henhaobio
Sign in

NatureDevelopmental BiologyPeer-reviewed journal article

A single-cell transcriptional timelapse of mouse embryonic development, from gastrula to pup

Chengxiang QiuiD, Beth MartiniD, Ian C. WelshiD, Riza M. DazaiD, Truc-Mai Le, Xingfan HuangiD, Eva K. NicholsiD, Megan L. TayloriD, Olivia FultoniD, Diana R. O’DayiD, Anne Roshella GomesiD, Saskia IlcisiniD, Sanjay R SrivatsaniD, Xinxian DengiD, Christine M. DistècheiD, William Stafford NobleiD, Nobuhiko HamazakiiD, Cecilia B. MoensiD, David KimelmaniD, Junyue CaoiD, Alexander F. SchieriD, Malte SpielmanniD, Stephen A. MurrayiD, Cole TrapnelliD, Jay ShendureiD

PDFDOIPreprintCodeData
In-depth full-text digest available · Jump to digest ↓

Abstract

The house mouse, Mus musculus, is an exceptional model system, combining genetic tractability with close homology to human biology. Gestation in mouse development lasts just under three weeks, a period during which its genome orchestrates the astonishing transformation of a single cell zygote into a free-living pup composed of >500 million cells. Towards a global framework for exploring mammalian development, we applied single cell combinatorial indexing (sci-*) to profile the transcriptional states of 12.4 million nuclei from 83 precisely staged embryos spanning late gastrulation (embryonic day 8 or E8) to birth (postnatal day 0 or P0), with 2-hr temporal resolution during somitogenesis, 6-hr resolution through to birth, and 20-min resolution during the immediate postpartum period. From these data (E8 to P0), we annotate dozens of trajectories and hundreds of cell types and perform deeper analyses of the unfolding of the posterior embryo during somitogenesis as well as the ontogenesis of the kidney, mesenchyme, retina, and early neurons. Finally, we leverage the depth and temporal resolution of these whole embryo snapshots, together with other published data, to construct and curate a rooted tree of cell type relationships that spans mouse development from zygote to pup. Throughout this tree, we systematically nominate sets of transcription factors (TFs) and other genes as candidate drivers of the in vivo differentiation of hundreds of mammalian cell types. Remarkably, the most dramatic shifts in transcriptional state are observed in a restricted set of cell types in the hours immediately following birth, and presumably underlie the massive changes in physiology that must accompany the successful transition of a placental mammal to extrauterine life.

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 1 (5 Apr 2023) · CC BY 4.0

12.4 million single-nucleus transcriptomes from 83 staged mouse embryos (E8 to P0) yield 190 cell types, a zygote-to-pup cell-type tree, and rapid postnatal transcriptional shifts.

Earlier whole-embryo single-cell atlases of the mouse covered gastrulation or E9.5 to E13.5 at roughly 24-hr intervals, and later fetal stages went unprofiled because embryos hold so many cells and bone is hard to pulverize. This atlas covers E8 to P0 at 2-hr, 6-hr and 20-min resolution with deeper profiling per nucleus, which supports a curated zygote-to-pup cell-type tree and candidate driver transcription factors (TFs).

Key results

  1. ShownA densely sampled whole-embryo atlas runs from late gastrulation to birth. Compared with the group's earlier E9.5 to E13.5 atlas, it has more nuclei, deeper profiling and finer time steps. It is annotated into 26 major clusters and 190 cell types.

