Scientist Stories: Ido Amit, The Power of Immunology in the Age of Single Cell Genomics

We have now come to the last session of this meeting. Our next speaker is one of the most important players in the field of single-cell genomics. Let's welcome Ido Amit, who is from Israel. Ido Amit is a professor at the Immunology Department at the Weizmann Institute of Science. His lab pioneered single-cell genomic technologies and their application to characterize the immune system.

Thank you, Gaojie, for inviting me for the talk here and for organizing the Asia Human Cell Atlas. This is such an important meeting in general and in these days more than ever. So I think we do not need to introduce why understanding immunology these days is of extreme importance importance in really understanding all of the different subsets of immune cells and how they may go weary under different conditions. So relevant to these days, I'm going to start by a bit of introduction of some of our work on COVID-19 and using single-cell approaches to understand how the immunology— immune system is modified under a COVID infection, and specifically under COVID infection that go towards more pathological states. Our entry into this was through work we did over the years to really understand the differences between viruses, cells that are infected with virus and therefore change their activity, so the interaction between the host and the virus, and the bystander effects that are going on in the tissue, given the signaling of the immune system itself towards the cells in the periphery. And the idea is very basic, although it took us some time to generate a robust working pipeline for this. And the idea is to use not only the genes that we align to the human database, or the host database, but also to a very large compendium of viral genome that we assembled. And then the idea is straightforward to try to understand how within our atlas of immune and in general of all of the cells within the tissue, where is the infection introduced, what type of infection, which type of viruses. And as you see, even in our very initial work on COVID, we see a lot of unexpected infections. So I'm going to primarily talk about two datasets that are already published. So work with Xin Zhang and Xu Zhang in Shenzhen Hospital. Again, amazing collaborative work that started by a tweet and our description of such an algorithm and the need to collaborate on this, and they picked up the ball and it was— and it still is a very fantastic collaboration in understanding a viral infection, not only COVID. And then increasing it to larger and larger cohorts that we are now assembling. So one really dramatic finding that we saw using this first single-cell atlas of the lungs of COVID-infected patients, mainly focusing on the difference between patients that have a more mild disease versus the more pathological cases. We saw very dramatic differences both in the compartment, the Myeloid compartment, really the most clear change was the apoptosis of alveolar macrophages and the elimination of the alveolar macrophages from the lung, which keep a more immune-suppressed environment. And their replacement with monocytes and the monocytes who differentiate into macrophages, which are much more inflammatory under these conditions. So that was the first and very apparent and dramatic change, and that is going alongside with secretion of a lot of inflammatory cytokines. And I dare not call it a cytokine storm, so it's a very specific set of cytokines that are secreted there. And the second event, again very dramatic, is that within the mild patient and patients who have a very a decay disease or a disease that is not showing a strong pathology, what we see is specific T cells to the virus, different clonal effector cells and memory cells, whereas in the severe patient, one observes much more naive T cells that do not seem to recognize the virus and generate clonality in phase of it. And again, we're talking about patients that when we see them in the ICU, most of them hardly have any virus at that point, something in the range of 3 weeks post-infection. We can see this basically summarized here. So there is some changes in the CD4 T cells. Again, these naive CD4 T cells are mostly seen in severe patients. There's a huge drop from the mild to severe in these clonal T memory cells and some increase in NK cells. And again, this is what I described in the alveolar macrophages and the change in the— and they're being replaced by monocyte-derived macrophages and monocytes in general. So that was a first in a more limited cohort that is growing at that time. Since then, we collected many, many more patients, and this observation that we had originally is extremely robust. So we see it in every cohort that we look at. Here is a cohort of around 30 patients from the Verona area. This is a collaboration with Vincenzo Ponte and his team from Verona. And, and the same observation holds very strong, a huge decrease from these memory CD8s in the more mild patient to the severe patient having mostly naive T cells. And that is completely correlated with the amount of pathology we would see in the patient, as one can see here in the SOFA curve. As I mentioned, our goal is to really understand the interaction of the virus with the host in these studies. And these are some of the benchmarking experiments we did where we, at that case, knew which virus we're introducing into mice in this case, and that allowed us to really calibrate and improve the algorithm. So we have better specificity. And again, viruses have repeated elements and so on to allow us to remove all types of noise and make the algorithm better. And then going into the real data, what we saw are what we expected. Again, we identified with our algorithm the SARS-CoV-2, again, in some of the patients, mostly in the severe patients. I have to say the levels were not extremely high except for one case that I'll show a bit later. But what was really surprising is that we started to see that there are secondary co-infections, that it was mentioned to other viruses from the same, from coronavirus. And this is, was suggested that this might be on these other viruses what generates a more severe disease. And indeed, in this cohort, we found human metapneumovirus. And in this case, the virus are not in a low amount, but actually take a very substantial amount of the molecules that we identify in the patients. And this diverts, again, the immune system in the lung of these patients very dramatically to a very high type 1 interferon signaling and secretion. And we think that has to do with a non-proper immune response. Again, most of the cases, patients that have these secondary effects do show a very severe outcome and in many cases dead. This is just an example of where, which cells we actually observe these infections. So while most of the corona, the SARS-CoV-2, we identify within epithelial cells, a little bit in macrophages, though we can't separate whether these are macrophages that are actually engulfing the epithelial cells. The pneumonavirus was mostly seen in monocytes. And again, these had a very abrupt and robust type 1 response, which we think interferes with the normal progression towards a productive immune response. Within, if we go to the second cohort, again we saw many co-infections. I'm showing in two patients here the infection were different, or so it wasn't with the pneumonavirus like we've seen in Shenzhen, but actually HSV-1 virus in the cases. So what we think is happening, that once the immune system becomes weak by the coronavirus, it allows for these secondary infection, which then really increase the pathology. Obviously, from