Another rainy day in the UK. Day two started off like the best of conferences: with an English breakfast! The themes for today’s seminars included development, computational tools, cancer, and neurobiology.
Spatially-resolved data is sexy and there are still a few optics-related things that need to improve. The throughput of scRNA-seq technologies is high enough that we can do entire atlases of specific model systems or thousands of cells from a variety of specific conditions. But do more cells = more information? I’m not sure.
Prisca described some interesting work on the dynamics associated with organoid formation. Working in the cancer & stem cell fields, organoids are of particular interest because they’re three-dimensional culture systems that attempt to recapitulate the some of the cellular complexity of tissues. Prisca’s work was centred on intestinal organoids, which I believe had been quite well-established. However, they were interested in looking into the dynamics that occur during organoid formation. It had previously been known that LGR5+ stem cells of the intestine could form 3d organoids comprising the different cell types of intestinal crypts at some efficiency, but Prisca not only showed that LGR5- cells could form organoids at the same efficiency, but also showed the expression patterns associated with symmetry breaking during the process. Using pseudotime analysis of bulk RNA-seq data, they showed that there is a transcriptional divergence of cells that go on to form “mature” organoids (ie. symmetry breaking with the relevant cell types), and those that remain as enterocytes, forming enterocysts. Interestingly, from this they were able to show that transient Yap signalling was required for symmetry breaking and formation of mature intestinal organoids. Using scRNA-seq they were able to construct pseudotemporal trajectories and show that at early time points, prior to symmetry breaking, they could see the consequences of Yap signalling that would eventually give rise to successfully-developed organoids (blocking transient Yap activation prevented proper symmetry breaking).
Bertie’s seminar highlighted some of the recent work to understand the process of murine gastrulation using scRNA-seq. Their dataset comprised ~100k cells from several time points (E6.5-8.5) around gastrulation. Together, these data represent a fairly comprehensive atlas of the gastrulating embryo. Manual annotation of clusters pulled out of the single-cell data matched the known cell types from embryology data. What’s nice about this dataset is that it can serve as a resource for those interested in embryo development. Although their group is following up in blood and primordial germ cell development, there still seems to be lots of space to explore developmental trajectories within the ensemble of their data.This could be a cool resource to play around with when it become available.
Couple 10 min talks
There were two talks this session: one (Daniel Wagner) really cool one using transposase-based DNA scarring to infer cell lineages, along with transcriptome data (TracerSeq). They did an impressive ~100k cells of the developing zebrafish embryo at several stages within the first 24hrs, tracking the cells’ expression and genetic lineages. The other talk explored an approach for spatially-resolved transcriptomics data (up to ~100 genes per cell) to understand neural crest development.
Admittedly there weren’t really many new computational tools in this sections
Scott discussed some remarkable advancements in imaging technologies that could dramatically improve signal-to-noise in FISH experiments. Admittedly, I don’t know much at all about optics, but he presented several methods for signal enhancement, and showed some incredible work decomposing fluorescence signal into its components (one example was separating GFP and FITC signals from the same image, which have a high amount of overlap in their spectra—I was surprised at how accurate this was!).
This was one of the talks that I was most excited for—particularly because Dana’s group has done some work that complements my own (epithelial-mesenchymal transition). Her talk deviated away form this a little bit though, and she described some recent work on building immune cell atlases for breast and lung tumours. I actually won’t talk too much about the findings of their studies because I wanted to emphasize a feature of their data/talk that stood out in my mind. Dana showed that the immune cell types that infiltrate a breast tumour are actually the same cell types that are circulating in blood, but the tumour-associated immune cells have a larger amount of phenotypic variance (based on their gene expression profiles). This is absolutely fascinating because it’s a feature that could not be resolved with bulk analysis and its biological role isn’t particularly clear. These are the questions that I find fascinating in the fields—ones that are not just “what cell types are here”, but actually interrogate some underlying cell biology. What does it mean if an immune cell occupies more phenotypic space? Is it more plastic in tumours, capable of different responses? I really don’t know, and Dana didn’t elaborate either, but regardless this is fascinating because it’s new. She also mentioned in passing that continuous cell phenotypes explain more variance than discrete clusters, which was particularly interesting because I think biology is much more…blobby(??) than we tend to assume. It will be interesting to see people explore their datasets as the field moves forward, looking for phenotypic gradients that exist within their discrete clusters.
More 10min talks
The two shorter talks of this session were about 1) decomposing fluorescent signal from spatially-resolved FISH data to identify spatial domains of cell types/expression patterns (Yuan et al.), and 2) deriving gene networks from pluripotent stem cells (McEwen et al.).
I had been familiar with the work from Itay and his group—in fact, they’ve made me nervous about exploring single-cell transcriptomics of any patient tumour model, because I’m sure they’re sitting on the data and working on publishing it!
His talk was about their recent paper on head and neck cancers31270-9), describing their findings about an epithelial-mesenchymal gene signature being associated with intratumoural heterogeneity, and this heterogeneity being associated with clinical outcomes (EMT = more metastasis). An interesting finding from this paper that had been seen in previous papers from melanoma and (I think) glioblastoma is that there is little variation in the non-cancer cells between patients, but the cancer cells vary quite a bit. This could be related to the mutational burden of the cancer cells, but not much exploration has been done into this as they’ve focused on intratumoural heterogeneity.
One thing that was particularly interesting is that they talk bulk RNA-seq analysis from The Cancer Genome Atlas and wanted to correlate their EMT signature with clinical features, but found that their EMT signature was founded with stromal component of the tumours (ie. some tumours had a more mesenchymal gene expression signature simply because they had more fibroblasts). They did some interesting work decomposing the bulk RNA-seq expression values (TCGA), to remove the effect of stromal and immune cell components, allowing them to look at a “pure” carcinoma cell expression profile from each patient sample and correlate that with clinical outcome. Turns out that an EMT signature is associated with more metastasis, which is consistent with most literature, despite some newer studies that suggest the EMT is not involved in metastasis.
10 minute talks
I was only able to stay for one of the 10min talks, but Klaas Mulder described their method (RAID) for quantifying (phospho)proteins and mRNA from the same single cells. It seemed very similar to the CITE-seq method, with a couple modifications, and they performed the quantification of 70 proteins in parallel! I thought this was very powerful because it can allow us to link signalling cascades to their transcriptional outputs. He showed pseudotemporal ordering of protein concentrations, but I’m wondering if you could explore spikes in phospho-proteins, shortly followed by transcriptional changes. Could be interesting to explore the expression programs associated with these various signalling pathways.
I actually had to duck out for the rest of the talks (one more talk on TARGET-seq and the Neurobiology sessions so unfortunately I can’t report on them today!
Anyway, that’s it for today. Tomorrow, we wrap up with sessions on stem cells and immunology. I’ll try to finish up a summary of that over the next few days!