henhaobio
Sign in

bioRxivDevelopmental BiologyPreprint

Decoding gene regulation in the mouse embryo using single-cell multi-omics

Ricard ArgelaguetiD, Tim LohoffiD, Jingyu LiiD, Asif NakhudaiD, Deborah Drage, Felix KruegeriD, Lars VelteniD, Stephen J. ClarkiD, Wolf ReikiD

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

Abstract

Following gastrulation, the three primary germ layers develop into the major organs in a process known as organogenesis. Single-cell RNA sequencing has enabled the profiling of the gene expression dynamics of these cell fate decisions, yet a comprehensive map of the interplay between transcription factors and cis-regulatory elements is lacking, as are the underlying gene regulatory networks. Here we generate a multi-omics atlas of mouse early organogenesis by simultaneously profiling gene expression and chromatin accessibility from tens of thousands of single cells. We develop a computational method to leverage the multimodal readouts to predict transcription factor binding events in cis-regulatory elements, which we then use to infer gene regulatory networks that underpin lineage commitment events. Finally, we show that these models can be used to generate in silico predictions of the effect of transcription factor perturbations. We validate this experimentally by showing that Brachyury is essential for the differentiation of neuromesodermal progenitors to somitic mesoderm fate by priming cis-regulatory elements. The data set can be interactively explored at https://www.bioinformatics.babraham.ac.uk/shiny/shiny_multiome_organogenesis/

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 2 (15 Nov 2022) · CC BY 4.0

A mouse embryo multiome atlas (E7.5 to E8.75) and an unsupervised transcription factor binding predictor; a Brachyury knockout supports its network predictions in neuromesodermal progenitors.

Transcription factor (TF) binding data are scarce for early embryos, and a motif in accessible chromatin does not guarantee binding. The authors' unsupervised score needs no ChIP-seq training data, unlike the supervised multi-omic predictors they describe, and it feeds gene regulatory network (GRN) inference and in silico TF perturbation.

Key results

  1. ShownThe atlas yields a catalogue of candidate cis-regulatory elements. Distal elements change in accessibility across cell types more than promoters do, as at the Gata6 locus, where the promoter is open in all cell types.

    61,781 cells passed QC (median 4,322 genes and 29,710 ATAC fragments per cell). 192,251 peaks were called, and 81% were differentially accessible in at least one cell type comparison. · Simultaneous profiling of RNA expression and…; A catalogue of cis-regulatory elements; Molecular characterisation of lineage-specific…

  2. SuggestedThe in silico ChIP-seq score combines motif score, accessibility and correlation with TF RNA expression. It agrees with public ChIP-seq for six TFs and agrees worse when RNA expression is left out. Agreement holds only for peaks positively correlated with TF expression.

    Benchmarked on Cdx2, Foxa2, Gata1, Gata4, Tal1 and Tbx5 ChIP-seq from in vitro models resembling the gastrulating embryo. Score thresholds of 0.10 to 0.30 were chosen by grid search. The text gives no summary AUC value. · Multi-modal prediction of transcription factor…; Multi-modal prediction of TF binding sites with…; ChIP-seq data processing

  3. SuggestedchromVAR-Multiome restricts chromVAR to predicted binding sites. Standard chromVAR scores for Foxb1 and Foxc2 (Foxc1 in the legend) are indistinguishable; chromVAR-Multiome fits TF expression better, with exceptions. It gives TF activity catalogues per cell type and suggests widespread TF pleiotropy.

    TF marker score is the fraction of significant pairwise cell type comparisons, with a minimum of 0.75 for the catalogue. Canonical markers are recovered (e.g. Tbx5 and Nkx2-5 in cardiomyocytes). Catalogues include neural crest and primordial germ cells. Foxb1 activity without expression in epiblast suggests priming. · Quantification of cell type-specific…; A catalogue of cell type-specific transcription…; TF marker scores

  4. SuggestedThe GRN of neuromesodermal progenitors (NMPs) places Cdx and Hox TFs at the centre, in an activating self-regulatory loop. Cdx2 is also predicted to repress TFs for somitic mesoderm and spinal cord. Cdx2 ChIP-seq shows binding across the Hoxb cluster.

