 I'm happy to introduce the second speaker of today, Barbara Treutlein from ETH Zurich, from the Department of Biosystem Science and Engineering in Basel. That's also my department. And we were very excited and happy when Barbara joined our department last year because she is a star in the field of single cell approaches. Barbara is a chemist by education and did her PhD on single molecule biophysics at LMU Munich. And then did a postdoc at Stanford University and became a Max Bank research group leader and a tenure-track assistant professor at Munich. And then last year joined ETH Zurich. Her group uses and developed single cell genomics approaches and in combination with stem cell-based two and three-dimensional culture systems to study human organogenesis. I think we'll learn more about this in her upcoming talk. She has won numerous awards. I'll list a few of them here, the Friedmund Neumann Prize of the Sharing Foundation, the Dr. Susan Lim Award 2019 for Outstanding Young Investigator of the International Society of Stem Cell Research. She's an EMBO young investigator and she also won the Young Investigator Award of the German Stem Cell Network this year. So she's really a great expert and world leading expert whom we welcome here to our summer school and we are excited to learn about your work, Barbara, and to see and explore where the connections to machine learning are. Welcome. Thank you very much, Carlsten, for this great introduction and I mean, as you mentioned, this is also I have to confess, I'm not a machine learner and I'm not a computational biologist. But yeah, I'm very much an experimentalist. But we, of course, generate a lot of data that is quite interesting and that machine learning approaches. And yeah, so I'm very happy to be here and present our work on organ development and regeneration through the lens of single cell genomics. So generally, we are interested in scenarios in biology where cells transform their identity and there are different areas, for example, development where you have a stem or progenitor cell that can transform into a mature differentiated cell type or different types of differentiated cell types. Then there's also regeneration, reprogramming, and disease where, again, you start off with a certain cell state and that cell state gives rise to other cell states, sometimes even very unrelated, different somatic cell state or due to a mutation, for example, it drifts off into a disease-associated cell state. So in all these scenarios, actually, we and others have shown that single cell technologies, in particular, single cell genomic methods are highly powerful in illuminating the intermediate stages that the cells go through. And by then applying these single cell approaches, we can learn what are intermediates, what genes are responsible for certain or are involved in certain processes. And yeah, so we can really understand more on a mechanistic level these cellular transformations. And we look at this in different areas where we have, as one major area, human organ development. But there's also reprogramming where one faces forces a transformation of a cell state, such as a fibroblast or a parasite, into a completely different cell state, such as, for example, a neuron. And then we also look at regeneration of organs. And today, I want to focus on these two aspects, human organ development and also, at the end, a bit talk about our work on regeneration. So starting with the human organ development, it's actually a really exciting time to look at human organs and organ development, in particular, because we are not limited to obtaining a primary material. But we can also model human development in the dish. So we can take a somatic cell from a human individual, reprogramming into induced purport and stem cells. And these IPS cells can then be differentiated into different lineages of the human body. But we can also use their self-organizing capacity to actually grow human tissues in vitro. And these are these so-called organoids, you for sure have all heard about it. One can also grow organoids from adult stem cells. This rather mimics then regeneration and maintenance homeostasis processes. We are very much focusing on this branch that mimics development. And these organoids are super exciting systems to us because they present avatars for healthy but also disease organ development. They are specific to an individual. We can grow them pretty much from any human individual. We can perform time-cost measurements. This you can't necessarily do with a primary tissue where you obtain a tissue and then you use it up in an experiment and then it's gone and you can't go back for the same individual to different time points. It is a genetically tractable. It grows in controlled environments and it's amenable to high footwork screening. So all in all, these are really exciting systems to us. And you can see there are different organoid models that we have growing in the lab. And today I want to very much focus on these brain organoids. But before I go into detail there, I want to just show you what we can actually, how we can use single cell approaches. And I'm sure you heard some things, I think Aviv gave the talk already about the human cell atlas and you know, currently. Not yet, not yet. Not yet, okay. So she won't really talk about it. It's yet to come. It's yet to come. But now Nikolaus Rajewski last year. Yeah, I mean, yeah, you for sure heard about it. There's a big international project on going the human cell atlas where we try to profile all the, all the cells that are existing in the human body using single cell approaches. And you know, when all these reference atlas emerge, what we can do is we can profile our engineered cells and tissues. And then really using this highly dimensional data to compare primary and in vitro systems. And then we can learn more about our in vitro systems and potentially also enhance or improve it in the future. So for example, we might get, profile the cell composition of a primary reference and an organoid. We might see that most cell types are found in both, but then there might be some cell types that are only found in a primary reference. So this is missing in the organoid or there might be also cells that are found in the organoid that are not found in the primary reference, which might be then off targets that emerge. So we can use these approaches to them or these comparisons to make the organoid in the future better by introducing the missing lineage and trying to prevent an off target. Also, we can reconstruct differentiation trajectories using this single cell data because we capture intermediates. We can obtain a differentiation pseudo time, let's say for the in vitro system and the in vivo system, we can align these pseudo times and then calculate similarities. And we might see, for example, that after a given maturation level, in vitro cells don't