A Nature study combines Stereo-seq with single-nucleus RNA sequencing to examine how gene-expression programs unfold across space, time, organs and anatomical substructures during early human development.
Recommended role: flagship deep-dive article
Between four and eight weeks after conception, the human embryo is undergoing a profound transformation. The foundations of the heart, brain, liver, kidneys, eyes and other organs are taking shape at the same time, each following its own developmental program while remaining coordinated with the embryo as a whole.
Single-cell sequencing has greatly expanded our understanding of the cell types involved in this process. But identifying a cell type is only part of the picture. During organogenesis, where a cell is located, which structures surround it and when a regulatory program becomes active can be just as important as the genes it expresses.
A study published in Nature, “Spatiotemporal transcriptome atlas of human embryos after gastrulation,” addresses this spatial challenge by profiling 77 sagittal sections from 13 whole human embryos spanning Carnegie stages 12 to 23. Using Stereo-seq and complementary single-nucleus RNA sequencing, the researchers built an atlas covering 50 organs or anatomical regions and 198 molecularly distinct substructures.
The value of this atlas lies not only in its scale. By connecting developmental stage, anatomical position, cellular identity and gene regulation, it provides a framework for asking a more difficult question: How do human organs progressively acquire their structure and molecular identity?
At a glance
13 euploid human embryos spanning Carnegie stages 12-23 (approximately four to eight weeks after conception)
77 sagittal sections profiled with Stereo-seq, producing about 14.7 million bin50 spatial spots
50 organs or anatomical regions and 198 molecularly distinct substructures annotated
607,093 snRNA-seq cell profiles from five embryos at CS19, CS20 and CS23 used to refine cellular context
Meet the author. Dr. Li Yuejiao will explain how the atlas was built, validated and interpreted in an online webinar on August 27. Register now
Reintroducing space into the study of organogenesis
Previous single-cell studies have described extensive cellular heterogeneity during human embryonic and fetal development. Spatial transcriptomics has also been used to investigate gastrulation and individual developing organs. However, a gap remained between these two types of studies: researchers lacked spatial transcriptomic data covering whole-embryo sections across multiple post-gastrulation stages, when organ systems are becoming increasingly distinct but remain developmentally interconnected.
To address this gap, the team analyzed a cross-sectional series of 13 euploid human embryos, including seven male and six female embryos, across Carnegie stages 12-23. The study compares different embryos at defined stages; it does not follow the same embryo longitudinally through development.
The paper reports approval by the relevant institutional ethics committees and states that the samples were collected after legal pregnancy termination, with voluntary written informed consent obtained before the procedure. This provenance is important context for interpreting and communicating the work.
The main spatial clustering used bin50 (50 × 50 DNB bins, 25 μm), generating 14,744,703 spots. Fifty organs and/or anatomical regions were annotated using established anatomical and molecular markers. RNA in situ hybridization on adjacent sections supported selected marker-based spatial assignments.
This produced more than a collection of static embryo maps. Across developmental stages, the atlas captured a broader reorganization of the embryo: early sections showed greater representation of the heart, spinal cord and somites, while later sections increasingly reflected the expansion of the liver, bone, cartilage and skeletal muscle.
Article Figure 1 (source paper Fig. 1)
Figure 1. The atlas combines whole-embryo Stereo-seq with single-nucleus RNA sequencing to map tissues and selected organ substructures across Carnegie stages 12-23.
Source and provisional credit for original paper Fig. 1: Pan et al., Nature 654, 751-761 (2026). © The Author(s), under exclusive licence to Springer Nature Limited 2026. Permission for commercial reuse is not yet evidenced in this package.
From organ maps to cellular and regulatory context. The team generated single-nucleus RNA-sequencing data from five embryos at CS19, CS20 and CS23, comprising 607,093 nuclei and 68 reference-guided cell types. Mapping these data back to the spatial sections combined Stereo-seq's anatomical context with finer cellular annotation.
This integration helped resolve 198 molecularly distinct substructures across the 50 organs and/or anatomical regions, moving the atlas from whole embryo to organ and then to molecularly defined compartments.
The researchers also examined predicted regulons and inferred developmental trajectories to generate hypotheses about how tissue identity emerges. These analyses are computational: they are not direct lineage tracing and do not, by themselves, establish causal regulatory relationships.
