheral B-cell follicles. But of course, we have high-resolution data here that we can now turn into virtual sections in any dimension we want. So we can, if we go to the right number of parameters, essentially phenotype and understand all the cells in all their organization entirely in this millimeter-sized object now. We have algorithms for segmentation of these densely packed cells, and I want to stress that we're looking at cells that are only 5 microns. So when you do a light sheet and you want to look at a whole brain and you're using a 0.3 NA lens, your resolution is 5 microns. You can't segment 5 micron cells doing that. So this is a whole different scale of technology here, and we can take this and apply apply it throughout the volume, so we can basically walk through and see— port everything in each of the sections over to our flow system, quantify it, and then play it back. And I've taken the slide out, but we now have VR-like mechanisms for showing you where all those cells lie in 3 dimensions if you want to go in and look at them. It's not a lymphoid-specific method, so here, for those of you who like the brain, this is a confetti mouse.— and this is— looks like a section. It's not a section. I'll show you that in a moment. But we can even see these very small processes here. It's a very high resolution, but this is a 1 to 2 millimeter volume that's being imaged under these conditions. And this is fluorescent protein, but I'm not, again, going to take the time to show you the image. We can go and stain on top on top of this and look at astrocytes or microglial cells or anything else we'd like to. So it's a combination method. Here is this method applied for 3D in the small intestine. And then if this plays— we're going back to the lung because I wanted to show you something here. So in the mouse, you don't get quite the same arboration and grape-like clusters for alveoli that you get in the human, so we're looking at all the conductive airways here to the alveoli. Now we're putting in a GFP population, but this is all at high single-cell resolution throughout the volume, even though here we're looking over centimeters in XY and multiple millimeters in Z. And again, we've now worked out ways of running this at 12 colors at a time. And so with some of the new objectives that have come onto the market, This is the same type of image. This image here and— I'm sorry, this image here and this one here are 3 millimeters apart. The working distance is 6, so in principle we can do 5 or 6 millimeters in all of these different colors. We're still limited in any one cycle to 8 to 12 colors. That's often not enough, so it turns out this method is compatible with stripping and restaining, so this is This is the original, that strip, and this is just autofluorescence. We can do this image. I'm not going to show you the strip because it's just black. And we can restain with the same antibody. This shows we're not just bleaching and leaving the antibody in place. We can restain with subpixel segment registration. And so if we take 10 at a time— this only is 6 at a time— and keep 1 or 2 for fiducials, in 2 or 3 cycles we're at what are called cytofluorescence levels, 30 or 40 parameters for looking at things. Now, it turns out other labs have been working on this high multiplex, and Garen Nolan's lab in particular. He's developed a method that goes back to the older MELC method but doesn't use bleaching or chemicals. It uses iterative nucleic acid-coated staining to do this type of iterative imaging. So you do 2 or 3 at a time, you image, and then you remove the fluorochromes that are associated with small nucleic acid probes, and you can iterate this. And so here I think there were 40 colors done. It's a little blurry in the slide he gave me. I apologize for that, but they can type all these cells, and they've worked out methods now that run in the cloud very efficiently to automatically define neighborhoods of cells that are touching each other under these conditions and show where they live.. And why is that interesting? Because that chemistry is compatible with our clearing method. And so we now are able to do this iterative staining in the clear tissue. So this is only about 150 microns, but it's using their chemistry to do this. And it's easier to see what we're really getting in this next image. So I'm just going to move along here. So here, this is 630 microns. Looks pretty blurry because it's all maximum projection, but if you go in and see, we've got fine individual, you know, 5-micron cell resolution throughout this volume using this approach. And the reason this is especially valuable is it's compatible with another nucleic acid method, which is RNA-FISH. And so we can then both do do FISH and stain. And this is an example here using the CODEX reagents. And so the green cells are T cells, and there are some of them that are in this so-called B cell zone here. And so you need to just remember that kind of distribution because the probe is specific for a subset of these CD8 green T cells. And you see the probe is really in that area and only in these few cells that were invading. So it's really quite specific. Specific, and we're working on knockout controls to validate this. And so the point is that if we combine the CODEX iterative staining with iterative FISH and the C3D clarification methods, we'll be able to do highly multiplexed 40-parameter or so 3D datasets in complex tissue that builds off of the new data that's being developed out of single-cell RNA-seq and other methods. And we think this is going to be extremely valuable at a tissue level for really understanding how things are organized, what's expressed at both the protein and at the RNA levels, because they're not always equivalent, and give us some good information. And so I want to end by saying that I've shown you only this kind of imaging part, but as we heard earlier, you know, imaging is a substrate for collecting data that needs to be interpreted. We've developed software tools in the lab, particularly Martin Meyer, a company called Cellushyme that enables the creation of highly detailed models that actually will do simulations of dynamic spatially resolved processes within cells. So it's not a stirred cell model. It actually looks at where things are. If you have a schmooing yeast cell, you see things in the schmoo and not in other places. And it does it as the cell potentially is changing its shape in essentially real time. And so we can begin to create these kinds of synthetic cells using the software tool. We built real biochemistry. These are not just cartoons. And these are essentially the swarming neutrophils. And those of you right in the front may see some lavender here. That's where those cells are interacting via integrins and signaling via those molecules built into the software tool. So we can take the type of imaging data I've shown you and begin to create quantitative models of of how things are behaving and ask questions about whether our models of how cells are moving and finding each other matches the kind of data I showed you from Antonio where there's a 3-hour delay in cells finding their partner cells, for example, and understand that better. And so putting that all together, we're really trying to get a multiscale model of how the immune system works. But any of the technologies I talked about are not immune-specific. Gary and I are trying to work with the Human Cell Atlas group to try to complement all the single-cell sequencing they do by providing this method I talked about at the end as a strategy for putting that into a tissue context, you know, in any particular tissue that you care to look at.