this data, one can really learn the structure of the viruses and start to understand how this impacts the immune system and how this influences not only the infected cells but also the bystander cells. So we think this is a very productive direction, and we encourage people to use this algorithm more and more. So just to summarize, and within the study, in the Verona study and the others, we're trying to take these experiments which are starting observational, as we like to do, to move them to more functional experiments. And we do see very dramatic changes in these severe patients. So while in the mild patient what we see as a consequence of the infection, there is also initiation of feedbacks which suppress and are actually checkpoint to stop the immune system, like expression of arginase 1 in various myeloid cells. I'll discuss that a little bit later. And other checkpoints on granulocytes and T cells. Within the severe patient, we don't see these breaks coming on. So if we take from the blood of a mild patient these myeloid cells, they will suppress quite effectively T cell responses, while we don't see that same phenomenon when we take such cells from the blood of the severe cases. So there is still a lot to learn here, but we think this is— a lot of this is driven by this inability of the immune system to stop its activity, almost like autoimmune response, and that with the secondary infection really causes this dramatic pathology. I'd like to move forward to two new technologies that we think are of extreme importance to incorporate into the Human Cell Atlas. A lot of this was also funded and helped by the CZI, but the CZIC networks and so on. So the roots of the HCA, and that is to better understand within complete tissue and the immune system how interaction between cells and how the signaling within the cell is really impacting the overall physiology or pathology. And I think that will allow us to build a more deeper understanding of the atlas we currently have and we're building. So, it's not— I think it's obvious for everyone in this consortium that we need better technology to understand the aspect of time. And as is very clear, the immune system is a very dynamic system, which is changing due to perturbations and diseases, pathologies, and so on. Spatial organization, cell-cell communication, the clonal relationship, being able to understand this data alongside genetic perturbation and the TCR and BCR. And these are all technologies we are constantly trying to develop and expand, as the rest of the community is. And our initial goal to understand communication between various cells within the tissue was this study by Mehrab and Amir, who tried to use ligand-receptor pair to understand the communication of various immune subsets with themselves and with the rest of the lung tissue. And such analysis by our group and others, I think, really allowed to understand new types of biology, which is critical in immunobiology, which is extremely critical in understanding such communication in the lung. And what we found is that through IL-33 signaling, AT2 cells modify and prime basophils to become lung-resident basophils who, using various cytokines, signal to macrophages to adopt them to become alveolar macrophages, which are very critical to restrict overactivation in immune responses, such as is the case of COVID-19. That finding really pushed us to understand that the tools we actually want is to understand how cells communicate with one another in a way that would be deterministic, not using such ligand-receptor pairs which give you a very wide spectrum of potential interaction, but one does not really know which one of them are actually true and happening within the tissue at the point of time that one is looking. And What Mehrab and Amir developed is this approach to understand such interaction, and usually such interaction, for example, between T cells and myeloid cells forming these synapses of antigen presentation or repression of specific T cells is a rare event. And to capture it, it may not be enough to look at ligand-receptor pair, and it may not be enough to do other approaches, but perhaps we need an approach to specifically target those pairs, and that's what may have an immunity. And that basically uses, again, the ability of the facts to look at individual markers. For example, a marker for a T cell, a marker for an antigen-presenting cell, and then look for these rare events where you see both markers on the same conjugate, okay, on the same cell basically, which is two cells that are communicating together. And the basic idea is to sort each one of these different subsets, individual of cell 1, cell 2, and the conjugate, and that helps us to start to understand their signaling. And again, this was published, so I'm going very quickly at it, at this data, but the basic idea is we have these different cells, and initially we sort them as singlets, and this is used to actually simulate all potential doublets. And this simulated doublets allows us to analytically remove all kinds of artifacts in such data, and that is doublets that are formed in vitro and so on and so forth, and really understand which interactions are happening more frequently than we would expect. Which are likely happening in vivo, in situ, in the tissue. And then not only that, but also understand which molecules are overrepresented, meaning they are a part of the likely interaction signaling between these two subsets, and then understand by deconvolution which molecules are relevant to which one of the cell types and which one are induced because of the interaction. Again, this system actually is improving on a daily basis, so we are going more and more into complex human tissue, again making it relevant for the Human Cell Atlas goal. This is just an example of one such cell that we sorted and then visualized, showing this beautiful synapse with the TCR here. Of myeloid cells, genetic cells, and T cells. Where we're going with this, as I mentioned, is going into human tissue. I'll show you a collaboration with the lab of Miriam Merad from Mount Sinai and the lab of Amos Tanai, where we specifically look at such interaction of T cells and myeloid cells in the non-small cell lung cancer tumors and the adjacent tissue, the healthy tissue, if you will. Again, looking myeloid markers, T cell markers, and the peaks. And the goal is to understand which such interactions are happening in TME versus the normal tissue. Long story short, we find very interesting and non-expected interaction. One specific interaction that we are focusing on is an interaction with the, what we call, tumor T helper cells. So it's a specific T helper, specific CD4 to the tumor with the very specific characteristics of molecules, both signaling and checkpoints. Again, this is a CD4 which is completely different from the Treg signature. We can identify it in various sites within the tumor. So this physical interaction. We moved on then to make a mouse model that recapitulates this phenomena and are constantly investigating this CD4 T cells and what potentially it might do function in the tumor and how we can modulate it. So this is a very exciting direction, and again, the entire discovery and understanding was generated because we can start to see such interaction and identify this relatively rare CD4 T cell. But when you look at it in the face of interaction with the APCs and dendritic cells, you see that these are very major cells in the tissue. So in this case, this is such interaction visualized mouse tissue. Moving forward to the last technology. How much time do I have left? Raulji, can you tell me how much time do I have left?

