Notes taken during the Single Cell Biology 2018 conference in Cambridge (6 - 8 March 2018). Notes were typed during the talks, have not edited and will contain mistakes. May be useful for others. Some session I missed hence no notes. Notes taken with bookdown. @ZakiWillBe.
Notes by others ;
In situ imaging of the organizations of transcriptome and genome in single cells
Xiaowei Zhuang, Harvard University/HHMI, USA
Imaging the transcriptome at the single-cell level. Single-cell transcriptome imaging method, multiplexed error-robust fluorescent in situ hybridization (MERFISH). Uses combinatorial labeling and sequential imaging to massively multiplex single-molecule FISH measurements.
Application of MERFISH
Decoding Cell Circuits at the Maternal-Fetal Interface
Sara Teichmean, Sanger
Begin by mentioning the technology behind scRNAseq. Calculate the accuracy of the spikeins. Different methods (Smart-seq etc), are accurate. Discussed saturation limit, the read depth plays a role in the detection limit. Sequencing benifits up to 1 million reads. The human cell atlas, the idea is simillar to google map vs street view.
Maternal immune response at the maternal-fetal interface at the decidua (layer only exists during pregnency, decidua is the maternal side). In order to vascularise the maternal tissue, provide oxygen and nutirets. Question : What cell types and states exists at the maternal/fetal interface.
Dissect the decidua & placenta interface. 45k cells of the maternal and fetal side. A progenitor population of the fetal trophoblast. The cells branch to the maternal tissue. The fetal cells do not express normal HLA, but finding a way to hide the HLA. They express the HLA-G. In the maternal sites, we see 3 novel dNK cell types. THe non-immune compartment (stromal cells) are important for interaction with the immune system. The fibroblast population seems to suggest 5 sub-population. The fibroblast subpopulation have differnt implication for fetal growth. Found some cell surface markers unique to the sub-population. Question : What are interactions between the new NK and stromal cell states. Ie - what type of dNK cells interacts with which fibroblast. CellPhoneDB http://www.cellphonedb.org/ a database of cell-cell communication network.
Finding and interpreting genetic interactions using Perturb-seq single cell RNA-seq CRISPR screens
Mapping the genetic landscape of a cell
Jonathan Weissman, UCSF/HHMI, USA
To be able to understand the function of a cell and find out how cells transition between states. We need a tool to allow us to (in high throughput) manipulate the cells in a controlled environment. Using dead CRISPR/Cas9 to turn on or off genes. Now we shift gear to synthetic lethality. Search for pairs of genes where loss of a tumour suppressor gene is conditionally synthetic lethal. The ability to look at huge number of combinations to find patterns of genetic interaction.
The basic behind genetic interaction : Look at the phenotype consequence of altering gene interaction (delete / KO). No interaction, buffering / synthetic lethal. How to study the relationship - Look at the interaction of gene to each of the CRIPR KO. Tmem261. So far it is about matching groups of genes into interaction. Now move into looking at synthetic lethargically action. Example FDPS & HUS1 gene interaction. We can do genome-scale library perturbation.
Perturb-seq http://www.cell.com/cell/abstract/S0092-8674(16)31610-5. Examples from synthetic lethal. CEBP synthethic lethal. CEBP goes to CMP lineage and CBL moves to eryrhtoid lineage. Example of buffering interaction. DUSP0 and ETS2. Where to go from preturb-seq : Use adaptive learning to predict gene interaction.
Exploring the dynamic benefits of intracellular noise
Mustafa Khammash ETH Zurich, Switzerland
The role of noise. And the interaction of noise and dynamics. Introduce theoretical motif. Perfect Adaptation - Responds to a stimulus but once situmus is removed, the cells return to basal level. Example : Calcium homeostatis. Importnant to maintain calcuim level. Data from breast feeding cow. Integral feedback from engineering. How can we achive perfect adaptation in a cell? A cell is very noisy. A probablistic approach to model noise. Problem of moment closure? Main challange in solving the moment dynamics. Adding second order (covariance). Can we build an integral feedback controller.