recapitulate in vivo situations anymore. So we might have a failure in the final maturation of the tissue and then we can identify which genes might be involved and we can try to help the organoid mature better. So this is kind of giving a broader picture. In the past, we applied these kind of ideas to brain organoids, these so-called cerebral organoids and also to liver organoids. And this is just giving you a short summary. For these brain organoids, we very much focused on human cortical region in these organoids and then analyzed these cortical regions in organoids and also in the fetal tissue using single cell RNA sequencing. And then in this network, you can see a core embedding of the primary and the organoid cells. And you can see that cells are here organized based on similarity of their transcriptomes. So this is a core expression, correlation network of these individual cells. And you can see that these cells very nicely mix. So this really shows that these organoid cortical cells are highly similar to the primary cortical cells. And so this was exciting. For the liver organoids, we, in addition to profiling human, in this case, actually even vascularized liver organoids in three dimensions, we also profiled hepatocytes that you can grow in two-dimensional culture in vitro. And so these cells in 2D and in these organoids, these are all hepatic cells. They start off at the same state, the definitive endoderm and hepatic endoderm. And then as you place them into the organoid, they actually mature on a different trajectory than if you keep them in a two-dimensional culture. This was very interesting to see. And when we compare to a primary reference of adult or fetal human hepatocytes, we could see that in the 3D culture, in this organoid culture, these hepatocytes are more similar to fetal human hepatocytes. Then if you grow them in 2D culture, and also in general, these organoids are similar to fetal hepatocytes rather than adult hepatocytes. So this was very interesting. And this shows that it really makes sense to study human development in terms of these organoids in three-dimensional systems where you have multiple lineages available rather than differentiating cells into dimensional culture, which is sometimes much easier. But then you really lack the interaction with other lineages. So based on that, today I want to talk more about how we use a combination of these organoids and single cell technologies to study different aspects of developmental mechanisms, but also look at what could go wrong in disease when a mutation occurs, for example, and how the human brain, in this case, has evolved uniquely as compared to our closest living relatives, the great apes. And so I want to focus on this evolutionary aspect and this mechanistic aspect. So here you can see a cerebral organoid cross-section through a cerebral organoid stain for Plexix and C-tip 2, which marks these radioclea cells in the cortex and early bone excitatory neurons in the human cortex. And you can see that, for example, this region here very nicely recapitulates the layered architecture of a developing human cortex, but there are all kinds of regions in this tissue and you can appreciate that this is a quite heterogeneous tissue. And so we started by asking what are all the cell states that actually emerge in these cerebral organoids, which we grow in a very unguided way. So we aggregate the embryo that give rise to any playoff, we then push it into the neuroectoderm and then pretty much provide a differentiation media that is very unguided. So we don't push into any brain region, but we let the organoid go through the whole morphogenesis and patterning. And yeah, so using this protocol, we then ask what are the cell states that emerge and we performed a time-cost single-cell RNA-seq experiment from to a potency through these stages up to four-month-old organoids. And this is a summary of the data. Again, you can see such a network graph, each dot is a single cell and this is a false-directed layout of a KNN graph. And you can see that we start from pluripotent cells and go quite homogeneously through a neuroectoderm in your epithelial state and then give rise to this group of neuroprogenitor cells that give rise to different branches, which actually reflect different neuronal cell types. And these different neuronal cell types also correspond to different regional identities. So for example, cortical, excitatory neurons, inhibitory neurons of the ganglionic eminence, there are some mid and hind-brain-derived neurons, also some very few de-encecalic neurons. And we have astrocytes that emerge later in the organoids at four months mainly, then we have some mesenchymal supporting cells and also some retinal cells that we see not in every organoid, but once in a while. So actually, how do we annotate such a graph? Just to go quickly through it, you can of course go gene by gene through this data and see where's the gene expressed. So for example, FoxG1 is marking the whole telencephalon. This is in C2 data of the developing mouse. And so you can appreciate that all these cells express FoxG1, which very likely shows that they are telencephalic cells. So they are cells of the forebrain. Then we can go through, and for example, neurod6 marks neurons in the cortex, whereas DLX5 marks neurons in the ganglionic eminence. So we can see very clearly here, we have our cortical cells in our ganglionic eminence derived cells. However, we weren't really satisfied going one by one through these genes and especially other regional identities are much harder to identify. So what we, or what Jonas in the group developed is a way to use this in C2 data that I just showed you, not just by looking at it and comparing it to our data, but actually this data includes more than 2000 genes that are spatially mapped across the whole mouse brain at various stages of mouse brain development. And so what this represents is actually a spatial transcriptome atlas where we have all these different regions in the brain covered. And Jonas generated a digital version of this where now you can look at the expression in the mouse brain in this, can look at all the voxels that were profiled here. And what we then do is we use our organoid data and we can project the organoid cells to this three-dimensional voxel map. And in that way identify whether a cell is a cortical cell, derived from the ventral telencephalon, the dencephalon, or also mid or hind brain. And we can also take clusters of single cells as we can kind of obtain, for example here, you see this U-map embedding of the organoid data, we can take clusters of cells and then correlate these clusters with all the voxels and obtain kind of a correlation landscape of these clusters. So this cluster is the cluster of cortical neurons. We have the GE inhibitory neurons, thalamic neurons and mid brain neurons. And so this very much helps to unbiasedly