The specimen was a well-preserved male embryo at late Carnegie stage 12, corresponding to post-conception week 5. It had 28 pairs of somites and came from a single voluntarily consenting donor undergoing elective termination of pregnancy. The investigators prepared 24 sagittal sections from six sets distributed across the embryo, assigning adjacent slices to haematoxylin-and-eosin staining, single-stranded DNA fluorescence imaging, Stereo-seq transcriptomics and matrix-assisted laser desorption/ionization mass-spectrometry imaging, or MALDI-MSI, for metabolomics.

Copy-number screening detected no whole-chromosome aneuploidy or large-segment abnormalities, although the assay was not designed to exclude balanced rearrangements, point mutations or low-level mosaicism. After quality control, the transcriptomic analysis retained 39,847 genes across 153,634 fixed spatial units, each 15 micrometres across and containing a 30-by-30 array of DNA nanoballs. The researchers treated these as near-cellular tissue units rather than individual cells and estimated a roughly 5% rate of mixed or “doublet-like” profiles, concentrated mainly at stromal–epithelial interfaces.

Unsupervised clustering produced 17 major populations whose positions corresponded to recognizable anatomy: hepatocytes below the heart, skeletal-muscle progenitors along dorsal somites and limb buds, erythrocytes in cardiac, hepatic and mesenchymal regions, endothelial cells along developing vascular routes, and neural progenitors around the ventricles. Fluorescence in situ hybridization and immunofluorescence supported representative assignments involving cardiac troponin T, alpha-fetoprotein, skeletal-muscle markers, SOX2 and SHH. Reference mapping to an independent human organogenesis single-cell atlas showed broad agreement, while a separate segmentation method produced concordant major domains.

The metabolomic layer retained 2,687 mass-to-charge features, including 1,395 putatively annotated metabolite features. To align the 20-micrometre metabolomic pixels with the finer transcriptomic grid, the team developed a Joint Spatial Registration Strategy using anatomical landmarks, rigid-body optimization, affine transformation and area-overlap-weighted interpolation. Independent metabolomic clustering recognized cardiac, neural, hepatic and mesenchymal regions, but their boundaries were less discrete than those inferred from RNA, indicating biochemical gradients rather than sharp cell compartments.

A weighted nearest-neighbour analysis combining both data types did not improve anatomical separation and sometimes blurred or merged clear populations. The researchers therefore used transcriptome-defined clusters for tissue annotation and treated metabolomics as complementary evidence about regional biochemical environments. One of the atlas’s most striking findings concerns the hepatic primordium, where the analysis separated five hepatoblast-related subpopulations.

Hepatocyte-like cells expressed metabolic, synthetic and plasma-protein genes including APOA2, APOB, AFP, AHSG and ALB, consistent with metabolic preparedness and hepatocyte-lineage bias in the developmental context. Centrally located progenitors coexpressed EPCAM, CD24, SOX9, HNF1B and ONECUT1/2, a profile consistent with hepatobiliary hybrid progenitors. Computational communication analysis inferred NOTCH-related signals involving DLK1 and NOTCH2, while liver mesenchymal cells expressed HGF and hepatocyte-like cells enriched the MET receptor.

These observations support spatially organized early liver specialization, but the study did not functionally test whether the inferred signals cause lineage selection. Co-registered chemistry added another dimension to the liver. Transcriptomic and metabolomic signals for glycerophospholipid, sphingolipid and purine or pyrimidine metabolism occupied overlapping territories.

Long-chain polyunsaturated fatty-acid programmes were accompanied by regional enrichment of arachidonic acid and docosahexaenoic acid, while bile-acid biosynthesis genes and glutathione metabolism were detectable. The researchers cautioned that enrichment might reflect lipid transport, storage, membrane composition or inter-regional exchange rather than local synthesis. SCENIC analysis also identified enrichment of the CEBPA and NR1H4 regulons in hepatoblasts, linking the spatial observations to inferred transcriptional regulatory programmes.

The neural atlas captured a closed neural tube with a layered arrangement of neural progenitors, neurons and a floor plate at the ventral midline. Neural progenitors expressed SOX2, PAX6 and cell-cycle genes, while neurons expressed STMN2 and DCX and enriched gene-ontology terms related to axon development and projection guidance. The floor plate expressed high levels of SHH and FOXA2, and inferred SHH–PTCH1 communication connected it to surrounding neural progenitors, including cells carrying the ventral-patterning marker OLIG2.

