9 Single-cell transcriptomics reveals a novel developmental pathway
Wajid Jawaid1,2,3,4, Ximena Ibarra-Soria5, Vasilis Ladopoulos1,2, Fernando Calero-Nieto1,2, Carla Mulas1, Antonio Scialdone5,6, Jennifer Nichols1, John Marioni5,7,8, Berthold Gottgens1,2
1Cambridge Stem Cell Institute, University of Cambridge, Cambridge, United Kingdom. 2Dept. Haematology, University of Cambridge, Cambridge, United Kingdom. 3Addenbrooke's Hospital, Cambridge, United Kingdom. 4Evelina London Children's Hospital, London, United Kingdom. 5CRUK CI University of Cambridge, Cambridge, United Kingdom. 6Helmholtz Zentrum Munchen, Munchen, Germany. 7EMBL-European Bioinformatics Institute, Cambridge, United Kingdom. 8Wellcome Trust Sanger Institute, Cambridge, United Kingdom



Deeper understanding of the embryological origins of congenital anomalies is likely to provide insights into novel preventative and therapeutic strategies. Emerging technologies now offer the possibility of fine grained measurement of cellular activity at the level of the single cell rather than ensembles. We aimed to use the rapidly developing technology of single cell RNA sequencing (scRNAseq) combined with the development of new computational methods to study gastrulation and early organogenesis at a hitherto unprecedented granularity.



Single cell suspensions were generated from embryos harvested from time-mated dams, processed and sequencing libraries generated by one of either the plate based Smart-Seq2 or the microfluidics 10x Genomics protocol. Libraries were sequenced in multiplex, aligned and processed using standard methods. Pseudotime/pseudospace inference and artificial intelligence approaches were developed and applied within the context of this dataset for in-silico gene perturbation prediction.



Across experiments >80,000 cells were harvested and analysed at 9 embryological stages. Clusters representing pseudotime (embryonic blood development) and pseudospace (anterior to posterior mesodermal axis) were identified and were consistent with known biological knowledge and subsequent validation experiments, Figure 1. Sub-clustering identified a novel leukotriene pathway previously not associated with haematopoiesis that was validated using an embryonic stem cell differentiation model. By training a deep artificial neural network on a subset of genes assayed by qPCR and using in-silico perturbation we were able to show that it faithfully recapitulates a developmental progression.



scRNAseq can comprehensively characterise cell types within the early embryo and provide insights into ontological and spacial relations. These relationships can be used to train computer models to learn gene interaction networks and simulate perturbation experiments. Furthermore finer scale characterisation can reveal new pathways that can be targeted using small molecules and guide stem cell differentiation assays for regenerative medicine.