annotate these organoid cells, which since they grow in vitro, it's really hard sometimes to assess what is actually growing in there. And so this is how we came up with these labels. And after having profiled this whole development from fluency, we were actually wondering how do these cell states vary across different individuals and different lines that one might start with. So in this case, we took now seven different pluripotent stem cell lines and grew two month old organoids and then profiled them at two months. And what you can see here is already the completely integrated data, where again, just like before, so this is actually pulling out the neuro-progenitors and the neurons only. And what you can see is again, there are these three major branches in this network and they reflect the differentiation of cortical NPCs to cortical neurons. We have again a big branch of these ventral tenencephalic cells and then we have in the middle here a branch of all the other dencephalic mid brain and hind brain differentiations. And so what you can see is that these different lines actually contribute more or less to all these states but at very variable proportions. So this is quantified here where you can see that for each bar, this is one organoid and then you can see the composition of the organoid and you can see there's organoid to organoid variability. In addition to that, there's actually also batch to batch variability. So this was two organoids from one batch and these are three organoids from another batch. So you have this variability which makes it sometimes hard to work with these organoids. But the good thing is that since we generate the single cell data, we can pull out all the cells for example that contribute to cortex development in all these organoids. We can order the cells along a developmental pseudo-time from the NPCs to the neurons and then we can ask how our genes expressed across this pseudo-time. So this is shown down here. You can see how for every of these lines, we measure the expression of these different genes, like three EOMes, neurodegesis and pre-R2 along the cortical pseudo-time and you can see that these expression profiles are highly correlated. So despite the fact that these different lines and maybe different individuals contribute differentially to the different cell types that emerge in the organoid, if you focus on one cell type and of course with single cell methods we can pull out these relevant cell types, then you can see that the expression profiles are highly consistent. And so this for us provided a baseline for now actually going into other species. And what we wanted to do here is to see whether we can use organoids to identify human specific features of brain development. So this is based on, this was a very exciting collaboration with the CAMPLab. So in great CAMPLab, this was Michael Boyle, a PhD student and from my lab, Chisholm He, a computational postdoc and Sabina Kanton, a PhD student that is actually now moving on to a postdoc. They work together to now in addition to the human organoids grow also chimpanzee and macaque organoids, brain organoids and then profile all these using single cell RNAseq but also single cell ATACseq. And so we created this grade A, Atlas of Grade A Brain Development. And as a background to this, we showed previously that indeed you can grow cerebral organoids also from chimpanzee IPS lines, also macaque IPS lines, these organoids. So here now focusing on cortical region in chimpanzee and human organoids. These organoids are highly similar, which is very encouraging because that's what we would expect since chimpanzees are our closest living relative. And this suggests that chimpanzees are organoids model chimpanzee brain development. And so we think we can use this approach of organoids to make these comparisons. So now also for chimpanzees, we wanted to profile this development from IPSCs to four month old organoids. And this is the data that we obtained there using single cell transcriptomics. Again, we start down here at the pluripotency state. And then we go through these early development stages to neuroprogenital state that then gives rise to different neuronal identities. And just to compare the human and the chimp data, you can see that in both cases we have these neuroprogenital cells marked by Glide III. We have cortical neurons marked by NeuroD6 and also ventral telenzophilic neurons marked by DLX1. So now what we want to do is we want to compare human and chimpanzee development. In this case, we wanted to focus on the cortex where we see a lot of cells in both species. So this is again, this is this branch on the right. And we didn't wanna simply take all the cortical neurons here and there and then do a differential expression analysis between these two clusters. But we wanted to really find matching cell states in both species since this is a developing system that is very much changing rapidly. And so what we did was we generated a pseudo temporal ordering of the cells from pluripotency to the cortical most mature neurons in human and chimp and then aligned these times using dynamic time warping. And what we found was that early on, from the pluripotency, roughly to the neurodegenerate cell state, we can see a good concordance of both pseudo times. So we are moving here along the diagonal, but starting at the neurodegenerate state in the cortex and then going to the neurons in the cortex, we see a divergence from the diagonal which shows that actually in this case, the neurons in the human organoids have a delayed maturation. So at a pseudo time hundred of human, we only can match these cells to a pseudo time of roughly 70 in the chimpanzee. And if we look at these networks, work graphs, this is how this then looks. The most mature state in the human organoids corresponds to only a kind of a state here in the middle of these chimp cortical neurons that emerge. And so we wanted to look more into that. That was very interesting to us. And what we realized actually was that, and you could probably see it in the graph, the chimpanzee graph shows this kind of bifurcation among the cortical neurons. And this is actually a distinction between deep layer and upper layer neurons that have already diversified in the four month old organoids. Whereas in humans, we don't see that yet, which very much supports again the maturation of the neurons in the human organoids is delayed. Also, if we calculate the neuron projection score, that would be a measure for maturation of the neurons. And we compare that between the species along an unaligned pseudo time, we can see that the chimps cells reach a higher maturity. So this very much supported that the neurons mature less fast in the human organoids compared to chimpanzee. So now after actually having aligned the pseudo times, we can throw out all the cells in chimpanzee, actually also in macaque, which seems to develop