Oh, sorry. Yeah, it's about 5 minutes left.

5 minutes. All right, so it's work that just came out, so I'll try to go fast and finish on time. I know We want to keep a tight schedule here. So again, this is another new technology just published, and I think it has two major areas of importance thinking about the UNC Atlas. On one hand, it can allow us to improve our maps by adding additional information, signaling information, metabolic information, information about transcription factors, and as we know, play a very important role in defining the different cell states that we look at. So this is one obvious advantage. The second advantage is that this fixative also allows us to keep the tissues for a very long time and keep them in a condition that are almost as good as fresh tissues. I'll show that in a second. I think that can be widely used for distributing samples across the world, which is, I think, currently a bit limiting. So the basic idea that we struggled with and have solved, and I think this was a struggle for many, is to actually be able to fix all of the content of the cell protein and so on, while— and RNA— while being able to permeabilize and open the membrane, which allows us to then introduce antibodies and other contents inside. And it— all cases we looked at, all fixative and other reagents we tried, once you permeabilize the cells or fix them, you degrade most of the RNA. Again, with our specific fixative, which is a combination of different salts and solvents, it allows us to actually permeabilize the cell while maintaining the RNA intact. And then one can label the different protein modification and so on with antibodies, either fluorescently labeled, and hopefully want to introduce soon that are labeled by sequence-specific, and then run it through other, or MARS-seq, or through droplet approaches, and understand the context of the transcriptional state in front of various other layers of information, metabolic, transcription factor signaling. Spearheaded by these two brilliant, or actually three brilliant PhD students and Asaf Weiner, staff scientist in lab, and Victoria Fair, postdoc in the lab. I'll go very quickly, I think I'm out of time. Again, several methods and InSec is really allowing you to permeabilize the cell and introduce antibodies, others are not, and what it also does Unlike any of the other known currently fixatives, it allows it to preserve the RNA. And these are the levels that one gets with fresh cells, and this is also preserved RNA for a very long amount of time. Again, it's hundreds and thousands of fold more than any of the other fixatives we've tested. The cell composition Once you run it on 10x, it's exactly similar to the first sample. And basically, one sees the same genes and gene signature that one has on the fresh cells. This is true for a very large number of tissues we've calibrated by now. And these are PBMC. As I mentioned, this allows you now to go and start to understand the context and even sort and focus on specific transcription factor. For example, over here we sorted Tregs from PBMC, and you can see they're quite rare. And only once we sort for FOXP3-positive cells with an antibody, we can really enrich for these cells, almost 60%. So this allows us to then go and do all kinds of combinatorics of different transcription factors and identify various T cell subsets, for example, different memory, rare memory cells in tissues, and so on. Looking at such combination of transcription factors. I think it's going to be extremely enlightening for many of us to push on this approach. It also has a translational aspect because we can start to understand what is the difference in different subsets of cells in specific tissues. For example, we would want to target Tregs only in the tumors and not in the periphery to avoid secondary effects, and this really allows us to do such important discovery. Last point, I'll try to do it in 2 minutes, I see time is running out. Going— better understanding various myeloid subsets, for example, some of the most potent inhibitory myeloid cells, or myeloid-derived suppressive cells, what they're called, are really cells that have no extracellular marker. They're impossible to define by such markers, and only these intracellular metabolic activities are defining these these cells, which make it very difficult to study them and understand them in this context. We always get a very mixed population. And what we did is very simply label these specific metabolic activities. In this