Thanks very much. John, for a really amazing talk.

We have time for questions.

IAS, Immunology Avoidance Syndrome. Not at all. So I'll start with the cloak of invisibility, which just blows my mind seeing those neutrophils do that. Do you think— you showed that that's every tissue that you've looked at has this type of cell that's capable of moving in. It's sort of embedded within the matrix and then just extends out to shut down any response to cells that are dying. I'm curious as to how you think that might be related to what's happening in the brain under brain inflammation, et cetera.

Are there cells like that in the brain? Yeah, so the very first demonstration of this process was actually in the brain. So laser damage causes microglial cells to do that same extension process to the site of damage, but there was no follow-up of what that actually did. There was a response, but it wasn't looked at beyond, oh, there's a response. Typically you don't get neutrophilic infiltration, so whether it's playing the same role of preventing neutrophilic damage Or whether those cells are playing a different role in protecting neurons in that region through another mechanism, I don't know at this point. Peter. So you've shown a couple of examples where immune cells go persistently either along mechanical cues or probably chemical cues. So my question is, can you envision how to visualize active chemotactic chemotactic response by imaging approach rather than inferring it indirectly? Well, we've done a couple of things to look at that. So in one particular case, we measured what the T cells and dendritic cells were making together and then blocked those chemokines and showed that that blinded the system to recruitment of cells in vivo. So we weren't watching the signaling, but we were manipulating that component and showing that it affected the behavior. And we also did that by removing the chemokine receptor from the responding cells. So whether we block chemokines or remove the chemokine receptor, then the system becomes blind. So that can be done with those types of controls. For actually imaging it, even though you're supposed to get calcium from chemokine signaling, we haven't, with any of the calcium sensors we've added, seen that compared to level we get from a sarokinase-linked receptor like a TCR or a BCR. So that isn't the obvious method for doing it. One could think about whether recruitment of beta-arrestin or something in a feedback mode would give you an implication of that type of signaling. There are a few chemokine receptors that have been built with fluorescent tails where you can look at their concentration, but now you're looking at both the accumulation and the downregulation, so whether that would be visible with enough intensity, I really don't know. So mostly what we've done is not look at the signaling per se, but look at regulation of those processes by manipulating the molecules involved.