even faster, that don't find a matching state in the other species. And then really look only at the states that are matching up between the species. When we aligned the pseudo time, we now can see, for example, for the expression profile of these three genes that they very nicely align, this peak of humans at the intermediate progenital state very nicely matches up. So we are quite confident if we now calculate differential expression, we don't get any artifact due to comparing different maturation states. So this is what we then did. And we obtained seven different clusters of genes that you see the number of genes in each cluster in the parenthesis and where you can now see different differential expression, different pseudo temporal differential expression patterns. So for example, this cluster of genes has a uniquely high human expression in the cortical progenitors compared to chimp in macaque, here's an example gene. Then we have genes that are uniquely highly expressed in the cortical intermediate progenitors in humans compared to the other species. The same we have for the cortical neurons. And then we also have genes that are highly expressed across the whole pseudo temporal trajectory. But we also have genes that are uniquely lowly expressed in humans compared to the other species in either the neurons or the progenitors. So this is a catalog of very interesting genes that we identified, but we wanted to go further and ask how is this differential expression potentially regulated? And so this is where the single set of taxic analysis comes in, so we profiled also these different species, organoids from these different species using single set attack in order to identify regulatory regions that are active, that are open in all of these species during brain organoid development. And so just to show you this data a little bit for human and chimpanzee, you can see again now individual cells as dots in a T-sne embedding and you can see that this is now based on the single set attack data where you can again arrange cell on a developmental trajectory from pluripotency up to cortical neuronal cells and you can then look at transcription factor binding site enrichments across this pseudo time and we can identify transcription factors that, yeah, well, we see an enrichment in the binding sites. We can do the same thing for chimpanzee and then we can of course also get a pseudo time and compare and calculate differential accessibility between these species. And so I want to walk you through that data a little bit based on this graph that I showed you. So what this shows is pretty much just a jitter plot for all the DA peaks that we identified. So all the differentially accessible peaks and then we can ask which ones are more accessible in humans. You go up here or which ones are less accessible in humans, meaning more accessible in chimps and you go down here. And so you go down in the number of DA peaks. Then we can actually ask of all these high accessible peaks in specifically to humans, how are they linked to genes that are differentially expressed? And this is shown here. So actually we find that, so we link these DA peaks to the nearby genes just based on proximity in the genome. And we can see actually that the majority of differentially expressed genes links to a differentially accessible enhancer. So I say enhancer because these are mainly distal regions. There are a few promoter regions as well. But roughly 70% of the genes link to a differentially accessible region. Then we can ask are these DA peaks. So now we are up here or up here. Now we can ask are these DA peaks found across the whole development from pluripotency to the formative organoids or do we see this very specifically in the neuro-progenitor cells or in the neurons? And when we look at that, we can see actually that most of the differentially accessible peaks are specific to these stages in the brain, cell stages in the brain and not so much to these early developmental cell states which makes these are of course more interesting to us in this case. And we want to link differentially accessible peaks that we filtered out now to genomic features, to genomic signatures. And here one thing we looked at was whether these differentially accessible peaks somehow have an enrichment of single nucleotide changes that would be fixed across all humans and different from the other species. And we can see actually that the DA peaks compared to non-DA random peaks have really an enrichment of single nucleotide changes which suggests that changes in the genome are then leading to differential accessibility which leads to differential expression. And so finally I want to show a few examples. So we can now using this data and the linking to the DAE genes we can pull out very interesting genes and corresponding enhancers or putative enhancers. For example, we can see cell state specific gains. So in this case, for example, we see a gain in human neuro-progenitor states cells and we don't find this in any other species or we can see neuron specific gain of an enhancer. We can also see human specific loss. So in this case, this accessibility is not seen at all in humans whereas it is seen in the other species. And then in this case, for example, this is a very interesting gene, Katerian 7 that is specifically highly expressed in human neurons not in the progenitors and not in the other species. Then we see that there is a DA peak nearby that is specifically accessible in human neurons not in the progenitors and also not in the other species. And in addition to these expression and accessibility changes, we see that this region lies in a human accelerated region and there are all kinds of fixed single nucleotide changes present. And so this makes this, for example, a very interesting enhancer. And currently we are working towards a reporter essay in the super organs to confirm these things. And so with this, I'm at the end of this part where I showed you how we use these grade A brain-organoid atlases to identify human specific features. This whole data is browsable on this webpage. It's a shiny app and we published this last year. So I want to go back to this network graph and we present the single cell data in this way. And this very nicely shows these or visualizes differentiation trajectories and how cells transform from one to another identity. But what this cannot show directly is any lineage relationship. And this is always a bit unsatisfactory to us because we would want to know which progenitors give rise to which neuronal lineage and also when do cells actually become committed or restricted to a given cell state, a neuronal cell state. What one can do to address these questions is one can pull out certain genes that may be here transcription factors that mark just one of the trajectories. And then we can trace back which progenitors express this transcription factors. And that