    Main matrix: 11,441,407 nuclei from 74 embryos (15 sci-RNA-seq3 experiments, 21 NovaSeq runs); median 2,700 UMIs and 1,574 genes per nucleus. Versus the prior atlas: 2 million → 12.4 million nuclei; median 671 → 2,545 UMIs. Pseudo-bulk PC1 (77.3%) tracks developmental time. · Introduction; Single Nucleus Transcriptional Profiling of…; Preliminary Annotation of Major Cell Clusters…; Cell clustering and cell-type annotations

  2. ShownBirth brings abrupt transcriptional shifts in only some cell types (hepatocytes, adipocytes, lung and airway, some endothelial and blood types); neuronal types change little. A C-section series placed the hepatocyte, adipocyte and lung shifts within the first hour, with gluconeogenic and thermogenic programs.

    k-nearest-neighbor timepoint analysis of 71 cell types (E16 to P0) showed extreme P0 separation for these types but not for neurons. A single-litter birth series (962,697 nuclei from 9 pups plus quality-control nuclei; 6 C-section pups at 0 to 80 min) replicated the hepatocyte, adipocyte and lung & airway outliers. Hepatocytes induced Ppargc1a, Pck1, G6Pc and Got1; brown adipocytes induced Irf4 and Ppargc1a. · Rapid Shifts in Transcriptional State Occur in a…; Collecting mouse pups immediately after birth; Cell clustering and cell-type annotations; Figure legends

  3. ShownA rooted, directed cell-type graph from zygote (E0) to pup (P0) was built by scoring mutual-nearest-neighbor (MNN) edges within 14 subsystems. It joins four published early datasets to this atlas and was then curated by hand.

    283 annotated nodes. Of 1,155 candidate edges with normalized MNN score > 1, manual review kept 452 (developmental progression or spatial continuity), plus 55 inter-subsystem edges. Inputs: 110,000 cells (E0 to E8.5) and 11.4M cells (E8 to P0). The displayed graph has 262 nodes and 338 edges. · A Rooted Tree of Cell Type Relationships…; Figure legends

  4. SuggestedEach of 436 tree edges (excluding dataset equivalence and pre-gastrulation edges) was split into early, inter-node and late phases using MNN-defined cell groups. This nominates candidate TF and gene drivers, recovering known regulators and proposing new ones.

    Across 436 edges: median 28 (IQR 12-51) differentially expressed TFs and median 171 (IQR 76-389) differentially expressed genes (DEGs) per edge. Known regulators recovered: MITF (melanocytes); Ebf1 and Pax5 (B-cell progenitors). Sox17 was the only early-phase upregulated TF for anterior primitive streak to definitive endoderm and preceded the others in pseudotime. · Systematic Nomination of Transcription Factors…; Figure legends

  5. ShownIn the posterior embryo, gene programs for anterior-posterior (A-P) position and somite stage are shared, and change in the same direction, across germ layers (notochord, gut, neuromesodermal progenitors or NMPs). Shared early-stage genes include Myc targets, Lin28a and Hsp90 isoforms.

    A-P genes, notochord (n=591) vs gut (n=502): 198 overlap, 86% concordant (p<1e-28, χ2). Somite-stage genes, NMPs (n=257) vs gut: 82 overlap, 70 (85%) concordant (p<1e-15). 54/70 (77%, expected 25%) replicated in 12 independent embryos (8 to 21 somites; 104,671 nuclei). · The Posterior Embryo During Somitogenesis; Cell clustering and cell-type annotations; Figure legends

Methods, data and reproducibility

Whole embryos were staged by somite counts, limb-bud morphometry or digit morphology, or (from E17) by harvest time and gestational age, then profiled with optimized sci-RNA-seq3 and clustered and annotated in Scanpy. Mutual-nearest-neighbor analyses inferred progenitors, cell-type tree edges and phase-resolved TF and gene nominations.

