Scientist Stories: Roser Vento Tormo, Immune Cells in Tissues

Good afternoon, everyone. It's a pleasure to be here and present the work we have been doing in the last 2 years at the Sanger Institute. So our lab has an interest in how cellular microenvironments are shaping the identity and function of a cell. During development, cells organize in complex tissues and organs and is there where they acquire their function. Therefore, the identity of a cell is dependent on the context. So, what we are interested in the lab is to understanding how cells in tissues talk to each other and how this triggers specific regulatory programmes in a cell. So, to address our question, we utilise two main technologies: single-cell and spatial genomics. In single-cell genomics, we get a comprehensive view of a single cell, and this allows us to understand the identity of the cells but also the regulatory programmes that mediate that identity.. And then with spatial genomics, we get spatially resolved information, and this is important to, for example, understand the location of a cell in the tissue, but even more important, the architecture of the tissue. And this is important to understand or infer the cell-cell communication.. And it's the combination of both single-cell and spatial genomics that allow us to obtain these high-resolution maps of development. So, the computational tools is what really allow us to guide ourselves on these maps and to be able to move from one side, to understand how to move from one side to another.. And in particular, in our lab, we have an interest in developing computational tools to understand cell-cell communication from these single-cell genomics datasets. But why is that important to have a catalogue of single cells? So in my view, this catalogue gives us the pieces, which are these single cell information, but also instructions, because we have the cell-cell communication processes, the pathways, transcription factors, and these two things allow us to reconstruct capital junk, or in other words, allow us to improve or even develop in vitro models. So, in vitro models are useful because they are a proxy to the reality. They help us to mechanistically dissect a specific process or to perturb them at a high-throughput manner. And this is of particular interest for scientists like myself studying human organs, because often those are not accessible to us. But of course, in vitro models are also— are often limited to one specific lineage an organ. And it's actually the integration of both in vitro and in vivo models what really give us the more accurate view of the reality. So how we do this process in our lab? So when we want to study a specific process, the first thing we do is to deconstruct these specific tissues by using single-cell and spatial omics. Then we utilise computational tools to integrate all this data and start reconstructing these tissues from these single-cell identities. And then we predict also behaviours of cells. And then we utilise in vitro models to dissect specific processes, or also we utilise these high-value references to start improving the in vitro models in a very iterative manner. And this is where we stand so far, but moving forward, we want also to start utilising in vitro models to interrogate disease and to perform screenings at high-throughput level. And that's possible because due to the exponential growth of the single-cell genomics, we can now multiplex samples and do this in a very high-throughput manner and at accessible cost. And also moving forwards, we are also interested in connecting the different identities and in vitro models to make more complex systems. And this is important because one of the cell types that we we are mostly interested in are the immune cells, and we know that they have the capacity to move from one tissue to another. But for today's presentation, I'm going to focus on the work we have been doing so far to utilise single-cell and spatial methods to profile the tissue, and then reconstruct the tissues and learn about the features of them and utilize all this information to inform the in vitro experiments. And in particular, I'm going to focus on our recent work in the human endometrium. So, the endometrium is the mucosal layer of the uterus and is one of the most intriguing epithelial tissues in the human body. Every month, the epithelium will completely regenerate into these highly structured tissues, and basically the, the main blocks of this tissue are the epithelial cells. So, we have the luminal epithelium, which is here in the surface, and these establish connections with the embryo during implantation.. And then we have these invaginations or glands that are full of secretions and will provide nutrition to the embryo during development. So, the endometrium goes through these cycles of shedding, proliferation, and differentiation, and this is mediated by the two ovarian hormones, estrogen and progesterone. Progesterone. During the proliferative phase, the estrogen induced the regeneration of the tissue. And then after ovulation, there is production of the progesterone by the ovary, and this induced the differentiation of the endometrium. So, this differentiation of the endometrium in humans occur even in the— well, occur in the absence of embryo implantation. And this actually doesn't happen in other mammals such as mouse. And that's why it's so important to study this tissue utilizing humans as its own model. So, the questions that we have in our work are very basic. The first one, we want to understand what are the cell types that make up the endometrium, what is their location, and how are they changing across the menstrual cycle. Then we want to understand the signaling pathways that are mediating this regeneration and differentiation. And