Single cell transcriptional dynamics underlying gene family expansion
How do gene families evolve
Jonathan Chubb MRC LMCB, UK
Example, humans has 14 histones which encodes for identical protein. Why the reason for duplication? Actin gene family. 17 gene family encodes for the same protein. scRNA suggest there was no compensation. Use live imaging to check the level of actin. Some gene s are associated with bursting. The transcription dynamics are differnt. What drives these differnt types of actin behaviours? Nuclear factors in gene expression. Swtiched promoter of differnt genes. “Bag-of-Patterns” approach to classify gene bursting. At shorter timescale the genes are transcriped differently.
Tracking hematopoietic stem cell kinship during in vitro clonal expansion provides insight at the most detailed level of cell fate and differentiation
Marijn van Loenhout, Hubrecht Institute, The Netherlands
HSC (CD49F+). Long term repopulating HSC. Low purity / heterogeniety of HSC. Fate decisions. Analogy of black box. Observing cell lineage is required. Take a platform (PReMiSe robotic platform) which can take +200 cells and can track each cell through its lineage. The captured sisters and cousin cells can be
Microfluidics for live cell imaging and cellular context
Mark D. Lynch, Fluidigm Corporation
Self-organization and symmetry breaking during intestinal organoids
Prisca Liberali FMI, Switzerland
Talk about emerging behavious in cell. How a population of cell interact, organise and create a tissue. Only few cells differentiate (in the self-organised cells) differentiate. Use intestine organoids model as models for self-organisation [https://www.nature.com/articles/nature07935]. One single cell (stem cell) can develop into organoid. The question : How does a single cell know where and when to differentiate? Studying one organoid may not be sufficient, there is a need to look at thousands of organioids. Developed a long term imaging framework for intestinal organoids. Indicates 3% of cells are able to generate organoids. Found two distict sub-population (based on imaging). ONe that forms ENterocyst and the other population did not form entrocyst. Used a custom build microscope to image the organoid formation. Tje enterocyst begin to spin. Looked at Lgr5+ and Lgr5-. Both of them developed organioid at the same frequency (3%) able to genrate organoid. Trajectory of organoids development found two branches. One branch to enerocyst. The Lrg5- cells acquire back Lrg5 expression after 5 days. Both organoid begins with Lrg5 then transitions into a Lrg5 negative population. Look at patterns of gene expression. Interested in finding genes which begins to turns on. Look at the transcription factor binding motif of the expressed genes. Found some motif that interacts with Yap1 motif. Small cells have Yap & big (aggregated) cells dont expresse Yap. Take home message, transiet yap1 activation is required for organoid formation. The activation of Yap1 is required only in the first two days but the deactivation of Yap1 is required to generate organoids. Deacivation is rquired to maintain homeostatists. The question : How is symmetry broken. What is expressed when Yap is variable? Conduct scRNA-seq on day 3 and day 5. Found Yap target genes. Look at the expression difference between Yap high vs Yap low population in the Day 5. Found Notch ligands (DLL1 and DLL4). Multiplexing and clarity imaging. DLL1 expression is high in the initial stage of the. Yap variability induces notch DLL1. The development of organoid does not follow developmental but regenerative process. Follows a Lrg5 intermediate. Yap1 variability is necessary for Notch/DLL1.
Charting the diversification of mammalian cells at whole organism scale
Bertie Göttgens, University of Cambridge, UK
A project to map mouse development using single cell. Process of grastulation. From a single cell, the zygote have to generate diverse types of cells to create a human embryo. You tube movie shoing the grastulation https://www.youtube.com/watch?v=9XV9c-97Wc4. Sampled from E6.5 to E8.5. Total purified single cells from more than 350 Embryods. ~96k cells that passed initial QC. Median number of 15k UMI & 3.4k genes generated. Looking at the temporal dynamics, shows the proportion of cells depending on the embryo days. There are celltypes that begin to emrges at particular time. The number of epiblast cells does not go down. Differentiation landscape of ony the intra-embryonic cells. Use force-directed graph to show the differentiation landscape. Can we create a reference atlas of the mouse? Combine data from different study. Tried to integrate data from these papers ;
The data mathces well on the current landscape of 96k cells. Overlayed the previous data (from the 3 studies above) on the current 96k cell reference plot. The current data can be used as a reference for wild-type vs mutant study. Can use this data to reconstuct the cellular roadmaps. Used graph abstraction to reconstruct the roadmaps of mammalian development https://www.biorxiv.org/content/early/2017/10/25/208819. Branches in the diagram captures well the aspect of ‘real time’ development. Look at the hemato-endothelial trajectory. Ask for clusters of co-regulatory genes. Creating a grastulation cell atlas that can be used as a definitive resource.