might tell us something about these being the progenitor to this state and these being the progenitors. Yeah, so this is one way to do it, but it's of course indirect. We can also employ a RNA velocity which I'm sure you have heard of. So in this case, one uses the intronic reads in addition to the axonic reads that we can sequence. And the intronic reads since these reads pre-splicing tell us something about the most newly generated transcripts. And so it tells us something about the kind of, it gives us a temporal measure of the transcriptome and it gives us a directionality of each cell's transcriptome. Where will it go in the future? And so we can use this. So this was developed by Julia Lamano, Stan Limousen and Peter Katzenko. We can use this to put these arrows onto our graph. And so we can see clearly the cells drift towards these neuronal and astrocytes states. But we can, for every individual cell here that is shown in large, we can calculate a transition probability matrix and see what is the most likely future state. So in this case, for example, the most likely future state of that cell is a ventrotinencephalic neuron. The most likely future state for that cell is a cortical state and so on. But still, this is very nice and exciting, but we can mainly link quite nearby states. And we can't make any statement about commitments of early stages in, with respect to what neuron they might become. And so here, again, we teamed up with Bray Camp and we wanted to develop a system where we can directly measure lineage relationships in organoids. So this has been great teamwork by two postdocs and two postdocs, Rebecca, Petri and Chisholm here and then two PhD students in the lab, Tobias, Galba and Ashley Maynard. And what they developed is this eye tracer system where we bring into IPS cells where we can induce Cas9 expression with doxycycline. We bring into these cells a vector that contains barcoded GFP. So this barcode has a high diversity and such that, in principle, every cell gets a unique barcode. And then in addition to this barcoded GFP, we have a guide RNA on the same vector driven by the U6 promoter. And so upon induction of Cas9, this guide RNA will guide the Cas9 to the GFP locus and induce the scar. And so this scar, if it has high enough diversity, which we can actually achieve, then is another lineage mark. And so we bring this vector into the IPS cells, we sort for those that have received the vector and then using these eye tracer IPSCs, we can grow super organoids. And now from the beginning, we have barcodes present that mark every cell lineage in the organoid. But in addition, we can induce scarring at a later time point and induce a second lineage mark. And by changing the time point of scarring, we can try to assess when cells commit to a certain cell fate. And so we can use this data to establish these lineage trees. And so how does this system work? Just for some QC, we have these organoids that are now green. So here these represent cortical regions that were micro-desected. We can detect barcodes in most of the cells, roughly 70%. We also see scarring. This is, I mean, the CRISPR system is not working 100% of the cases. So we see scarring not in all cells, but in a good fraction of the cells. And when we look at the barcode, the size of barcode families, we can see that there are quite some barcode families that where we can see tens of cells per family, we can also sometimes see hundreds of cells for a barcode family. So that we mainly detect the majority of barcode families we detect with one cell. This is just showing that our sampling is not complete at all. So we would need to sample much deeper in the organoids to get better there. But we think that this is already quite informative. If we look at the scarring, we can see both insertions and deletions, rarely both. We can see variable length of scars. And since cells can receive multiple vectors, we can also detect multiple barcodes per cell. And in that case, if a cell has multiple barcodes, we also detect multiple scars. And so now let's have a look at the data from organoids. So in this case, we generated a data set from six whole organoids that grew in two batches. And since this is a single RNA-seq based method, we get the single cell transcriptomes for every cell. And then we can, again, in this 2D U-map embedding, you can see how the different lineages or how the different cell states emerge. Here's a telencephalic lineage. There's a hindbrain lineage, a dencephalic and the midbrain lineage. Now, in addition to just using these transcriptomes, we can detect barcodes and scars in every cell. And so we can use this information to then establish these lineage trees for the organoids. So in this case, if we zoom in here, you can see on this first level, we have the barcodes. So this is a barcode family here that you can see up here. All these cells have the same barcode. So they arose from one original cell. And then in addition, we scarred at day 15, which you can see down here. We scarred at day 15. And we have different scar families within this barcode family that then share the same unique barcode scar combination. And so this is very rich data. And we started to explore that. And when we did that, we actually got quite surprised in the beginning because what we could see is that already when we look at the level of barcodes, these barcode families segregate out in these regional identities. For example, the red barcode family is mainly found in the terminal cephalone and this top-wise barcode family is mainly found in the rounded cephalone. Barcode families tend to segregate and accumulate in brain regional identities. And we wanted to look more into that. And what we did was we actually combined our eye tracer organoids with a spatial transcriptomic method. So in this case, we now don't dissociate tissues and measure single cell transcriptomes, but we take an organoid, a fresh frozen organoid, we slice it and then we put slices, in this case, three different slices from the same organoid, onto slides that contain spatially barcoded oligos from which you prime your reverse transcription. And what you at the end get is a CDNA library where every transcript has a spatial barcode. So we can map back where the transcript came from. And these spots where we have these spatially barcoded primers located are of the size of they are in diameter, 55 microns roughly in diameter, which shows that it's roughly one to 10 cells that you profile with that. So this is not a single cell level method, but it gives us the spatial resolution. And yeah, so this is the 10 X genomics visual method. And so we can actually deconvolute every individual spot that is in principle an average over one to 10 cells into composition of single cells based on the single cell data we have. And this is what we did here then in order to annotate the spots. So again, we see telencephalon