Deep dive: locating the embryonic pacemaker
The heart provides one of the clearest examples of how spatial information can generate biological hypotheses. The researchers extracted cardiac data from 58 sections and annotated 26 cardiac substructures, including atrial and ventricular myocardium, endocardium, epicardium, compact and trabecular myocardium, vessels and the sinoatrial node.
The sinoatrial node, or SAN, is the specialized structure that initiates the heartbeat. Because it occupies a small and anatomically specific region, its molecular program is difficult to resolve without preserving spatial context.
At CS17, the atlas localized the SAN and identified a spatially enriched gene module containing SHOX2, a known regulator of pacemaker-cell differentiation, and VSNL1, which has previously been associated with SAN function. The analysis also identified KIAA1324L as a gene colocalized with the SAN and revealed regulatory activity involving RORA, a transcription factor better known for roles in circadian regulation and neural development.
RNA in situ hybridization supported the localization of KIAA1324L and RORA in the SAN region, where they colocalized with the pacemaker marker HCN4. The researchers then perturbed the zebrafish orthologues of SHOX2, VSNL1, KIAA1324L and RORA using CRISPR-Cas9. These perturbations reduced pacemaker-cell numbers and lowered heart rate in zebrafish embryos.
Together, these results support potential regulatory roles for KIAA1324L and RORA in pacemaker-cell development and function. The zebrafish experiments add functional evidence, although further work will be required to define their precise roles in the developing human heart.
Article Figure 2 (source paper Fig. 3)
Figure 2. Spatial mapping resolves embryonic cardiac substructures and sinoatrial-node programs, followed by functional testing of selected candidates in zebrafish.
Source and provisional credit for original paper Fig. 3: Pan et al., Nature 654, 751-761 (2026). © The Author(s), under exclusive licence to Springer Nature Limited 2026. Permission for commercial reuse is not yet evidenced in this package.
Deep dive: revisiting the timeline of early brain development
The developing brain presents a different spatial challenge. Its regions become progressively compartmentalized, while neural progenitors diversify into multiple neuronal and glial populations.
Using data from 65 sections, the researchers annotated major regions and substructures of the nervous system. The number of identifiable brain substructures increased from five at CS12-13 to twelve at CS23, while transcriptional differences among these regions became more pronounced after CS18.
One notable result concerned the relative timing of inhibitory and excitatory neuron development. Markers of inhibitory neurons were detectable in the subpallium as early as CS12-13. In contrast, excitatory-neuron specification was detected later, around CS19, in the pallial region. The early appearance of inhibitory-neuron markers was supported by RNA in situ hybridization and comparison with an external dataset.
This observation refines the developmental timeline reported in earlier studies and illustrates why spatial context matters. A marker detected across an entire embryo or brain sample may be difficult to interpret; locating that marker within a defined brain region makes the developmental event more specific.
The atlas also highlighted HMGA2, a regulator associated with neural progenitor maintenance. During earlier stages, HMGA2 was broadly expressed across neural regions but showed relatively low regulon activity. At later stages, its activity became more spatially restricted, particularly within the ventricular zone of the pallium.
Integration with single-nucleus data linked HMGA2 to radial glial cells, including a PAX6-positive, FOXG1-positive population in the pallial ventricular zone. An in silico HMGA2-knockout simulation predicted accelerated neuronal and glial differentiation at the expense of maintaining this radial glial population.
The study further reported a striking difference between species: HMGA2 expression showed rostral enrichment in the human pallial ventricular zone but a more caudal pattern in the corresponding mouse region. Additional analyses connected HMGA2 to gene sets involved in neural maturation, supporting the hypothesis that species-specific spatial regulation of HMGA2 may contribute to differences in how progenitor pools are maintained across the developing brain.
This does not mean that a single gene explains anatomical differences between human and mouse brains. Instead, it identifies a regulatory axis that can now be investigated more directly.
Article Figure 3 (source paper Fig. 4)
Figure 3. The nervous-system analysis links regionalization and neuronal timing with HMGA2 activity, inferred cell trajectories and cross-species spatial comparison.
Source and provisional credit for original paper Fig. 4: Pan et al., Nature 654, 751-761 (2026). © The Author(s), under exclusive licence to Springer Nature Limited 2026. Permission for commercial reuse is not yet evidenced in this package.