finally, we want to utilize the organoids to model this in a dish. So, we came to, to this organ from a very basic perspective of regeneration and how tissues regenerate, but as we started studying, we realized how much of an impact have our study for women's health. And I just want to highlight, for example, endometriosis, which is a painful disease that is suffered for 1 out of 10 women in reproductive age. And so far, there are not non-invasive diagnostic tools to detect this disease. So, in our experimental design, we are combining two types of tissues. So, the first one is biopsies from the functional layer, which is the layer that changes mostly during the menstrual cycle.. And the other type of samples that we are getting are uterine biopsies from the organ donor program. And in this case, we can profile both the functional layer and also the basal layer, which is enriched on progenitor cells. So, we capture that information from different individuals, and then we integrate it with other datasets of single-cell RNA-seq from biopsies from the functional layer. So, we integrated all the data produced by us and by others, and we get this manifold. And the way to look at this is that each dot is representing a single cell, and cells will cluster based on similarities on their transcriptome. And for this particular work, I want us to focus on these two main lineages. The epithelial cells and the stromal cells. So, as you can appreciate by the colours here, the identity of these epithelial and stromal cells, or the cell states that are present, really depends on the menstrual cycle. And this is because those subsets can respond to hormones. So, what we want to understand is how epithelial cells can respond to hormones and differentiate into ciliated cells, which are enriched in the lumen, or into secretory cells that are enriched in the glands, and what are the transcription factors mediating this process. We also want to understand how— what is the cell-cell communication that mediates the activation of these transcription factors, and that could be mediated by the epithelial cells themselves, but also by these stromal cells that are able to respond to hormones. So, let me first introduce the different players, and the first one are the stromal cells. So, as I mentioned earlier, the stromal cells can respond to hormones. And here we have the endometrial stromal cells enriched in the proliferative phase, and here we have the ones that are decidualized or differentiated and in the secretory phase, and you can see how they express specific molecules. So, in addition, in our study, we identified this novel population of stromal cells that we called fibroblasts 7. And what we did to know more about them was to look for the spatial transcriptomics data.. And in particular here, what we did is to integrate spatial transcriptomics with single-cell genomics utilizing cell-to-location. And as you can see here, these fibrocytes that we discovered are enriched in the basal layer of the endometrium. And indeed, if you look at the transcriptome program, we see a lot of molecules such as RES, for example, from the Wnt pathway, which is essential to maintain the stem cell niche. But then let's move into the epithelial cells, which are the ones that we mostly care about. So, during the differentiated phase, that is called the secretory phase, we have these highly structured glands that are full of secretory cells. And then we also have the lumen where we have few secretory, but we mostly have ciliated cells that will spread the secretions into the uterus. So, here is how we get this manifold when we zoom into the epithelial cells. And as I mentioned before, the cells are highly dependent on the menstrual phase. So, in the proliferative phase, we discover these two Sox9 populations, the ones that are Lgr5 negative and the ones that are Lgr5 positive. The ones that are Lgr5 positive sit close to the luminal cells, while the Lgr5 negative sits close to the glandular cells. So then we decided to locate them in the tissue to know more about them. So again, we use Cell2Location to deconvolve these spatial transcriptomics data.. And we can nicely see how the Lgr5 negative are sitting in the basal layer while the Lgr5 positive are sitting in the luminal layer. And we further validated this using high-resolution microscopy. And again, we can see markers for this luminal population, SOX9 positive, Lgr5 positive, expressed in the lumen. Similarly, we can do the same exercise for the secretory phase where the tissue is differentiated, and again we can see the luminal in the lumen and the glands where the glands are. And what is relevant for this is that then we can start extracting markers for these populations and also validating these at the protein level, which means that we can also in the future sort them and study their function. And that's important because that wasn't known before. But then we thought, can we utilize all these healthy maps to understand more about endometrial disorders? And here we focus on two main disorders, endometriosis. And in this case, what happens is that the endometrium is able to migrate to peripheral tissues and proliferate there. And then we also focus on endometrial cancer. So, for the first disorder, the endometriosis, there was available microarray data of the lesions, and we wanted to understand what is the composition of the lesions based on our dataset.. And what we found is that there is an enrichment of these SOX9-positive, LTR5-positive progenitor in the lesions, just highlighting this could be the population that is migrating. So, for the cancer, we utilise bulk RNA-seq data from TCTA. And here, what we did is to deconvolve this data using single-cell transcriptomic signatures.. And what we could see is that the