Systematic mapping of cell state trajectories, cell lineage, and perturbations in the zebrafish embryo using single cell transcriptomics
Daniel Wagner, Harvard Medical School, USA
Interested in the generation of adult organism from a single cell. Transition from a single egg to an animal. Studying the cells from each embryo development stage. Used the inDrops bethods for single cell RNA-seqencing. Sampling the first 24 hrs of zebrafish development. Profiled ~63k cells. Colelcted 7 different timpoints begining 4, 6, 8, 10 ,14, 18, 24hpf. Interested in how the cells develop/evolve over time. Settled on nearest-neighbour graph approaches (sc-kN). A graph is a collection of nodes and edges. Stiching together graphs which was build at one timepoint. The earliest time-point (4hpf) sits in the middle of the graph and the others sits at the outside of the graph. Build a web-portal to naviage the tree of the zebrafish development. SPRING web portal. Build a tree graph by collapsing cells from a cluster. How does a cell progress through the landscape of three? Developed TracerSeq to track lineage barcoding casset. The process of lineage tracing. Ability to relate cells to lineage information or to the graph information. TracerSeq can resolve non-tree like state relationship. Looking at loop-structure on the diffusian map.
Identification of a neural crest stem cell niche by Spatial Genomic analysis
Antti Lignell, University of Helsinki, Finland
Developed method to multiplex gene-expression preserving spatial context. Hybridization chain reaction. The methods are published https://www.nature.com/articles/s41467-017-01561-w. Wanted to study neural crest development. Visualied the gene expression and the position of the cells spatially. Found certain population of cells acts as intermediarry between stem cell population. Take home message and lession learned ;
Multimodal and multiplex molecular imaging tools for single cell analysis
Scott Fraser, University of Southern California, USA
We are very good at single cell omics quantification. But we are not very good at spatial context. Hybridization Chain Reaction (HCR). The idea is to do imaging at single cell level. Talk about some options to address the problems. Combined two-photom and light sheet microscopy. This gives a chance to look at live imaging at differnt layers. Fourier transform to remove signal noise. HySP segmentation. MUSE (Multimodal Universal Signal Enhancement) HCR. STain more quiclky and more accuraly then HCR. Hyperspectral?
Reference ;
* https://www.ncbi.nlm.nih.gov/pubmed/28672397
* https://www.nature.com/articles/nmeth.4134
* http://bioimaging.usc.edu/TICpubs.html
Cell states in the tumor ecosystem and the transitions between them
Dana Pe’er, Columbia University, USA
In-drop characterization of tumour immune cells. Collecting immune cells. Saw batch effects in tumour of different patitents. Build entropy graph, found low entrophy. Normalisation of cluster-dependent. Biscut normalisation http://proceedings.mlr.press/v48/prabhakaran16.html. Breast cancer immune system. Cluster are driven by co-variance rather than variance. Bhattacharya Distance (overlap between two distance). Using new techonolgy 10x and a new patient. How well can the known immune category cluster in the new patient. Found that even with new techonology (differnt platform) and sequenced to 3x depth managed to cluster well with known category. How much of the data can be described by gradient. Most of the variation can be explained by the gradient. How can robust cluster exists when most of the variation are driven by co-variance? The clusters are driven by metabolic changes. Look at TCR clonotype using the new 10x sequencing. Andrew Cornish (did the analysis)
https://www.biorxiv.org/content/early/2017/11/25/221994
Looking at latent metastatsisi-initiating cells. Understanding the latent matestatis initiating cells. Late metastais that can stays months and years after tumour. Example : patient can relapse after two years. Mouse system is build that can study the latent metastasis https://www.ncbi.nlm.nih.gov/pubmed/27015306. Profield the lung from normal and cancer patients. Created human lung adenocarcinoma atlas. Calculate entorpy matrix to measure cluster stability. Focus in sections of the non-immune cancer cells. Created phenograph for classification. Check the probability of how each cell reacing the other cell state. The method used random walk. The cell falls into multipotent state. The intermediate cell types are more of a mix. Diversity and Sox9 lineages.