case, I'm showing you ARG1-positive myeloid cells. And once we did such an effort, we really identified the various subsets of myeloid cells in which inhibitory activities they conquer. So really breaking the myeloid suppressor cells into very specific groups, which we understand now very well, all of their markers. I go very rapidly on this one. A very interesting receptor which we worked on in the past, which is called TREM2, was found to be very much correlated with such myeloid suppressive activity. As I'm showing here, TREM2 is of high interest to us. Again, we show this signaling pathway is almost not active within immune cells in health, but only within disease. This is— and actually, we had a beautiful picture of how in the human ciliatus, only in brain microglia and the bone, you have a little bit of expression of TREM2, while in pathology, this expression in many areas. And what we've learned over the years is this receptor activates a very robust immunosuppressive activity. We have to stop now. Control the time. One slide and a half. All right. And then again, with knockout, we've shown that tumors would grow much slower without TREM2, again because of its immunosuppressive activity. And that really modifies the suppressive environment within tumor, dysfunctional T cells, and so on and so forth. Um, need to thank the amazing people who did the work, wonderful collaboration as I mentioned with Amos Tanai, Tom Schulmacher, Miriam Merad, and people I mentioned all along, and our great funders. Thank you, and sorry for going a bit over time.

Great talk. Great title. I mean, I really, I'm not an expert on immunology, but really like, you know, all of these fancy technologies that you have, you know, developed to advance the single cell genomics. I think one question I wanted to ask exactly, your beautiful work in combining index sorting with single cell RNA-seq, and how would those results compare to CITE-seq? And is there any interesting thing that is different between the two methodologies?

Yeah, so Sightseek is a great technology and I think now it's really very widely used and that's, I think, allows it to constantly improve. The point that Sightseek is focusing— I would say there's two differences people should consider. So SiteSeeker is for quantifying proteins, and it's for quantifying currently extracellular proteins. So what the approach we like to do is to use the FACS also to zoom in and focus on specific population, right? And for that, you need to see the amount of protein in, in, in, in live. And what these fluorescent markers allow you to do, and, and I have to say, through our very fun and collaborative work with BD. We are constantly increasing the amount of fluorescent markers that you can use and actually on the fly really decide, okay, here are the important cells I want to focus on and really focus on them and only sequence them. You know, the new sorters that are going to come out or are coming out now can do 40 such proteins. That's going to be very important for our community, immune and other. So that's one feature. Okay, so the CySec, you have to predetermine and you quantify, but you cannot use it to— the second difference is that it looks— it only uses the extracellular markers. We are also very interested in the intracellular signaling metabolism, and But the good thing is we hope to merge these and have our technology also relevant for site seek. There's still some problems in the fact that introducing nucleic acid into the cell at this point generates quite a substantial amount of noise, so the signal-to-noise is not as good as with fluorescent, and we hope to solve that by various blocking mechanisms. If people have ideas, we really welcome collaboration on that end.

Great, and then we have a question from the audience. So Xi'an asked, is there a version of the viral track algorithm that uses sequence information from both ends of the paired end read?

Yeah, so this algorithm was developed by a brilliant— actually, I now remember I just introduced yesterday the slides on this because you said you wanted to have us. Yeah, so a brilliant PhD student called Pierre Both, and we are working on that. So that will be introduced hopefully soon. So, but at this point, I think if you go into the current version, I think it will not support this. But I can respond, you know, if it's— and answer the community, if you can forward, when it's expected to come out. Hopefully soon.

Okay, because of time, we have to move to the next speaker.