Scientist Stories: Prisca Liberali, Self-organization in Collective Cell Behaviors

Good morning, everybody. So today I will describe a bit of different spatially resolved methods, and I will mainly focus on imaging and the current state of the art on what we do in the lab, but also on different other methods that are developed worldwide. So to start and to combine with some of the other discussion we had this morning is really a definition what for me is a cell type or a cell state. So for me, So really, cell type is something that is very phenotypically different, so something that looks different. Because you can imagine, a neuron looks very different than an epithelium from a certain organ, from the skin or the gut, or different even immune cells. But in between these very phenotypically different cells that also have different function, they secrete different hormones, we have a very big heterogeneity. This heterogeneity arises from different cellular states that can be cell cycle, as we heard today, it is a very big part of it, but the other also metabolic state that I think are very important. So one question I have is really, if we are going to understand billions or trillions of cells in the body, how many do we really need to sample to understand the state and the type of cell? So this is ideally the best-case scenario because we have one human, and we sample a single human. But real world is different. And without considering human-to-human differences, gender differences, or disease state, our own, we are really developing, regenerating constantly, and unfortunately also aging. And so how can we combine all this? And these are really also transition states that are very important. So as we saw today in the different— in the previous talk in lineages, We don't jump from one state to the other. It's a continuous where we go from being young, and then we slowly develop, and we slowly age, and we regenerate. So how can we deal with this transition? And something I will go back to later on is that this transition might be very fast. So we need to sample enough cells to get this transition. We might have very fast transitions that are also molecular. So I'll come back to this at the end. Then, last but not least, single cells don't do this alone. And as we heard from Gary Nolan, a single cell, and a single cell type and state, is determined by its environment, by its cell-to-cell neighbors. And this self-organization that feedbacks on single cells and drives the collective behavior of an organ and the communication is very important to keep in consideration. It's clear now that A stem cell is not a stem cell. A stem cell has a potential to be a stem cell, and depending on its environment and its neighboring, will become a stem cell and will have such an activity in a certain period of time. But you need an intrinsic potential to be a stem cell to then have your environment. Also because the genotype is clearly defining a phenotype, but we clearly know that this is a clear feedback, that then depending on the phenotype— and this comes to the last part of this slide, that is, What's the cause and the consequence of a cell type or a cell state? What is determining it, and what is just a consequence? So which type of genes get expressed just because this cell type needs to secrete it, and which set of genes is really required to make that cell? And so now I go in my ideal world. So what would I like to measure in every single cell at any instant in time? I would like to look at gene expression, and protein abundance and localization, because this brings us much closer to the phenotype— that is, my definition of a cell type— and really to the phenotype, the activity of the neighbors, but also the microenvironment and the mechanical cue of the tissue surrounding it, because this is clearly driving a lot of differentiation steps. And this will give us information about the cell type, the signaling status, the metabolic status, and it would be also important to understand how can, from a gene expression profile, we can determine the signaling status And this is, for me, still computationally— can I look at gene expression profile and really say in one second, this is a cell that has AKT and ERK signaling activated? But going back to something that I think Dana pointed out, that this dimensionality reduction, but really feature selection— how many measurements do we really need to define a cellular state? And at what resolution? Do we need cellular resolution? Do we need subcellular resolution? And can we really also computationally find these genes? Can we find the genes that could really explain everything? I don't believe that we need a full gene expression profile to define if a cell is in a certain state or not. But there will be key measurements that will be important to define it. So now, going back to the real world, I'll try to give you a bit of overview of what different spatial methods have been used in the past to look both at gene expression and also at protein. So I will go very fast on the gene expression and focus mainly on the protein. That is what we do. So there are many spatial sequencing methods where— and I will go in detail because most of you are the ones that made this work. So you always need landmark or specific orientation, in this case from the tomography. And it's not always one-to-one applicable to everything and every system that we can use. This is why in the years since 1998, we really had an explosion of ways to look at gene expression in single-cell and single-molecular resolution. Starting in '98 with very complex probe, in 2008 Raj's work was really booming the way we started looking at gene expression because of also the simplicity of how we can synthesize this one and how Applicable to most of the samples we use, it is. Then in 2013, with the bDNA-FISH technology, we could move really high content to look at 1,000 transcripts in 1,000 cells. And recently, the MERFISH for really the multiplexing and starting to look at multiple transcripts simultaneously in the single cell. I think the combination of this and being able to move multiplex in multiple cells is really powerful. But what about protein abundance and localization? This is, I think, very important to get closer and closer to the phenotype and to combine all this, and also to go back and to really define what are cause and the consequences of a phenotype. So for this, I won't go through this slide because we had this morning a very good— Gary Nolan really showed his work and how impressive we can look now at, you know, multiple, like 40, 50 proteins simultaneously. However, even if I've seen the talk this morning, this is not really true, but resolution at 3D is still a little limiting, and you need to create all these antibodies that are labeled, that is not trivial for every lab. What we do in the lab is mainly using immunofluorescence, but immunofluorescence at large scale and content, and in a very quantitative way. So here you see a population of cells. Every cell here, the DAPI, and you can really appreciate the cell-to-cell variability that we can have. And really, we can capture most of the heterogeneity in the sample. And we can have an intracellular resolution where we can really follow intracellular localization of most of the protein. And you see here the heterogeneity that can mainly be explained by cellular state, cell cycle, environment, or metabolic state. We can segment all these cells, we can cluster them, we can look at their— and combine both morphological and molecular