Decomposing spatially dependent and cell type specific contributions to cellular heterogeneity
Guocheng Yuan, Harvard University, USA
Spatial information is lost in the scRNA-seq. Gave a nice lego analogy, we know the bulding blocks but we dont know the structure. How do we put all the blocks together. Moving beyond scRNA-seq. We want to try to move beyond scRNA-seq and incoprorate spatial information. seqFISH - in situ single cell. Spatial patterns of gene expression. http://www.cell.com/neuron/abstract/S0896-6273(16)30702-4. HMRF : a computational method to detect spatial domains. Use neighbouring cell information and gene expression pattern. Applied this method to the cortex data. Spatial organiztion is better represented by domain then cell type annotation. Overlay the gene expression into the spatial information computed by the HRMF. Inference of spatial information to public scRNA-seq data. The model HRMF can be useful to infer spatial domain from transcriptomic data. Validated the information by seqFISH as a guide to map scRNAseq data to spatial location. Qian Zhu develop the method. Available in bioRxiv https://www.biorxiv.org/content/early/2018/03/02/275156.
Concerted computational approaches to interpret transcriptional heterogeneity in pluripotent stem cells
Kirsten McEwen, Imperial College London, UK
Computational approaches beyond clustering. Focus on the additional analysi on Gene Regulatory Network (GRN). Many biological process are controlled by network. Interpreting gene regulatory networks. scRNA is quite noisy. But the biological noise aids the indentication of gene interactions. Often there are non linear relationship between interacting transcript. Move away beyond correlation using information theory https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5624513/. Build a network using top-scoring threshold approaches. Eg top 100 edges. Developed an emirical Bayes approach to perform hypothesis testing. Control for FDR where there are many edges and find the edges with statsitical significant. Compared to bulk RNA-seq the scRNA-seq showed better resolution. Focusing on pluripotency network. Use a measure called assortativity to increase the accuracy of network measure. Assortativity test basically is to test if the functional group in one category (in this case pluripotency) interact with each other more than chance. Can incorporate prior knowladge (chipSeq, bulk RNAseq). Example shown were information with chip data prior increases to network accuracy. The use of chip can also indicate directonallity. Package https://www.biorxiv.org/content/early/2018/02/13/264853. Another work is to look at the computational approaches for single cell bursting. Developing a method to infer gene bursting. And the second is to look at molecular mechanims of heterogeneity
Tracking the fate of single mutant stem cells in oesophageal pre-cancer
Philip Jones, Wellcome Sanger Institute, UK
Talk about the previous work of seqeuncing eye lids https://www.ncbi.nlm.nih.gov/pubmed/25999502. Infering the size of mutant clone. Now can extend the work to use engineering mouse mutations. Use mouse with single cell reporters to track the fate of the cells as the progress over time. Used human oesophageal epithelium as model. The human oesophageal epithelium has basicaly only one tisse (keratinocyte). Keratinocyte life cycle.
Get normal tissue from organ transplant donors. Cut of part of the oseophagus and do targeted exome sequence with deep sequencing. Mutation burden of 25 years old is 1-4 compared to 75 years old (apologies cant remember). Most frequent mutation is NOTCH1. How do you know if the mutation is driver? Use a method to estimate which genes are in positive selection https://www.ncbi.nlm.nih.gov/pubmed/29056346. The genes (Notch1, TP53) are under selective process. LOH of Notch1. Compared the mutation between TP53 in esophageal SCC vs normal esophageal. The mutation are very simillar. The mutations in notch are constrained to 8-12 domain. These are the domains which are conserved in mouse. Notch mutation is an aging associated process.
Radiation inuce a change of differentiation. If given radiation to the p53 mutatnt cells. The colony of mutatnt p53 will increase alot. The good news is that it is reversible. If anti-oxidant is taken before treatment of radiation, this reduce the p53 mutant percentage by 8 fold.