emerge, we have dencephalon, mesencephalon and also rhombencephalon. Here's just expression of some genes shown on the spatial maps. And now we can also detect the barcodes and the scars using this method. So here are all the red spots where we detect the barcode. So this is quite a large number of the spots. And then you can see how the barcode families that here are some exemplary barcode families. And you can see also in this data now how barcode families segregate and accumulate spatially, which is very interesting. So this suggests that cells that are uniquely barcoded in the very cell stage in this embryo body, this three-dimensional early tissue, will give rise to more and more cells. But all the cells that a cell gives rise to stay in the proximity of that cell and are most likely to give rise to the same regional identity later on in the organoid. And we wanted to look at that actually directly. So in this case, a cancer gen, a postdoc in the lab did in total imaging of the organoids using inverted light sheet microscope. And she imaged the organoids really over many, many hours. In this case, for example, over 100 hours. And you can see here a section through this organoid that we image in three dimensions over time. And you can see how initially this is a round ball. And then you see more and more of these lumina that form and these regions that bloom out. And we can use this data because we have nuclei fluorescently enabled to actually track, directly track lineages. So in this case, this is the tracking of one nucleus and what nuclei it gives rise to. And this is the lineage tree that belongs to this initial nuclei. So here you can see where then all the cells, all the daughter cells of that initial cell are located. And we did this for a different starting nuclei. And what we could find that indeed the original location of a nucleus really matters. And the cells that belong to one lineage are very much clustering around the location of the initial nucleus. So the internuclear distance within the same lineage and between different lineages in the same lumen are very small. And then the internuclear distance between nuclei in different luminal regions are large. So this shows that in these organoids we have an initial proliferation and neuroactoderm formation step. And then you have aluminization and regionalization. And as a consequence related cells tend to contribute to the same part of the brain. And so this was really nice to us because this is consistent with in vivo data from model organisms such as zebrafish and mouse. So this shows that these organoids follow developmental mechanisms that are found in vivo. Finally, what I didn't yet touch upon is this, what I said initially this method is really good for is to scar at different time points and identify when cells get committed. And of course, because we found barcode families mainly look at it in a single region identity this analysis was a bit confounded but we found enough barcode families that would be spreading across different regional identities. And for these barcode families, we can do this analysis. So in this case, this shows one barcode family. It's a KNM graph, a false directed layout. And you can see within this barcode family the scar families that we detect. And for example, focusing on the red scar family you can see that that family very nicely just locates to one of the cell types. So these are progenitors and then the different cell types that emerge here to the left and to the right. And so it seems like at this day 15 when this organoid was scarred, this cell that gave rise to this scar family was already committed to that cell type. And so when we do this for now for different time points of scarring and we always assess whether scar families are distributed across different cell types or whether they are restricted to a given cell type or cell lineage. We can see that between scarring at day four or seven and day 15, this is when the most happens where scar families really start to restrict to a cell type. So it seems that cells start to commit to brain regional identities before day 15 which got us very interested in these early time points. And yeah, because we didn't necessarily expect that and we now started profiling also these early time points for example, day 15, where we can see nicely how the organoid is getting patterned actually. You see expression of all these different morphogens. And this is where we are going right now. And so this is the summary of that part where I showed you that we can use hyphalopathy and it's only seek to illuminate that multiple brain regions emerge in these organoids. We can use chimpanzee field organoid atlas to identify predictive human specific regulatory changes and we identified this delayed neuromateration. And then I showed you that we use novel or develop use novel technologies to track lineages directly in the organoids and study regionalization. And this revealed this clonality of brain regions similar to what has been observed in vivo in model organism. It's now fully switching gears and we go to actually living model organism away from the organoids but I think this is a very exciting system. So I try to fast tell you about this project that we already completed. So what I showed before is really work in progress but here these are these axolotls which are really regeneration champions. They can regenerate major parts of their body. And we worked in collaboration with Ali Tanaka who is a world leader in axolotl research to analyze the regeneration of axolotl limbs. And just to give you an impression. So the axolotls, these are salamanders that where you can amputate the limb here, the fall limb for example and then within roughly 30 days the limb regrowth which is of course really fascinating especially since this limb very much resembles from the architecture and human arm. And the most important part of this regeneration process is the establishment of this little tissue at the amputation side which is called blastema. And this is a tissue that forms through migration of cells into this area. And this blastema has been studied for a long time but what has not yet been studied was very much was these connective tissue cells. These are all the gray cells. And these are really the most abundant lineage in the blastema and they also express, they are known to express patterning factors that are required for proper morphogenesis of the limb that is regenerating. So we teamed up with the Tanaka Lab and they had generated this reporter line where all connective tissue cells can be labeled fluently. And using that line, we tracked the connective tissue cells through the limb regeneration. And so here's a cell atlas now using single cell transcriptomics on all cells from an upper arm. And you can see all kinds of different cell types including immune cells. But there's this big block in the middle the labeled