Floor-plate cells also showed regional palmitic-acid enrichment, while inferred N-cadherin interactions linked the floor plate with neural progenitors and the otic placode. Peripheral Schwann-cell precursors occupied locations along developing nerves, expressed SOX10, PLP1 and MPZ, and were linked to neurons through inferred NRG1–ERBB2/3 signalling. The otic placode was clearly positioned beside the hindbrain and carried markers including OC90, PAX8, GATA3 and EYA1.

Skeletal-muscle analysis resolved four states arranged from peripheral dermomyotome progenitors through migratory and differentiating populations to embryonic muscle fibres. Dermomyotome cells expressed PAX3, HMGA2 and MEOX1; migratory progenitors occupied surrounding somite regions and extended toward the limb buds; differentiating cells expressed MYF5, MYOG and MYF6; and nascent fibres contained MYH3, ACTA1 and TTN. Trajectory and RNA-velocity analyses ordered these transcriptional states but did not directly observe cells moving over time.

Inferred communication included IGF2–IGF1R and FGF–FGFR loops in differentiating muscle, while SIX1 and PAX3 programmes were associated with migratory progenitors. The heart was resolved into 13 cell-related clusters. Left- and right-ventricular populations shared cardiomyocyte markers but differed in IRX4, RYR2, CSRP3 and MYL2 expression, with the left-ventricular state appearing later in an inferred transcriptional ordering.

The full outflow tract was captured from the distal arterial pole to the ventricular junction and showed an outer contractile layer, atrioventricular-canal cardiomyocytes, endothelial cells, cardiac jelly and mesenchymal or erythroid populations. Cardiac neural crest cells were detected at the caudal outflow tract, where they enriched TGF-β and BMP signalling associated with cushion development and outflow-tract remodelling. Two sinoatrial-node-related populations expressed TBX18 and SHOX2, while a transitional population combined SHOX2 with atrial genes, placing conduction-associated states near the developing atrium before complete looping and septation.

Glycolytic and tricarboxylic-acid-cycle metabolites aligned with cardiac genes, and fatty-acid transport or oxidation signals were already detectable. The atlas also resolved macrophages, endothelial cells, epidermis, fibroblasts, renal progenitors, craniofacial mesenchyme and two classes of mesenchymal progenitor. Erythroid domains expressed GATA1, TAL1, KLF1 and NFE2 together with haemoglobin genes, while mass-spectrometry imaging detected regional enrichment of haem or haemin.

Renal progenitors showed a strong HNF1B regulon and target genes including LHX1, OSR2 and JAG1, consistent with early nephron patterning. These findings constitute a stage-resolved anatomical and molecular reference rather than a general developmental rule. The experiment used one embryo, so it cannot establish biological reproducibility, inter-individual variability or developmental changes over time.

Metabolomics captured only part of the metabolome and measured relative, steady-state ion intensities rather than metabolic flux, while pseudotime, RNA-velocity and cell-communication results were computational inferences rather than direct observations of temporal change or functional signalling. Additional embryos, later Carnegie stages and functional experiments would be required to determine how broadly the observed programmes persist and whether they actively shape organ formation.

Subject of Research: Spatial transcriptomic and metabolomic mapping of organogenesis in a Carnegie stage 12 human embryo

Article Title: Spatial multi-omic atlas profiles early organogenesis of a Carnegie stage 12 human embryo

Article References: Wang, L. J., Yang, H., Song, J., Zhao, R., Liu, S., Hou, H., Zhang, C., Tian, H., Gao, J., Chen, J., Chen, Z.-J., Zhao, H., Liu, B., & Wu, K. (2026). Spatial multi-omic atlas profiles early organogenesis of a Carnegie stage 12 human embryo. Nature Cell Biology. https://doi.org/10.1038/s41556-026-02086-5

Image Credits: AI Generated

DOI: 10.1038/s41556-026-02086-5

Keywords: human embryology, organogenesis, Carnegie stage 12, spatial transcriptomics, spatial metabolomics, human liver development, heart development, neural development, skeletal muscle, MALDI mass spectrometry, Spatial, multi-omic