Deciphering intra-tumoral heterogeneity by single cell RNA-seq
Itay Tirosh, Weizmann Institute of Science, Israel
Tumour heterogeniety within a tumour. Try to profile tumour by scRNA-seq. A lot of the work focuesd on Glioma and Head and Neck (HaN). Get the surgical samples of primary tumour and metastatic tumours. Sequenced ~10k cells and ~5.5k passed QC. The cells were classed as cancer or non-cancer. The non-cancer cells cluster by cel type (ie, fibroblast clusters with fibroblast). But in cancer cells, the cells cluster by patients, indicating these patients have individual regulation (tumour hetreogenity). Looked at the similarity between primaray tumours and the lymph node metastatsis. Overall there was no big difference btwn primary and metastatsis. Suggesting that once the cells have metastasised the cells revert back to the primary tumour. How can the intermediate cell stage (or transition between primary and metastasis) be studied? To test these possibility went back to the data of matched primary and metastatics tumour. Found difference in Cycling, Stress responds [FOS, FUN, IER3] (generic and found in most cancers). Found changes specific to individual tumours, EMT-like process and EPidermal difference.
Is the EMT prgram associated with metastatsis?
Can correlate EMT expression vs meastasis. Map back TCGA data to the single cell data. Found that out of the 4 TCGA HaN classification (Atypical, Mesenchymal, Basal, Classical), 3 subtype (Basal, Atypical, Classical) can be found in the cancer scRNA-seq data. The Mesenchymal group mapped back to the fibroblasts population. What dominates the variablity between the bulk sample is partial EMT. Partial EMT is associated with metastasis. Similarity btwn primary tumour and lymph node metastasis.
IDH-mutat glioma follows tree like pattern. https://www.nature.com/articles/nature20123. Synergy btween mutation H3-K27M and developmental contex. Found oligodendrocyte prgenitor cells. Wanted to find the Mechanism of K27M mutations. Perdiatric K27M driven by OPC-like glioioma cells. Potential to inhibit BMI-1.
Intracellular (phospho-) protein and mRNA quantification from single human epidermal stem cells
Klaas Mulder, Radboud University, The Netherlands
One message : A way to measure intracellular and RNA-level on individual cells. How does a cell respond to a signal? Cells have different receptors on the outside and these signals are relayed into the inide of cells. If we measure the co-variation of signaling pathway activity and transcriptional output in single cell, we might able to infer the regulation of the cell. Use a model of human epidermis. Do mRNA and protein detection concurrently using the method developed called Single cell RNA and Immuno Detection (RAID). First stain with antibody. Doing scRNA-seq after immunostaining. Use reversible fixation to retreive the endogenous RNA. As proof of princial correlated the expression of unfixed single cells with the RAID single cells. Number of genes detected in the RAID protocol is slightly less (5k) genes compared to unfixed protocl (i think it was 6k). As the RAID will require good antibody, spend a lot of time to have a panel of 70 antibody panels which can be trusted. The confidence of single-cell prtein detection. Correlation btween two barcode. Can we use the antibody to learn about biology. Use the epidermis model system where cells are at differnt stage of differentation.
TARGET-seq: a novel method for single-cell mutational analysis and parallel RNA-sequencing
Alba Rodriguez-Meira, University of Oxford, UK
A method to integrate mutation and transcription. Question of interest hematopoitic systems. Percentage of gene detected are low in the dropout rates. Sort the cells by immunophenotype. Use Poly A combined with specific gene targeted primer. Compared to Smart-seq detection, TP53 (5%) in Smart-Seq to 100% in TargetSeq. As an example used the Jurkat cell line where known mutations. TP53, Notch,Pten, RUnx1. From 0% detection in SmartSEq to 98% detection in tartetSeq. Heterogneity of HSPC comparment. JAK2 hetrozygous and JAK2 homozygous mutation. Can find the resolution of complex hierarchy. Using high sensitiviy mutation detection.