cells labeled with the reporter which are the connective tissue cells. And if we zoom in, we can see heterogeneity in the connective tissue cells which includes tendon, tenor sites which link the muscles to the bones. Then we have periscal cells surrounding the bones. We have bone cells and then we have all kinds of fibroplastic connective tissue clusters. So our question was what happens to this diversity of connective tissue cells when you form this blastema? Is this maintained? Is this lost? What happens to all these cells? And so we then amputated the limb and then profiled single cells in a time course of limb regeneration and always sorting out these connective tissue derived cells. So we knew they did the cells are derived from connective tissue. And just looking at this data day by day, you can see that this initial heterogeneity which you can see in this heat map now plotting the same genes for these other cells from the other time points. We see that this heterogeneity is lost. We can very much see a loss of any of these cell state identities. And this is also shown here where initially we have this heterogeneity that then samples into a quite homogeneous state. And I skip this. So then we can use this data. Again, we can align cells on a pseudo time in this case using diffusion maps. And we can then show what is actually happening very early on you have an inflammation response. You lose the connective tissue identity. So cells kind of exit that connective tissue state. The extra serum matrix gets disassembled and then we enter a proliferation and cell division state. And at the end, it's not shown here, but at the end, the connective tissue identity gets reestablished. So this very much suggests that the cells that initially heterogeneous connective tissue cells funnel into a homogeneous progenitor like state that then gives rise to the regrowing limb. And so we were wondering whether these homogeneous cells in this plasthema tissue in any way resemble or are similar to cells that you would find in a developing limb. So in a limb butt that you find in an embryo. And so in this case, we then profiled the developing limb across these different stages of development. And then using signals RNA seek. And then we compared the states that we find in the developing limb to our plasthema cell states. And what we can see is that now here calculating the similarity to the limb butt stages, which is a very homogeneous state actually, we can see that initially the uninjured cells are very different. And then as you proceed through regeneration at day 11 post amputation, we have a state in the plasthema that very much resembles the embryonic limb butt state. So this was interesting. So this shows that the cells in the plasthema really resemble at day 11 a developmental state. And you can see that also in terms of expression. So here are all kinds of genes. You can see that these later plasthema stages resemble these developmental stages in the expression profile. Whereas these early stages are actually very unique to the regeneration and are not found in development. So this was very interesting. And we can also see that in terms of patterning in the developing limb, we see patterning, proximal distance, a distal and anterior posterior patterning ongoing. And we can see the same thing then in these later plasthema stages. And finally, this is the last point we wanted to see now, how does the arm grow out? And what are the cell states that are being occupied during this process? And so we profite now the late stages of regeneration from 18 to 38 day post amputation always focusing on the upper arm. And we can now see that this is again a diffusion map of this data. And we can see that we come from the plasthema progenitor state here and then we move along here and then divert into different branches here that reflect the cartilage, the bone and the nonskeletal connective tissue. And this very much suggests that these progenitors in the plasthema are actually multipotent. And we also confirmed this using a kind of rainbow axolotl line that cells can be derived from one connective tissue lineage. And then enter this progenitor state that is actually multipotent and can then give rise again newly to any of these connective tissue identities. And so with this I'm at the end of this short second part and I want to thank my group for all the fun that I have working with them. I mentioned all the people that did the work. So this last axolotl work was done by Tobias Gerber in my group. And I want to thank our collaborators and funding sources. Thank you. Thank you very much Barbara. It's 59. I went a bit longer than five minutes. I think you can go a bit into the lunch break no problem. Sorry. No, this was a great overview of the field and of your great progress. Should I unshare my screen or? You can leave it on in case there are questions about specific slides. Is there a question from within the network at the moment? I don't see a raised hand, but you're still welcome to do so. To raise your hand if you have a question for Barbara. There are two questions on Slido. I'll ask them in chronological order. The first one was in the first half of your talk. It is nice work. I'm curious about the two to four months differentiations, viability. How often do they fail? What are the usual technical problems that you face, if any? To differentiate from two to four months. Yeah. So, I mean, when actually when an organoid, if an organoid looks good at one month stage, usually it proceeds quite nicely. I mean, things can happen technical things like they are growing on these orbital shakers. If that one stops, then usually the organoid collapses and things like that. But usually one can assess quite early on whether organoids will stay nicely, will develop further in a nice way or not. I mean, there can be things like that the organoid, that the different cell types in the organoid get out of balance and you get the full organoid full of astrocytes or something. We see this very rarely. Usually, if the organoid is the whole batch looks good early on, then it also proceeds nicely. Of course, at some point there's a problem with the maturation and we can't, so far we haven't profiled, let's say one year old organoids. At some point there's a big necrotic core in the middle because the cells in the middle don't receive as much nutrients and oxygen. And this is of course where the whole field tries to engineer these organoids better, introduce a vascular term, for example. Also introduce, as I said in the beginning, cells that are missing such as, for example, immune cells. So kind of letting microglia infiltrate the organoids to then help also get rid of dead cells and so on. So I think there's definitely a good opportunity to get these systems mature better. But I think up to four months, this is relatively straightforward. And there's a general question about