  • Whole-embryo sci-RNA-seq3 timelapse of C57BL/6NJ embryos in 2-hr (somitogenesis) to 6-hr bins, E8 to P0; about 1% (0-12 somites) previously published · mouse (Mus musculus, C57BL/6NJ) · n = 11,441,407 nuclei from 74 embryos (75 selected of 523 staged) · GEO GSE186069; GSE228590
  • Birth series from one litter: 3 vaginally delivered pups and 6 C-section pups harvested at 0, 20, 40, 60 or 80 min · mouse · n = 962,697 nuclei: 9 pups plus residual nuclei of 3 earlier samples
  • Independent early vs late somite-stage validation set, sci-RNA-seq3 sequenced on Illumina NextSeq · mouse · n = 104,671 nuclei; 12 embryos (8 to 21 somites)
  • Four published pre-gastrulation and gastrulation scRNA-seq datasets (E0 to E8.5), including the Pijuan-Sala 10x Genomics gastrulation data (E6.5 to E8.5) · mouse · n = 110,000 cells (2,250 pre-gastrulation; 108,857 gastrulation)
  • Mosta spatial transcriptome sections of mouse embryos (E9.5 to E16.5), used to impute the spatial origin of mesodermal subtypes · mouse · n = 53 sections across 8 timepoints
  • Embryos digested for total genomic DNA quantification to estimate whole-embryo cell number from E8.5 to P0 · mouse · n = 12 embryos
Methods & tools
sci-RNA-seq3 (optimized protocol; 'Tiny Sci' adaptation for E8.0 to E9.75 embryos), Embryonic Mouse Ontogenetic Staging System (eMOSS) plus a modified Python stager, bcl2fastq, deML and trim_galore for demultiplexing and adaptor trimming, STAR aligner (mm10, GENCODE VM12), HTseq strand-specific UMI counting, Scrublet plus iterative subclustering for doublet removal, Scanpy (PCA, Leiden/Louvain clustering, UMAP), Seurat v3 FindMarkers/FindAllMarkers, Monocle 3 (doublet subclustering, pseudotime), Tangram mapping onto Mosta spatial transcriptomes, Mutual nearest neighbor (MNN) scoring for progenitor inference and tree edges, Cytoscape v3.9.1; Qiagen Puregene DNA quantification for cell counts
Software & environment
bcl2fastq v2.20; trim_galore v0.6.5; STAR v2.6.1d with mm10 and GENCODE VM12; HTseq on python v2.7.13; Scrublet v0.1; Scanpy v1.6.0; Monocle 3-alpha; Seurat v3; Cytoscape v3.9.1
Reproducibility gaps
A website for browsing the cell-type tree alongside the scRNA-seq data was still under development. The text does not say which data subsets are deposited under GSE186069 and which under GSE228590.
Limitations (6)
  • The authors call the cell-type annotations preliminary. They expect both the annotations and the tree structure to contain errors that will need correction. · stated by the authors
  • Despite the scale, cellular coverage stays modest: summing across 6 embryos per stage window, it falls from 0.5-fold at early somite stages to 0.002-fold just before birth, as the embryo grows from 210K to 670M cells. · stated by the authors
  • Early (0-12 somite) and later somite-stage embryos were processed in separate experiments, so stage differences in NMPs could partly reflect batch. An extra 12-embryo experiment was run to check this. · stated by the authors
  • Transcriptomes carry no explicit lineage information. Tree edges come from a heuristic plus manual curation; the blood and peripheral neuron subsystems needed manually added pseudo-ancestor edges; and some links, such as posterior intermediate mesoderm to collecting duct, need lineage tracing. · stated by the authors
  • How long after birth the vaginally delivered pups were harvested was not recorded precisely. Estimating it from C-section kinetics assumes similar dynamics for both delivery modes, and the E18.75 to P0 gap may be longer than 6 hours. · stated by the authors
  • Most timepoints rest on one embryo, and the C-section series has one pup per 20-min point from a single litter, so differences between individuals, sexes and litters cannot be separated from time. No TF nomination was tested by perturbation. · noted in this summary

Read this if you need a finely time-resolved single-nucleus reference for the whole mouse embryo from E8 to P0, a curated zygote-to-pup cell-type tree with TF nominations, or data on rapid birth-associated transcriptional change.

Citations
202
FWCI
26.58
Field percentile
top 1%

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISHe '20Qiu '21Cheng '24Menchero '25Shetty '25Boroviak '18
68% semantic matchDevelopmental Biology
Peng He, Brian A. Williams, Diane E. Trout+13 more·Nature·Jul 29, 2020· 177 citations