readout, and we can also move in 3D. So this is mainly what we are doing now in the lab. We work with organoids, we look at the formation of these organoids, and we can really look at single-cell resolution for different amount of days and look at different proteins. So you can see here, we can really follow the formation of organoid for 6 days, we can really acquire hundreds of thousands of organoids and look at their variation between organoids or variation between cells in the same organoid, the abundance of the different cell types. This is just to show we really have single-cell resolution. We can segment all these cells, and we can extract, as I said, morphological and molecular features. What we then used all these features for, and you saw it before in the previous talk, this was one of my works where I can— we were able with Dana and Lucas Pelkmann's lab to really use these features to make trajectories, these trajectories of progression. And there are two things I would like to stress here. First one, when we start working with this, we clearly don't have the depth of information. We have a lot of measurement, but most of this measurement can be correlated. A cell area is quite correlated with the major axis, so we don't have as many independent measurements that we can have with RNA-seq data. But what we have is very reproducible data, and we know what we want. So in this case, we were looking at the cell cycle, so we just need 4 measurements, 4 or 5 depending on the cell type, to have a very accurate trajectory that looks at the progression through the cell cycle. And what was very impressive is that we have very tight transition point. We can sample thousands of cells, so we can sample very well all the transitions. And these transitions are very tight, so we can really look at G1S transition and what is molecular involvement in that case. The other aspect we heard about pseudotime or time progression, I mainly call it molecular progression because it's really not time. You know, we put it here, and for us, This is a pseudotime, so they go to one state and another, but what's happening is really a molecular progression that is a pseudotime. But I think that to consider it really as a molecular progression, that in fact reflects time. So it's a pseudotime, but to imagine it really as a molecular progression in there, and how to combine this with the information we have from live cell imaging. So ideally, again, back to my ideal world, This is what we would like to have. We start from a certain point, we would like to have all the different trajectories, but again, as I said, we have limitations in how many things we can measure by immunofluorescence. Recently, there have been many methods to start multiplexing, so to really start using normal immunofluorescence with normal antibodies, but to be able to have different ways to remove them. So they can be bleached, And so they can be chemical or by light. They can be eluted, or they can be spectral and mixed in the sense that you do a combination of different fluorophores, and then computationally afterwards you just can see the contribution of the different protein in there. These have really— these methods have not been used that much because they're not really applicable right now. And so we have been working a lot in the past years to try to get to a method where we can have a multiplexing assay where we can then measure multiple proteins simultaneously for many rounds. And so the method of choice was eluting. And in this case, you see, this is in collaboration also with Lucas Feldman's lab. So you see here, we can take a plate, we can take thousands of cells, all segmented, and we can have a sequence of immunofluorescence. This is really stripping away the antibody so there is no cross-contamination of like species. And so you can go from one to the other. And just to give you an idea that it's getting out of time, this is the resolution that we get for thousands of cells, and we can have different type of localization. And these are really important to get to the molecular function and to understand the cell type. And here you can see and can appreciate some of the localization. This method is extremely applicable to all different types of settings. These are, for example, yes, differentiation to neurons. And we can look at all different types of transcription factors also in 3D in organoid formation. So this is where we are on fixed sample, the multiplexing. But again, going back, we want to see if this is really reflecting life. Live cell imaging is important because we need to see if our pseudo-time, our molecular progression, is really correlated. And for this, we are also using different methods to look at development for a long period of time. Here we are speaking about almost 6 days of continuous imaging. I will show you 2 colors, and I think in the near future we could probably get to 3, 4, maybe 5, with a lot of imagination. And this, you can then follow the formation of an organoid from a single cell to a fully formed organoid. For 6 days. And then you see we have a cyst-like— this organoid really reflects— this is our intestinal organoid reflecting the function of a normal intestine, and they will form crypts. These crypts really are like in the normal intestine, and you can see we really can look at the formation of an organoid from a single cell, and probably in the future combining also more colors. So just to finalize, these I think are the challenges. I think how to combine the different data sets. These are timescale difference. So how to combine imaging with sequencing at different scales. What are the cause and consequences of phenotype, molecular versus temporal progression, and how to deal with fast transition that brings how many do we really need to sample to get this transition accurately. How many measurements do we need to really get this aspect through, and what's the spatial resolution? With this, I would like to thank you for the attention, and I'm happy to take questions.

Thank you, Prisca. Again, time for one question. Pam Sharma, MD, Anderson. I've spoken to Hans Clevers about these organoids because he's very interested in now how to get the immune system in. From our understanding of how the organoids are, you have to throw a lot of hormones, a lot of factors, a lot of glutamine, a lot of things in to make this work, which of course changes what you're really growing out. You're missing the microbiota, you're missing a lot of other. Do you think that there's a way to try and take this data then to compare to what you're really getting from all of the human samples that you'll get from the donors or the patients that we put in the Human Cell Atlas and try and come up with maybe a signature that's organoid-specific versus what's truly biology?

Yes, I think so. What exactly an organoid recapitulates and it's at which step of development we are looking at and that we clearly miss immune cells, but they're still extremely powerful. And that could be a very big base to have some of the information to understand which are the measurements that we need to then go where we have more limiting samples that maybe come directly from the patient where we cannot do all these intestine. So I think the organ order will be essential to really understand, having a— looking at which measurements are important and using— and then a certain point when we have both, really compare what is really the immune part in the body and having glyco-cultures on our different aspects.

Okay, let's thank Prisca again.