Molecular organization of the mammalian nervous system at the single-cell level
Sten Linnarsson, Karolinska Institute, Sweden
Engaged in understanding brain arcitechture and brain cells. Profiled ~500k cells. Approach the cells using supervised approach. Then used graph clustering. Functional split between neurons vs non-neruons. The 2nd stage of the tree is split between CNS neuron and PNS neuron. Important point to note is that just because the cells are at the same anatomical origin, they do not cluster together. Rather clustering is based on functional genes. Focuses on astrocytes. Finding astrocyte diversity. Certain astrocytes are anatomically located. Example Telenchepalon. Validated the data with antibody and found a clear speration between the two sections. Switch to spatial distribution of cell types. The data is available from Allen Mouse Brain Atlas. Correlate the expression from the mouse atlas with the scRNA-seq data. osmFISH for spatial mapping of cell types. Build the flow cell and machine to automate omFISH. Stain the cels with dapi and totalRNA. A way to align scRNA-seq data with spatial tissue organisation. Develop a method to find the regions/layers from omFISH data. Taking from static picture to dynamic picture. Want to infer developmental trajectories. The idea is that use the idea of RNA Metabolism. There will be difference between unspliced and spliced RNA. unspliced mRNA will be seen first then the spliing occurs. Therefore by looking at the splice and unsliced RNA can predict the velocity. Validated the alrorithm of velocity with circadian oscillation data. Validated on another dataset https://www.ncbi.nlm.nih.gov/pubmed/29230054. Applied it to the hippocampus development, can predict the developmental origin. ALso did the velocity in human embryo.
https://www.biorxiv.org/content/early/2017/10/19/206052
http://velocyto.org
http://loompy.org
Gene delivery across the blood-brain-barrier, whole-body tissue clearing, and optogenetics to understand and influence physiology and behavior Viviana Gradinaru, Caltech, USA
Theme of the lab : Understanding how the circuits of the brain relates to brain behaviour
What circuits act as bypass for sleep? For example when a deadline is due, we can re-wire the brain to not sleep. Monitor Dorsal Raphe Dopamine Neurons (DRNda) neronal activity during natural behaviour. When there is an encounter with another female the DRNda would spike. http://www.cell.com/neuron/fulltext/S0896-6273(17)30458-0. On going work : At the single cell level are there single cells that respond to specific stimuli. Calcium imaging of deep brain to see which cells are stimulated upon exposure. Worked on method to gain optical access (ie to be able to view under microscope). Created a method to remove the bone structure. Tisue purifiing, How can we stain the whole brain? One agent is to use the blood. But there is a blood brain barrier. Engineered a method of labeling with vascular using a virus http://www.caltech.edu/news/novel-viral-vectors-deliver-useful-cargo-neurons-throughout-brain-and-body-78785. Used a method that carries an inducer that can. https://bmcbiol.biomedcentral.com/articles/10.1186/s12915-017-0421-3. Single cell transcript resoultion, noninvaisve behaviour monitoring.
Single-cell transcriptome profiling of Drosophila optic lobe neurons
Chaimaa Fadil, New York University Abu Dhabi, UAE
Investigating the neural diversity of Drosophila optic lobe. Isolate the entire optic lobe on dropseq in 3 runs. Resulted in 57k cells. Clustering found 52 cluster. The first branch of the tree resulted in Glia vs Neurons. Find the top10 markers in each cluster. Present approach to find markers genes of each cluster. Took transcriptome of bulk RNA seq and identify which of the scRNA-seq cluster.
Another aproach is to use a Gal4-UAS cluster system??. Applied machine learning to find gene network. Weighted gene co-expression network (WGCNA), Each cluster display a unique TF fingerprint and these are not shared. https://www.biorxiv.org/content/early/2018/01/04/243113
Spatiotemporal trajectories in the developing human cortex
Aparna Bhaduri UCSF, USA
Organizing principles of gene regulatory programs in somatic evolution
Itai Yanai, NYU Langone, USA
Talked about the book Yanai & Lercher http://www.hup.harvard.edu/catalog.php?isbn=9780674425026. Perspective of evoulution to give insight into everday biology. Talk about 4 differnt things
CelSeq uess early barcoding and linear amplification. Introduced CelSeq2 which has a sensitivity of 21%. Introduced scDual-seq where the method can gain the transcriptome of pathogen https://www.ncbi.nlm.nih.gov/pubmed/?term=scDual-seq. Note pathogen does not have polyA, therefore CelSeq2 will not detect pathogen sequence. An idea to do a mouse atlas is to use DNA-microarray to do spatial transcriptomics https://www.ncbi.nlm.nih.gov/pubmed/27365449. Each probe on the microarray capture a sections of the tissue. BSeq-SC use scRNA-seq to deconvolve bulk heterogeneous tissue data. Can we use scRNA-seq to deconvoluve the bulk? Applied this idea to pancreas https://www.ncbi.nlm.nih.gov/pubmed/27667365. Now try to move on to spatial transcriptomics. Seperating the library preparation to do bulk and scRNA-seq at the same time. Can take each spot of the microarray. Applied ‘smoothing’ to the scRNA-seq data. The logic is that if two cell are close by in the high dimention space, indicates these cells are simillar. Applied kNN-smoothing in a step-wise manner. Also in bioarchive.