the second part. Do you have errors in measurement problems in your data? I think the answer is clearly yes, but maybe you can comment a bit more on that. I mean, of course, there are all kinds of problems with single cell data. I mean, we are usually capturing only a small fraction of the transcriptome. We also have, especially with these droplet-based methods, I didn't go into explaining the methods we use, but the axolotl project used a combination of chip-based kind of valve microfluidics-based data, the fluid ion system, and then also droplet microfluidic-based data. And with droplet microfluidics, you don't just capture single cells in the droplets, but also ambient RNA. And so every dissociation, you have a different pool of ambient RNA floating around that you will then encapsulate in the droplet and measure. So using these droplet methods, there is much more batch variation you see. So from experiment to experiment, there is some background RNA that will give you a technical variability. And yeah, so I mean, there are definitely problems with this type of data, but I think one can still extract a lot of important information from it. I don't know if that answered the question. No, I think this addresses this question. And I'll pick one more from Slido, which is that this was a jaw-dropping talk, really. Is there any research in inducing limb regeneration in mammals, or is it science fiction up to this point? Yeah, so of course, this is what everyone would like to know. Why can they regenerate the limb? And we cannot. And so, I mean, we also, again, in collaboration with Ellie, who is really from the biology side, the expert, we bring all the technologies. We have also now, for example, compared axolotls with Xenopus. This is of course not yet going to mammals, but Xenopus is interesting because when you amputate the limb, it will make a spike. It will grow a blastema and then some spike of cartilage, but it will never regrow the complete limb. So we thought this is a very nice intermediate system that can somewhat do the beginning of the regeneration, but then fails to proceed all the way through. And we can see that these connective tissue cells fail to actually go into this multi-potent progenitor state. And of course, now there's a, it's interesting to now look at, for example, mouse where you can, of course, injure the mouse skin, let's say, and then have also injure generally the limbs. And then there are cells that will migrate into that injured tissue. And then one can of course study how what is the profile of these cells and make a comparison. I mean, there are problems that are related to the different genomes. I mean, axolotls have, I think, 20 times, I don't know, I hope that is right, a much larger genome than humans. There are a lot of repetitive regions in the genome. And of course, it could even, the secret could also lie within some genes or regions in the genome that we don't have at all. So these comparisons will only give some information, but we can only compare genes if we have the ortho look in all the species, for example. I showed you that these late blastema stages correspond to the embryonic progenitor in the developing limb, but these early stages are quite different and quite unique to regeneration, to the regeneration process. So we think that a secret could lie within this early state and that needs to be populated. And so one thing that of course is also interesting to do is now knock out some of these genes in axolotl and see whether the axolotl can proceed through the regeneration or not anymore. So these are all kinds of things that is interesting to follow up on. Yes. Definitely very exciting. Usually the more complex an organism is, the less regenerative capacity it has. Of course, we can regenerate parts of the liver and so on, but that is the most, the largest thing we can, I mean, yeah, usually we can't regenerate whole organs. I think liver is the most, yeah, the best example there, but we don't really have other examples from humans. Thank you. I'll ask one question myself for the network, which is, Barbara, in your talk, in all of these very impressive examples, the big N is usually the number of cells that you are looking at. In other talks in our summer school, the big N is like all the patients that there are in a big bio bank or collection of electronic health records. Now, my question is, do you see that these two worlds will meet at some point in the near to mid future? Of course, we have heard about this lifetime initiative that, so the core of the question is, when will we have large collections of single cell data on many thousands of patients? Do you think this is happening very soon? So I think, I mean, especially with these in vitro systems, we are not so far, I would say. So because there are initiatives to generate IPS lines from many different individuals, such as the Hipski resource, yeah, and in the UK. So actually when we provide the seven lines, which is of course not large N, we use some of these Hipski lines. This of course all depends on having many cells profiled from many individuals. This depends on then also having methods that are less expensive, that can profile millions of cells in a quite cost efficient way. And then of course, it's also about the sequencing costs. But I think there are methods on the horizon and used by some labs. For example, the split pool barcoding based approaches that Jayshen Rur, for example, is really pioneering. There you can get two millions of cells with at least on the consumable cost for the experiment, you are much cheaper than with 10X genomics, let's say. But even with 10X genomics, you can actually barcode cells before putting them into droplets. And then you can get much more single cells out of one 10X run. And so I think this combination of using these Hipski lines, growing organoids or in vitro cells, and then doing these very high throughput single cell experiments will bring quite a lot of insight into now focusing on personalized health and so on. I think there's quite, it's exciting. Also, the organoids are currently not yet so high throughput. So one might have to employ robotics to grow them. What one can actually do, what we have started to do, one can grow them also musically. So you don't just grow an organoid from one line, but you actually mix all the lines and then grow a tissue that has many different genotypes. And of course, in this way, you also get too much higher throughput. So these are all kind of... ...exciting results to come. Yes, exactly. I think many exciting... That's where the field should push. Thank you so much, Barbara, for this talk. It was wonderful. We sent you a round of virtual applause. And thank you, it was great. And we now go into a lunch break and we restart at 1.30, it's a central European time, with Jenna Wien's talk. So enjoy the lunch break. Enjoy the rest of your summer school. Thanks.