    Besides Sox2 and Brachyury, 33 additional TFs were found, with 379 activatory and 48 repressive interactions. Validation is the Cdx2 ChIP-seq in epiblast stem cells treated with Wnt and Fgf. · Mapping the transcription factor regulatory…; Inference of the TF regulatory network…

  5. ShownCellOracle predicted that Brachyury knockout disrupts the NMP to somitic mesoderm transition. Brachyury KO embryos confirmed this: fewer somitic mesoderm cells, a lost velocity trajectory, and 344 peaks that stay closed, including elements near Tbx6, Mesp1 and Fgf4, supporting epigenetic priming.

    Multiome of 6,797 cells (3 WT embryos) vs 6,572 cells (7 KO embryos). 399 differentially accessible peaks in NMPs, 344 open in WT only, enriched for the T-box motif with higher in silico Brachyury scores than non-differential peaks. · Brachyury controls the differentiation of…; In silico TF perturbation with CellOracle; Brachyury gene targeting

Methods, data and reproducibility

Joint snRNA-seq and snATAC-seq (10x Multiome) of mouse embryos from E7.5 to E8.75 was combined with a correlation-based TF binding score. That score was used for TF activity estimation, metacell-based GRN inference and CellOracle simulation, then tested with a CRISPR/Cas9 Brachyury knockout.

  • 10x Multiome (snRNA-seq plus snATAC-seq) atlas of whole embryos, E7.5 to E8.75 · mouse · n = 61,781 cells passing QC · GSE205117
  • Brachyury KO and WT control embryos at E8.5, 10x Multiome · mouse · n = 6,797 cells from 3 WT embryos; 6,572 cells from 7 KO embryos · GSE205117 (raw sequencing data; the text does not separate the KO samples)
  • Public ChIP-seq for Cdx2, Foxa2, Gata1, Gata4, Tal1, Tbx5 (from in vitro models) used as ground truth · mouse (reads mapped to GRCm38)
Methods & tools
10x Genomics Multiome, CellRanger ARC, ArchR, MACS2, MOFA+, chromVAR, SEACells, CellOracle, scVelo, edgeR, fastMNN, JASPAR and CISBP motif databases
Software & environment
Versions are stated for many tools, e.g. CellRanger arc 2.0.0, ArchR v1.0.1, macs2 v2.2.7.1, scVelo v0.2.1, scran v1.20.1, but not for MOFA+, chromVAR, SEACells, CellOracle or edgeR. No container or requirements file is mentioned.
Reproducibility gaps
Raw data are in GEO (GSE205117) with a reviewer token. Processed objects are linked from the GitHub repository rather than a stated archive. Supplementary tables and figures are not in the text.
Limitations (5)
  • Few in vivo ChIP-seq datasets exist, so the in silico ChIP-seq benchmark relied on in vitro models that resemble gastrulating embryos. · stated by the authors
  • Too little ChIP-seq exists for chromatin repressors to benchmark repressive TF-peak links, so only activatory links were used downstream. · stated by the authors
  • Peaks are linked to genes by proximity (maximum 50 kb) because no 3D chromatin contact data are available. The authors advise against GRN inference at single-cell resolution because of ATAC sparsity. · stated by the authors
  • The in silico ChIP-seq validation covers six TFs and is described qualitatively in the main text, with no summary metric against competing TF-binding predictors. · noted in this summary
  • The Brachyury KO comes from one pooled experiment with 3 control and 7 KO embryos, and the in silico perturbation was checked for a single TF. · noted in this summary

Read this if you want a ChIP-free way to assign TF binding and activity from single-cell multiome data, or a mouse early organogenesis multiome atlas.

Citations
69

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
THISKim '24Argelaguet '19Yang '26Rendón '26Yan '22Deng '25
65% semantic matchDevelopmental Biology
Yang Joon Kim, Shruthi VijayKumar, Benjamin Giovanni Iovino+16 more·bioRxiv·Oct 18, 2024· 14 citations