Day one of the Single Cell Biology 2018 conference at the Wellcome Genome Campus has just wrapped up, and with jet lag taking a toll on my sleeping schedule, I figured I could take a few moments to write up a summary of the conference so far.
Great set of diverse presentations, from expanding spatially-resolved transcriptomics techniques (MERFISH) for other purposes, to using pooled perturbation experiments to understand gene epistasis (Perturb-seq). Lots of impressive posters, but similar to some of the conversations going around Twitter, I wonder about the appropriateness of so many tSNE plots.
Today was a half day, with registration starting at lunch, so after a bit of food and mingling, we dove right into presentations.
Note: Tweeting was encouraged and none of the speakers said not to report any info about their talks, so I’m assuming posting this is okay, but if any of the speakers want this removed, just contact me.
Keynote: Xiaowei Zhuang
Xiaowei gave a fascinating talk highlighting some of the applications that they’ve been MERFISH for. MERFISH is an modified version of single-molecule fluorescence in situ hybridization that allows for multiplexed transcript counting of up to thousands of genes in the same cell. The benefit of this is pretty obvious: single-cell resolution expression profiling while retaining the spatial contexts of cells. They have also extended the protocol to be amenable for tissue staining while reducing background signal by embedding the tissue in a polymer matrix and clearing away the tissue, retaining the RNAs in the matrix.
She then talked about some more recent work that was particularly interesting. They’ve started using the technique along with pooled CRISPR library screens, allowing them to identify each cells genotype while preserving the cells in space. One could imagine using this to assess morphological differences following deletion of various genes. Their work was using pooled CRISPR screens to mutate a fluorescent protein called YFAST to improve its photobleach time. This study is quite cool on its own, but it was the first time I had thought about using pooled CRISPR screens to deeply interrogate a single gene. If there’s a way test a given phenotype of interest, you could mutate every amino acid of a protein determine its consequence on that phenotype.
Sarah filled in this time slot due to some scheduling conflict, but it was a nice combination of my interest in single-cell genomics and my labmate Pascale Charette’s (@pascalerc) interest in the maternal-fetal interface. I was furiously scribbling notes of things I thought she may find interesting.
The maternal-fetal interface isn’t my specialty, but it is quite interesting because…well, it’s the interface between fetal (which contain paternal antigens) and maternal cells. It’s required for the transport of nutrients between mom and baby, and because of the paternal antigens, it makes for an interesting immunological model (a strong immune response against the fetal cells could lead to miscarriage). The Teichmann lab was interesting at profiling all of the cell types in the decidua (maternal side) and placenta (fetal side), and were trying to assess the interaction between the various cell types. Single cell RNA-seq of these tissues allowed them to define all the cell types present and construct pseudotemporal trajectories representing the development of different trophoblast lineages. They found novel populations of NK and fibroblast subsets, each with unique spatial distribution through the tissue. They also introduce a new method/database called CellPhoneDB (domain exists, but seems locked to internal users only???) that makes use of receptor-ligand expression patterns across scRNA-seq data to infer cell type interactions. I’m still not super clear about the method, but it seems interesting.
Jonathan discussed their Perturb-seq31610-5) method that combines scRNA-seq with pooled CRISPR library screens and how it can be used to explore epistasis. By comparing the expression profiles of cells containing deletions of gene A, gene B, or both gene A and B (note: can be scaled up to a higher number of genes), they’re able to look for interesting gene relationships. Right now, they’re particularly interested in synergistic synthetical lethal combinations, which could be valuable for developing new combination therapies, or even tailored single therapies for cancer patients, for example, whose tumours may have mutations that provide the first “hit”. It would be interesting if they could find interesting interactions with the common “cancer genes” like p53 or Myc.
He mentioned one thing that made me appreciate the ability to assess the cells’ phenotypes with scRNA-Seq in these screens. There was one example of cell depletion/reduced growth following the over expression (CRISPRa) of a specific gene—cell growth is a common assay used in these screens, seeing which genes lead to over-under enrichment after a period of time. Using the scRNA-seq data, they were able to show that this wasn’t a direct effect on proliferation/death, but rather, it was driving differentiation (and in the case of another gene, transdifferentiation). This was only really detectable because they used the expression profile to conclude it. It made me reflect for a bit because of how often we use do simple experiments like growth curves following a drug treatment or something, and we just assume that our treatment is affecting proliferation/death. Gene expression profiles really can be an invaluable tool for figuring out what’s going on.
What a neat talk. I understood pretty much none of the technical components of it, but I still thought it was pretty intriguing. He described some work defining a new gene-regulatory motif that allows for Perfect Adaptation, which is when a system initially responds to a system, but then returns to basal levels, even when the original stimulus is sustained. The levels of the various components of the network can change, but the architecture of the network allows for the response variable to go back to a specific level. He showed some great simulations of this, and even some synthetic networks that they built in bacteria. What was cool was that they concluded that for this regulatory motif, stochastic noise was required to achieve this precise regulation in scenarios where similar deterministic regulation fails (and can produce oscillations).
This made me think abstractly about cell phenotypes (some latent response variable) and their relation to expression profiles. Because we never measure the environmental variables that cells are exposed to (eg. Ligand concentrations, temperature, pH, etc), it made me wonder if there is some perfect adaptation-like mechanism to keep phenotype relatively constant by changes in gene expression patterns. This could be problematic because in the field we effetely equate gene expression profiles to cells’ phenotypes. I guess it depends how you want to define phenotype: if a cell activates a stress response to prevent itself from dying following a stimulus, it may look quite different on the inside, but perhaps it is functionally identical to a cell that had not received the stimulus. Maybe its all just semantics—I dunno.
A couple 15 minute talks
The 15 minute talks (did they shorten them to 10 minutes? They seemed shorter than 15) were also quite cool. I won’t go into each one, but we got to see some interesting information about using single-cell dynamics to learn about gene families and why ohnologues may persist throughout evolution, another study using live-cell imaging to track hematopoietic stem cell fates, and an update from Fluidigm about their current directions (scRNA- and ATAC-seq from the same cells! No data though, but the paper has been accepted in Nature Medicine).
There are some really cool posters here. It’s crazy to think how fast this field is developing and how many datasets are being generating. There seemed to be a good mix of methods papers and…not methods papers? (What would you even call them?). There were a handful looking at the tumour microenvironment, some looking immune cell response to various things, and others.
Looking around from the middle of the room, I felt surrounded by tSNE plots. I’m not thinking of any single poster in particular here, but it made me question the appropriateness of so much tSNE. Don’t get me wrong, tSNE can be a fantastic way to visualize the data, colouring everything by cluster IDs determined by some other approach, but I feel like the default project pipeline that’s taking over is 1) scRNA-seq on some complex tissue, 2) cluster, 3) tSNE to see clusters, 4) validate cluster by IF/FISH/IHC. I mean, this is fine and dandy, but the data can be more rich and informative than this (trajectories between cell types, intracluster phenotype gradients, etc) and I’m not sure we’re asking the questions that follow “what cell types are in my tissue?”, even though we’re seeing projects involving development, cellular responses, cellular heterogeneity, etc (and if these are the questions, tSNE is probably not the best option because the inability to draw many conclusions from cells’ coordinates and the weird stuff it can do with continuous data structures). I also understand that posters are often the beginnings of larger projects, and the first questions of these projects tend to be “What cell types are here? Are their new ones we didn’t know about?”, so I’m genuinely not throwing shade at any specific paper/talk here.
So with that, I’m really excited to see how the rest of this conference goes!
PS. Canadian IPA is much better than the Brits’ ;)