Gadams? Fantastic talk, thank you. So you mentioned the modeling of the single-cell interactions you have in which you want to map back now into the tissue organization. First question is, what are the length scales of interactions that happen? And second, how do you mathematically actually now tackle that problem that you—

it can get very quickly very chaotic. So there are a couple of things. Let's go back to Peter's question. In work that I didn't talk about at all today, we have to revise our thinking of how chemokines guide cells in vivo, it's not actually mostly spatial sensing, it's actually temporal sensing. And in all the systems we've looked at for inflammation, there's a feed-forward loop in which you have a rapidly increasing gradient due to the recruitment of cells that make their own chemoattractants. And so we've actually— we have some models for that in general, but we obviously can't measure all those parameters in vivo, so that becomes becomes an issue. But we can make certain assumptions, simulate, do things in 3D chambers where we think we have some numbers, and then ask whether the models begin to predict the real behavior. Jennifer and I were talking about some of her recent work about stiffness and compliance. That becomes an issue because you start with a relatively stiff tissue, you make it edematous by vascular leak, and now that parameter changes. So you're right, there are many, many, many parameters and not very many of them measured. We were mostly concentrating here on the signaling that occurs when cells are interacting or that control the directionality, and then for the speed part, you know, you're going to have to make certain assumptions. In terms of the length issues, in this LTB4 neutrophil system, the cells were will come in a directional manner from 200 or 300 microns away in the tissue. For most of the other things we've looked at when we were looking in lymph nodes, the turning angle occurs at about 50 microns from the centroid of the cell that's actually producing things. So we know the size of the cells and we know approximately how far the gradient or the chemokines extend. I don't even want to call them gradients for the reason I just said. Scott.

So very pretty. Sort of another temporal aspect question. I'm not sure if it's jet lag induced hallucination or what, but I thought I saw in many of the movies almost an episodic motion of many of the cells, almost like they're moving and then pausing. And I'm wondering if in the modeling there's a move— they decide where to move and and then execute sort of phase.

You weren't dreaming. Okay. It depends on which movies you're looking at. The neutrophils tend to be pretty persistent in what they do. The lymphocytes have been characterized by multiple laboratories as having a kind of stop-and-go motion. So in your model, do you put that implicitly in, and what do you say those different phases do? Different views of what causes that, and one of the things I emphasize is this issue of worrying about what's in the black. If you have things on a structure where they have to turn periodically because that's the intersection of the network, then they're not going to necessarily make a smooth turn. They're going to maybe hesitate at that turning angle. We've actually measured those turning angles to show the cells really do follow the fibers, but unless you co-visualize those and measure measure everything. If you just look in a black space at cells moving and then they sort of stop and go, you think they're pausing in emptiness, you know, and cell intrinsically deciding what to do. But they— we have beautiful movies showing that a whole bunch of cells just run into a blood vessel or a nerve and can't get to the chemokine center. And if you include those in your calculations, you screw it up. So that's why, you know, you have to worry about what's in the black.— but yes, there are different modes of motility. The dark matter. Dark matter.

So next question. So a few of your interesting videos, the macrophages, instead of moving the cell body, it is showing the pseudopodia and reaching to the dying neutrophils. Is there any interaction between those macrophages and the neutrophils from moving the circulating neutrophils to that position or it is preventing them?

We have no evidence that the macrophages attract the neutrophils at this point. I can't— that's no evidence. I just want to be careful how I state that. We know that the neutrophils will come in due to the tissue damage and mediators that are made under those conditions, and we often arrange for them to be in the moving field while we're doing things. Mostly what these macrophages do is to prevent the neutrophils from going into this feed-forward cascade that depends on them dying when they touch the actual damaged material. And we're looking— we don't have evidence for inflammasome activity or pyroptosis at that point, though that's a good candidate for what's going on because there's some evidence that touching a stiff substrate can cause potassium flux across the membrane and activate inflammasomes, and then that would cause the system to operate. So that's what's prevented by the cloaking. You don't get that touching, you don't get the activation of the neutrophil, it doesn't make LTB4, and then the other neutrophils can't make LTB4 and they can't undertake the swarming behavior. In that case, you don't get the damage, because the damage that I showed you is not metalloproteinases, it's actually physical pushing of the collagen fibers away by the mass, the integrin-dependent mass of the smorin.

Thank you. Okay, well, let's thank Ron again for really a spectacular keynote talk.