https://www.biorxiv.org/content/early/2018/03/05/254375
https://www.biorxiv.org/content/early/2018/01/24/217737
Moving to the biology. The cell lineage of c.elegans was done by Sulston. A name of the hall in Sanger is Sulston. How is cell fate established? Example in drosophila, where cell fate is established by position. The fate in c.elegans must be hard wired. The development of c.elegans was studied previously https://www.ncbi.nlm.nih.gov/m/pubmed/25487147/. Isolating cells by ‘shotgun transcriptomics’. Dissociate embryo mannualy, each embryo is done individually (i think total about 8 embryo?). Conduct supervised clustering on the 15 cell stage, supervised clustering is based on staining data. Because there is a lot of staining of c.elegans have been done, so its possible to annotate each cell to thier stage. We can look at gene expression through the lineage. Made a simplifying assuption is that the genes are either on or off at particular stage. Each lineage has different gene that is used to create a position.
Clustering across cell-types in non tumour while cancer clustered by individual patient (Refered to Itay’s talk). We cant study tumour over time in humans, therefore turn to zebrafish. Inject oncogene into zebrafish (Collaboration with Richard White). Look at the PCA of cancer cells, and look at the loading of the PCA what is pushing it. Re-analysed https://www.ncbi.nlm.nih.gov/pubmed/27124452 data. The triangular shaped PCA may give an indication of something. https://www.ncbi.nlm.nih.gov/pubmed/22539553. Refer to the ant paper where vertices may reflect distint transcrtional programs. There are 4 programs (stress,….). Looked at the copy number from the RNA-seq data and found 4 clones. Now do a time-course experiment on zebrafish. Took biopsy at 4 time pooints. Cancer cell seems to have 4 transcriptional program.
Natural selection in spermatogenesis. Example of spermagonia. Preparing scRNA-seq data from studying seprmatogenesis. Can identify stage specific. Hww many genes are expressed at each stage. 87% of human genes are expressed in testes. 13% are not expressed. Genes that are not epxressed has more mutation in the 13%. By transcribing the genes it reduces the DNA-damage. Removing DNA damage by transcription. Bidirectional transcription. Try to correlate expression with mutation. If the gene is epxressed too much the mutation rate is higher. If a gene is off it would evolve faster. The gene transcribed in the gonads because to redce DNA-damage.
Characterising cell state transitions in mammalian epidermis
Non-cell autonomous effects of Wnt signalling
Fiona Watt, King’s College London, UK
Looking at heterogenitey in mouse epidermis. Schematic picture layer of human skin. Epidermis is the most outer layer. Krt14 expressed in the eipdermis. One of the pathway invovled in epdiermis is B-catenin. Previously looked at the effects of expressing B-catenin https://www.ncbi.nlm.nih.gov/pubmed/15992546. The question now is how do neighboring cell respond to cells where B-catenin is activated https://www.ncbi.nlm.nih.gov/m/pubmed/29334988/. Looking at WT cells in the absence of Tamoxifen, Look at the difference between cell state. Selected candidate genes to validate. When B-catenin is activated - reduction in response heterogeneity and increase in protein synthesis and regulation.
