In preprints: unfolding the spatial genomics frontier of mouse gastrulation

In an international collaboration (centered around Berthold Gottgens', John Marioni's, Jennifer Nichols' and Wolf Reik's labs in the UK, but also involving Shila Gazanfar's lab in Australia, and Long Cai's lab in the USA), Harland and colleagues performed a detailed spatial transcriptomic analysis of 48 h in the development of the mouse embryo, beginning at embryonic day (E) 6.5, corresponding to the start of gastrulation. The authors extended their previously published spatial atlas of early organogenesis (Lohoff et al., 2022) by supplementing their original E8.5 data with spatial transcriptomic data from earlier E6.5 and E7.5 embryos, and also integrating their recently published extended mouse embryo atlas representing the transcriptomes of all cells present in embryos from gastrulation to early organogenesis (Imaz-Rosshandler et al., 2024). This resulted in a spatiotemporal atlas of the cell states present from E6.5 to E8.5. Central to their construction of a spatial transcriptomic atlas, Harland and colleagues used the seqFISH multiplex mRNA in situ hybridization method to document the expression of 351 reference genes on 20 E6.5-E8.5 embryo tissue sections, to provide positional information on cell states during this 48 h window. They also re-annotated cell populations through the inclusion of cell subtypes. Spatial information, including coordinates for the anterior-posterior or dorsal-ventral location of single cells derived from their seqFISH experiments, were computationally imputed onto their embryo scRNAseq atlas. The authors then used these integrated data to explore gene expression dynamics along the anterior-posterior axis of the developing embryo, revealing that transcriptional changes along the anterior-posterior axis occurred in migratory mesoderm cells rather than in nascent mesoderm cells exiting the primitive streak.

Stem cell-derived models, such as embryoids, gastruloids and organoids represent the new vanguard for recapitulating the development of an embryo or organ at scale (Terhune et al., 2022). By projecting scRNAseq data generated from mouse 3D gastruloids (Rossi et al., 2021) onto their spatiotemporal mouse embryo atlas, Harland and colleagues sought to validate and benchmark this in vitro model, and glean ways to improve protocols for generating gastruloids more consistently. While the emergence of populations of cardiovascular, endocardial and hematoendothelial progenitors in the mouse gastruloids was confirmed, imputation of embryo spatial information onto the gastruloid scRNAseq data, revealed that the anterior-posterior development of the different germ layers was not as congruent in gastruloids as in embryos.

A second study from the labs of Naihe Jing and Sengbao Suo in China (Yang et al., 2024 preprint) analyzed embryos over a 24 h time window, also beginning at E6.5. These authors combined single-cell multi-omic (scRNAseq for gene expression and scATACseq for chromatin accessibility) data with spatial transcriptomic information and ChIPseq data, to link the transcriptomic and epigenetic landscapes of single cells in space and time.

Over the past decade, the Jing lab have published a series of landmark studies in which they performed spatial transcriptomic analyses of mouse gastrulation through the laser capture microdissection of groups of approximately 20 cells from tissue sections of embryos – a method they refer to as Geo-seq (geographical position sequencing). Although not at single-cell resolution, Geo-seq has provided an accurate 3D spatial atlas of gene expression. Data are typically represented as ‘corn plots’, their signature method for spatially visualizing gene expression data in cup-shaped gastrulating mouse embryos (Cui et al., 2018; Peng et al., 2016, 2019).

In their latest study, Yang and colleagues take it up a notch by integrating multi-omic data, from early through later stages of gastrulation (corresponding to E6.5-E7.5), with their existing 3D spatiotemporal atlas. Gene expression is just one piece of the puzzle underlying the dynamic nature of the cell state transitions occurring during development. Multiple inputs, ranging from extrinsic signals, cell-cell and cell-matrix interactions, along with cell-autonomous and non-cell autonomous responses, to epigenetic modifications, are necessary for the execution of proper embryonic development. Excitingly, Yang and colleagues have begun to integrate data for some of these parameters at the single-cell level in a tissue context. Using two new computational pipelines, ST-MAGIC and ST-MAGIC (+), the authors queried their multi-omic data to investigate cell lineage allocation and symmetry-breaking events during mouse gastrulation. With ST-MAGIC (+), they added another level of detail to their ST-MAGIC platform through the integration of previously published ChIPseq data for the WNT signaling effector β-catenin (Blassberg et al., 2022; Zhang et al., 2013) and NODAL signaling effectors Smad2/3 (Wang et al., 2017). Therefore, the interplay of gene expression, chromatin accessibility and developmental signaling could be studied, and indeed modeled, in a spatiotemporal context.

By incorporating existing computational methods [e.g. SCENIC+ (González-Blas et al., 2023)] into their platform, Yang and colleagues determined hierarchies of gene regulatory networks (GRNs) by inferring enhancer-driven transcription factor binding to target genes. They also implemented a new algorithm, named Bi-Orientation Cis Regulatory Element (CRE) Predictor (BIOCRE), to enhance the linkage of CREs with the expression of candidate genes. Their integrated data could then be queried for domains of gene expression, gene-peak linkage and cell type distribution at specific spatial locations within the embryo and across time-points. This, for example, allowed them to identify previously unreported subclusters of specific cell lineages in the axial midline. The authors' analysis of the regulatory regions associated with the well-known and evolutionarily conserved mesoderm marker Mesp1 revealed that peaks are shared by cells of the same lineage – hinting at the notion that cells at the start of a lineage tree are more plastic (i.e. have more peaks in common) – but as they begin to differentiate and commit to a specific lineage, their associated peaks become increasingly restricted.

Yang and colleagues' data visualized as corn plots is easily queried for cell position and identity, gene expression and chromatin accessibility. Once this atlas is publicly available, it will be a useful tool for other investigators. While the atlas represents the wild type, one can expect it to also be foundational for projecting data from experimental perturbations (e.g. gastrulation defective mouse mutants) to interrogate changes in cell states, cell composition and tissue-level organization.

The trilaminar cup-shaped gastrulating mouse embryo comprises thousands of cells and represents an excellent model to begin building spatiotemporal multi-omic atlases for gleaning insights of cell neighborhoods at single-cell resolution and tissue scale. A key challenge will be to extend the methods Yang and colleagues have developed to later developmental stages when tissue cartography exhibits greater complexity.

Another exciting possibility is the integration of additional parameters into the reference spatiotemporal atlases provided by Harland and by Yang and colleagues. Beyond compiling the transcriptional and epigenetic landscapes of cells in a tissue context, there will be a desire to annotate these atlases with single-cell proteomics data. As with the studies of Harland and of Yang and colleagues, different approaches are being used to build spatial proteomic atlases (Method of the Year, 2024). Furthermore, the biophysical properties of a cell influence its gene expression and are impacted by its neighborhood. While linking information on mechanical forces to multi-omics data remains a challenge, studies in Drosophila have demonstrated that the mechanical forces in cells can be computationally inferred from their tissue geometry (Noll et al., 2020). In a recent publication (Hallou et al., 2025), Hallou and colleagues report a computational pipeline that integrates mechanical force inference with single-cell spatial transcriptomic data of the same E8.5 mouse embryo (Lohoff et al., 2022), analyzed by Harland and colleagues. One can, therefore, anticipate further evolution of methods towards establishing a multiparametric atlas of mouse development, at gastrulation and beyond, to better understand the complex interactions between cells in tissues.