majority of cancers have this SOX9-positive population as, as the main component. And actually, we can divide the endometrial adenocarcinomas in two main signatures. The ones that are SOX9-positive, ALGER5-negative, where there is a lot of endometrioid cancers,. And then the one that is SOX9 positive, LGR5 positive, where all the serous endometrial cancer sits together with a few of the endometrioid. And we know that serous ones are more aggressive than the endometrioid, and we wanted to know then if this LGR5 signature was related with higher stages of the cancer. And indeed, that was the case. As you can see here, we could see how higher stages of the cancer are related with or associated with this LTR5-positive population. Therefore, as the main conclusion of this first part of our work, we have been utilizing single-cell and spatial genomics to reconstruct the regeneration of the human endometrium and found that these populations in the proliferative phase are also enriched in some endometrial disorders. But our next question was, what are the pathways that are regulating the ciliated and secretory lineages? So, to do that, what we look at is the transcription factors, and we look at their expression but also at their activity by using Dorothea. And in this case, what we are looking at are the genes that are targets for these transcription factors. So what we see is that in the ciliated lineage, we have transcription factors that are known to be activated by Wnt, while in the secretory lineage, we have transcription factors that are known to be activated by Notch. So then we wanted to see if there is a disbalance of Wnt and Notch signalling in the tissue itself. And again, we made use of the spatial transcriptomics information. And here we did not integrate with single cell, just with the spatial. We did clustering on the epithelial spots. And what we could nicely see is that the clusters really depend on their location. Just highlighting how important the microenvironment and the spatial location is for those cells to acquire their identity. And again, what we could see is that in the lumen we have the Wnt pathway, the canonical Wnt pathway active, while in the glands we have more a NOT signal. And we validated this or quantify these using high-resolution microscopy, again demonstrating that Notch is higher in the glands and canonical Wnt is higher in the lumen. So, we have then an opportunity to look for the cell-cell communication that is different in the lumen and the glands and that is allowing the cells to acquire their identity. And to do so, we used CellphoneDB, which is a database that I developed during my postdoctoral stay in the Sarah Teichmann team. And basically, this database of ligand receptors and their interactions is also coupled with, with a statistical framework to look for the cell ligand receptors that are enriched in specific cell types.. But for this study, we also have spatial data, and then we utilise this to improve our Cell PhoneDB algorithm by also considering the cells that are coexisting in the same microenvironment. So therefore, this combination of specificity of the ligand receptors and also the coexistence in the same microenvironment allow us to get high-confidence cell-cell interactions. So, we utilise this in this specific study to look for the Wnt and Notch signalling. So, let us start by Notch. So, what we find is that Notch, for example here Jag1, which is the ligand of Notch, is really highly expressed by epithelial cells in the lumen, and all the Notch signal is mainly mediated by the epithelial cells themselves.. So, regulation of Notch signal is a bit complex, but there is a recent study that was done in the gut where they say that depending on where JAK1 and Hey1 are expressed, you get activatory signal or inhibitory signal. So, for example, if they are expressed in adjacent cells, then you have an inhibitory, and if they are expressed in the same cell, you have activatory signal. So then we thought to see what was happening in our case. So when we stain the luminal epithelium using Jag1 and Hey1, actually we can see that the cells— the signal for Jag1 and Hey1 tend to happen in cells that are adjacent, which means that we have an inhibitory signal.. And this will explain how we have not inhibition in the lumen and not activation in the glands. But what happens with the Wnt signalling? So, in the case of Wnt signalling, we see that the majority of the regulation are coming from the surrounding stroma cells that I presented before, and that I also mentioned that they can differentiate and respond to hormones.. So, what happens is that when these stromal cells differentiate or decidualize, they start expressing high levels of DKK1, which is a potent inhibitor of Wnt. And as you can visualize here, that this DKK1 is really expressed on the surrounding cells of the glands. And also, the epithelial cells in the glands start downregulating the receptor of Wnt. So, altogether what we have been seeing here is that we have in the differentiated or decidualised tissue two main regions: the lumen, which is higher in Wnt and lower in Notch, and the gland, that is higher in Notch and lower in Wnt. And how the Notch signalling, the Notch inhibitory signalling in the lumen is mediated mainly by the epithelial cells themselves, while this highly inhibitory Wnt in the glands is mainly mediated by the surrounding stromal cells. But then we ask ourselves, how does Wnt nod influence the differentiation of epithelial lineages? So, to resolve that question, we collaborated with Margarita Turco in Cambridge University. So, Margarita Turco developed these endometrial organoids that are derived from primary tissue and that they can respond to hormones. So, in response to estrogen, and they recapitulate the proliferative