Single-cell reconstruction of developmental trajectories during zebrafish embryogenesis

INTRODUCTION During embryogenesis, pluripotent cells gradually become specialized and acquire distinct functions and morphologies. Because much of the specification process is controlled through changes in gene expression, the identification of the transcriptional trajectories underlying cell fate acquisition is paramount to understanding and manipulating development. RATIONALE Traditional approaches have studied specific fate decisions by analyzing the transcription of a few selected marker genes or by profiling isolated, predefined cell populations. The advent of large-scale single-cell RNA sequencing (scRNA-seq) provides the means to comprehensively define the gene expression states of all embryonic cells as they acquire their fates. This technology raises the possibility of identifying the molecular trajectories that describe cell fate specification by sampling densely during embryogenesis and connecting the transcriptomes of cells that have similar gene expression profiles. However, the numerous transcriptional states and branch points, as well as the asynchrony in developmental processes, pose major challenges to the computational reconstruction of developmental trajectories from scRNA-seq data. RESULTS We generated single-cell transcriptomes from 38,731 cells during early zebrafish embryogenesis at high temporal resolution, spanning 12 stages from the onset of zygotic transcription through early somitogenesis. We took two complementary approaches to identify the transcriptional trajectories in the data. First, we developed a simulated diffusion-based computational approach, URD, which identified the trajectories describing the specification of 25 cell types in the form of a branching tree. Second, we identified modules of coexpressed genes and connected them across developmental time. Combining the reconstructed developmental trajectories with differential gene expression analysis uncovered gene expression cascades leading to each cell type, including previously unidentified markers and candidate regulators. Combining these trajectories with Seurat, which infers the spatial positions of cells on the basis of their transcriptomes, connected the earlier spatial position of progenitors to the later fate of their descendants. Inspection of the developmental tree led to new insights about molecular specification in zebrafish. For example, the first branch point in the tree indicated that the first molecular specification event may not only separate the germ layers but also define the axial versus nonaxial mesendoderm. Additionally, some developmental branch points contained intermediate cells that expressed genes characteristic of multiple downstream cell fates. Gene expression analysis at one such branch point (the axial mesoderm) suggested that the intermediate cells switch their specification from one fate (notochord) to another (prechordal plate). Last, analysis of single-cell transcriptomes from a Nodal-signaling mutant revealed that even at the whole-transcriptome level, mutant cells were canalized into a subset of wild-type states and did not adopt any transcriptional states not observed in wild type, despite abnormal developmental signaling. CONCLUSION These findings reconstruct the gene expression trajectories during the embryogenesis of a vertebrate and highlight the concurrent canalization and plasticity of cell type specification. The scRNA-seq data and developmental tree provide a rich resource for future studies in zebrafish: The raw and processed data and the URD software are available for download, and the data can be browsed interactively online. Last, this approach provides a broadly applicable framework with which to reconstruct complex developmental trajectories from single-cell transcriptomes.