From disease-gene mapping to cross-species model assessment
Because the atlas places gene expression within defined organs and developmental stages, it can also be used to examine genes associated with developmental disorders.
Using the Developmental Disorders Genotype-to-Phenotype database, the researchers analyzed the spatial and temporal distribution of disease-associated genes. Many showed enrichment in organs consistent with their known phenotypes: genes associated with structural heart defects were enriched in the heart, genes linked to intellectual disability were prominent in the brain, and cataract-related genes were concentrated in ocular structures.
Some disease-associated genes were expressed across more than one organ, reflecting broader clinical phenotypes. Others showed differences between human and mouse development. ARG1, which is associated with arginase deficiency, showed liver-enriched expression in human embryos from CS14-15, while comparable liver enrichment appeared later in the mouse dataset. CCBE1, associated with lymphatic development and Hennekam syndrome, also showed differences in lung expression between the two species.
These observations do not invalidate mouse models. Instead, they provide a way to identify developmental stages or organs in which model organisms may not fully reproduce human expression patterns.
Adding an allelic dimension to the spatial atlas
Most genes are transcribed from both inherited copies, or alleles. However, the two alleles do not always contribute equally to gene expression. This allelic imbalance can be particularly important during development. Imprinted genes, for example, are expressed preferentially according to whether an allele was inherited from the mother or father, and disruption of imprinting can contribute to developmental and neurological disorders.
The researchers identified 452 genes with reproducible single-nucleotide polymorphism-level evidence of allelic imbalance in at least three of five analyzed embryos. Using haplotype phasing in three embryos, they further identified 104 genes showing gene-level allelic imbalance across organs, including 20 known imprinted genes.
Canonical imprinted genes displayed expected patterns. MAGEL2, IGF2 and MEST showed paternal dominance, while MEG3 showed maternal dominance. The spatial data additionally allowed the researchers to examine how these imbalances varied among organs and anatomical locations.
This analysis adds another layer to the atlas. It is not only a map of where a gene is expressed, but in selected cases, also a map of which allele contributes to that expression.
Allelic layer. 452 genes with reproducible SNP-level evidence; 104 genes with gene-level cross-organ imbalances after haplotype phasing; 20 known imprinted genes.
What can the atlas tell us - and where are its limits?
The resource can help researchers locate gene expression across organs and stages, compare anatomical substructures, prioritize candidate regulators, examine developmental-disease genes in spatial context and assess where model organisms reproduce - or diverge from - human developmental programs. The paper provides repository accessions, analysis-code links and an interactive atlas, although access conditions may vary among human-data repositories.
Its boundaries are equally important: capture efficiency may obscure rare populations; sagittal sections provide two-dimensional views of a three-dimensional process; and inferred trajectories or regulatory networks generate testable hypotheses rather than causal proof. Many candidates will require perturbation experiments, additional stages, three-dimensional reconstruction and other molecular modalities.
What, then, can this atlas answer now, and which questions still require new experiments? These are among the issues the author will take up in the webinar.
Interactive atlas: HESTA interactive atlas
Hear the study behind the atlas
How did the research team align data across embryos and developmental stages? How were 198 anatomical substructures defined within complex whole-embryo sections? Which regulatory candidates were selected for validation, and what should researchers consider when reusing the atlas?
On Thursday, August 27, 2026, Dr. Li Yuejiao, a co-first author of the study and Associate Researcher in Biotechnology at BGI Research, Shenzhen, will present the work in the online webinar “Mapping Human Organogenesis with Stereo-seq.”
10:00-11:00 AM SGT (UTC+8)
11:00 AM-12:00 PM JST / KST (UTC+9)
12:00-1:00 PM AEST (UTC+10)
Recording access will be provided after the event.
Register for the webinar
Webinar artwork and registration
Webinar artwork. Dr. Li Yuejiao will discuss Mapping Human Organogenesis with Stereo-seq on August 27, 2026, 10:00-11:00 AM SGT (UTC+8).
Reference
Pan, J., Li, Y., Lin, Z. et al. Spatiotemporal transcriptome atlas of human embryos after gastrulation. Nature 654, 751-761 (2026). doi:10.1038/s41586-026-10545-0