phase, and in response to progesterone, they recapitulate the secretory phase. So, the first thing we did was to benchmark this model to first understand how robust it is and also how similar it is from the in vivo data. So we did single-cell transcriptomics at multiple time points and multiple individuals, and then we did cell clustering and we looked for markers of the specific cell states. In a first instance, we could see that the main clusters that we will expect for epithelial cells, such as ciliated and secretory are there. But now we have the opportunity to do this exercise in a more quantitative manner. So, what we did is to train a logistic regression model based on our in vivo data, and we projected this data over the in vitro system. And what we could see is that the estrogen-induced populations are mainly comprised by Sox9 population, the Sox9 progenitors, and then they can differentiate into pre-ciliated and ciliated with a really high match, and/or to secretory. And what we could appreciate here is that in the secretory we have Sox9 population differentiated into the secretory or glandular cells. So that was relevant, and that is indicating that not all the secretory cells are fully differentiated, but they are in the process to do that. And the other thing that we did is to derive clonal organoids, and we really demonstrated that there is a common progenitor for both ciliated and secretory cells. So, we mentioned that in the secretory phase, not all cells are secretory, but what we care about is not how well differentiated these in vitro models are, but if during the differentiation the programs that are being activated are the same that the ones that are active in vivo. So, in other words, we want to show that the cells respond to hormones in a similar way in vivo and in vitro. So to do that, we calculated transcription factor expression and activity as before but in the organoids. And once again, we can see that in the ciliated lineage we have more higher Wnt, while in the glands or secretions we have higher Notch. Well, that was great, and that means we can utilise the organoids to resolve our questions. So what we want to understand is how the lumen and the glands are shaping the ciliated and glandular lineages. And to recapitulate this, what we did is to inhibit Notch and Wnt with small molecules and to put them together with a hormone. And first, we just did imaging, and let me walk through it. So here we have NOT inhibition, two NOT inhibitors, and this is recapitulating the lumen where you have more Wnt and you have very few NOT that is inhibited by the epithelial cells themselves. And very nicely, we can see how we have a lot of ciliated cells and we have very few secretory ones. So, the opposite occurs when we have a Wnt inhibitor, and this is recapitulating the glands where we have the surrounding stromal cells producing Wnt inhibitor. And we can see very few ciliated cells and a lot of secretory. And here again, we have the opportunity to be more quantitative using single-cell genomics. So, we performed single-cell genomics on different time points, on different individuals, and different conditions with Wnt and with hormones. And what we can see is that we can define clusters, and for desylated clusters we only have one, but for the secretory, depending on the inhibition, we have one cluster or others, meaning that it's not only affecting the efficiency but also the identity of a cell. So with single cell, as I was saying before, we have the opportunity to quantify. So we can quantify efficiency, and we could see that Ciliated cells again are highly produced in the presence of Notch, and the opposite happens in the presence of Wnt. But also we can quantify the accuracy. And that here, as I mentioned before, we have new clusters of secretory cells popping up in the presence of Wnt inhibitor, and we could see how those are expressing higher levels of secretory markers, such as PAP, for example, or DFP1. And that means that in the presence of Wnt, we have cells that are secreting more and being more similar to the glands. And actually, we validated this also at the protein level by quantifying the secretions. But even more interesting, we're going to start exploring the interactions between hormones and pathways. So, for example, we only see the effects of Wnt inhibitor in the presence of oestrogen and progesterone. And when looking at the transcription factors, we see that transcription factors induced by Wnt are present in both cases. However, in the absence of hormones, we don't see upregulation of Notch factors, which are essential for the secretory lineage. So, altogether now we have completed our map and we know that in the lumen we have a higher Wnt inhibitor, which is pushing the efficiency towards the ciliated lineage, and then in the glands we have a higher Notch, which is really pushing the efficiency but also the accuracy towards the secretory lineage. And moving forward, we can utilise all these regulatory networks that we have built in vivo and in vitro and comparing both conditions to start improving the models. And then we can also derive all these models and the ones that we improve from pathological conditions to start understanding disease. And today I have focused on the work in the endometrium, but in the lab we have similar projects in other developmental tissues. And just to finalise, I would really like to acknowledge my fantastic team, and in particular for this work, it was led by Luz García Alonso and Louis-François Rempfle, which have been fantastic. And this was a close collaboration with other teams, including the Biodactyl team with all the imaging, the Taiwan team, and the Margarita Turko team in Cambridge University. And thank you everyone, and happy to take on questions.