Using deep learning approach to get more out of scRNA-seq data. Integrate diverse dataset and reduce batch-batch and lab-lab variation. Use concorruent training of two neural network. 1700k cells for training and 500 cells for validation. PCA - data cluster by the site. If uses the PC removal plus t-SNE, the data looks better. The GANt-SNE uses information from deep-nerual network to integrate data from the differnt site. Latent space interpolation. Find group of genes from bulk sequencing data. Can we use the anlaysis to infer relationship between genes. The GAN approach shows a unversal represenation of epidermal differntiation. Able to idenitify cell state-determening gene & gene regulatory networks. https://www.biorxiv.org/content/early/2018/02/08/262501.
Exploring the mechanisms of haematopoietic lineage progression at the single-cell level
Ana Cvejic, University of Cambridge and Wellcome Sanger Institute, UK
Introduction of HSC, where the HSC can differentiate to all blood lineages. Disruption of the balance can lead to blood disorder or in extreme situation lead to blood marrow failure. Various models of Haematopoietic tree. To define a particular cell type normally is using cell surface markers (CD family). Use zebrafish as a model syste. Zebrafish have one haematopoetic organs, the kidney. Combine a large number of transgenic line. 8 differnt transgenic line, from each line isolate the kidney. Captured 1.4k cells and orderd by monocole. https://www.nature.com/articles/s41467-017-02305-6. Validated the computational trajectory by looking into cell morphology. Conducted a transplantation expriment. Only the stem cell line remains in the kidney. Only the stem cell were able to mainatin in the kidney, as other cell types will be in circulation. The Runx1 have the highest number of engrafted. Zoomed in at the branching point. Assume the cells at the branch are the ground truth. Train deep network to class each cell in the transition. Transcriptinally the cells are similar but there are subset of genes which makes the cell decide in the diffent fate choice. Data deposited at BASiCZ (Blood Atlas of Single Cells in zebrafish).
Restriction of lineage potential within the human haematopoietic stem cell pool
Serena Belluschi Stem Cell Institute, UK
Overview of the haematopoetic tree. The restriction of lineage potential happens somewhere in the haematopoietic pool. Look into the human HSC pool. Look at CD90 and CD49f. Only 1:10 cells can re-populate. Assese the differentiation potential of 49f+ cells. Isolated ~5.5k cells with index sort. Index sorted of differnt markers and check the time for lineage division and colony forming capacity. Pre-existing heterogentiy in the 49f+ compartment. The My/Nk & My/Ery markers are polarised to in the PCA space. Used a gating stratery to islate CD34lo/CLEC9Ahi (Subset 1) or CD34high/CLEC9Alow (Subset 2). Look at the lineage of the two subset. GSEA on lineage purifiying model. Did a limiting diultion assay to test the frequency of repopulating HSC in subset 1 and subset 2. Subset 2 is Ery NULL and My/Ly commited. Subset1 have long term repopulating capacity. Concludes that the first lineage restriction already ocuurs in the earliest compartment.
Multi-parameter analyses of single primitive human cord blood cells reveals a complex topology of early hematopoietic differentiation events
Connie Eaves, BC Cancer Agency, Canada
Explaning waddington landscape in the context of stem cell. Developed the spleen colony assay 1961. Can identify cells which makes erythrocytes, T-cell. HSC not always differentiate along a linear path. Putting one cell into the tail vain of the mouse https://www.ncbi.nlm.nih.gov/pubmed/18371352. Alpha, Beta (more multilineage), Gamma (Become more myloid). The heterogenity of mouse HSC, the alpha and beta both repopulate. In terms of human, human 34+ regeneates lymphoid and GM. Barcoded CD34+ cells with a viral barcoded cells. An approach to look at the entire CD34+ CB cells. Did cytof of 40 proteins in 3k single human 49f+ CB cells. Did not see any specific sub-population. Design an experiment using in-vivo culture to look at 49f+ cells over 8 weeks. Only the CD33mid population gave repopulating activity in the mice. The idea of CD34+ CB is there could be multiple path to
Deciphering the regulatory control of γδT cell development by single cell transcriptome analysis
Dominic Grün, Max Planck Institute of Immunobiology and Epigenetics, Germany
Simultaneous lineage tracing and cell type identification using CRISPR/Cas9 induced genetic scars
Jan Philipp Junker, Max Delbrück Center, Germany