Introduction

The placenta, a sophisticated and unique organ, plays an indispensable role during mammalian pregnancy. It not only serves as a physical barrier between maternal blood and the fetus, but also performs crucial functions such as delivering oxygen and nutrients to the fetus and removing waste^1,2,3. A dysfunctional placenta is often linked to various human pregnancy-related disorders, including miscarriage, preeclampsia, and fetal intrauterine growth restriction^4,5,6.

Morphological and functional similarities between human and mouse placenta make the mouse a key model for studying the impact of placenta development on pregnancy and the mechanisms underlying placental dysfunction^7,8. The mature mouse placenta comprises the maternal decidua basalis and the fetal-derived labyrinth and junctional zone^9,10. The labyrinth mediates nutrient and gas exchange, and its disruption often leads to developmental failure and growth restriction, highlighting its essential role in nutrient transfer from mid-gestation onward^11,12. By contrast, the junctional zone is mainly endocrine and consists of spongiotrophoblast (SpT) and glycogen trophoblast cells (GC)¹³. GCs are named for their storage of glucose as glycogen, although the physiological significance of this storage remains unclear¹⁴. Beginning around E12.5, immature GCs gradually mature, accumulating glycogen and acquiring a vacuolated morphology, with GC numbers increasing nearly 300-fold between E12.5 and E16.5 before declining by approximately 50% by E18.5¹⁵. In humans, aberrant placental glycogen storage is associated with maternal diabetes and pre-eclampsia^16,17,18,19, and over 40 targeted mouse mutations demonstrate that defects in GCs compromise fetal growth¹⁴, suggesting placental glycogen may provide readily mobilized glucose during periods of high fetal demand^15,20. However, direct functional evidence is still lacking. Importantly, up to 68% of knockout mouse lines that are lethal at or after mid-gestation exhibit placental abnormalities, and early embryonic deaths between E9.5 and E14.5 are almost invariably associated with severe placental malformations¹¹. These defects correlate strongly with abnormal brain, heart, and vascular development²¹, highlighting the critical need to elucidate the cellular and molecular mechanisms by which placental dysfunction compromises normal embryonic growth and viability.

Previous studies have uncovered cell types comprising trophoblast derived lineages of mouse placentas at various developmental stages (from E7.5 to E14.5) using single cell/nucleus RNA-sequencing^22,23. Yet, our understanding of placental spatial organization, molecular dynamics, and functional maturation, and of how their disruption affects embryonic development, remains limited. Here, we reconstructed a spatiotemporal atlas of mouse placentation across E9.5 to E18.5 using Stereo-seq, the data is accessible via interactive data portal, named spatiotemporal transcriptomic atlas of mouse placenta (STAMP) (https://db.cngb.org/stomics/stamp/). This resource advances insight into placental ontogeny and cell-cell interactions at the maternal-fetal interface and establishes a reference for evaluating genetic or pathological perturbations. Mining this atlas revealed that glycogen trophoblast cells progressively migrate from the junctional zone to the maternal decidua while undergoing transcriptional state changes and forming previously unrecognized subclusters with distinct molecular profiles. Application to a defective placentation model with perinatal lethality demonstrated that disruption of glycogen metabolism compromises fetal survival. These findings highlight placental glycogen metabolism as a fundamental requirement for sustaining fetal viability.

Results

A spatiotemporal single-cell transcriptomic atlas of mouse placentation

To gain a comprehensive understanding of cell fate transitions and specifications at spatial and temporal resolution during placentation, we first profiled 9 mouse placentas spanning developmental stages from E9.5 to E18.5 using Stereo-seq and snRNA-seq (Figure 1A, S1A, S1B and S1C). After rigorous quality control procedures (see Methods), we identified 35 distinct cell clusters, broadly categorized into trophoblast cells (14 subclusters), stromal cells (4 subclusters), immune cells (7 subclusters), endothelial cells (4 subclusters), and other cell types from both fetal and maternal tissues (Figure 1B, S1D, and Supplementary Table 1)^{22,23,24,25,26}. We validated and confirmed these cell annotations by cross-referencing them with existing placental datasets from prior studies (Figure S1E)^22,23.

Figure 1.

Next, we employed DNA staining combined with the watershed algorithm²⁷ to achieve single-cell resolution, enabling the capture of transcripts within clearly defined nuclear and cytoplasmic boundaries. Through spatially constrained clustering (SCC), we delineated distinct clusters corresponding to key anatomical regions, including the labyrinth, junctional zone (JZ), maternal decidua, and metrial gland (Figure S2A). Using snRNA-seq data as a reference, we annotated the Stereo-seq dataset with the TACCO (v0.3.0) framework²⁸, assigning the most probable cell type to each individual cell (see Methods for details). This strategy allowed us to comprehensively map the spatial and temporal transitions of placental cell types across developmental stages from E9.5 to E18.5. The full dataset is publicly available on our interactive platform, the Spatiotemporal Transcriptomic Atlas of Mouse Placenta (https://db.cngb.org/stomics/stamp/) (Figure 1C, 1D, S2B, S2C and S2D).

Firstly, our study delineates distinct spatial distributions and dynamic fate transitions among placental cell types across key anatomical regions, illuminating their region-specific functions and overall significance during embryonic development. In the labyrinth region, we observed expansion through branching morphogenesis that creates intricate vascular spaces essential for maternal-fetal blood exchange²⁹. Notably, the proportion of trophoblast progenitor cells in the labyrinth decreases steadily from E9.5, while terminally differentiated trophoblast cells surge after E12.5, marking critical milestones in placental maturation (Figure 1C and 1D). Simultaneously, spongiotrophoblast (SpT) and glycogen trophoblast cells (GCs) begin to mature from E12.5 onward.

Remarkably, we detected a spatial transition of GCs from the junctional zone (JZ) to the maternal decidua. GCs first appear in the JZ at E12.5, peak at E14.5, and then decline, while their emergence in the decidua begins at E13.5 and increases steadily until E18.5 (Figure 1C and 1D). This pattern corroborates previous immunostaining studies using protocadherin 12 (PCDH12) as a GC marker^15,30. In contrast, other cell types, including endothelial cells, immune cells, and stromal cells are predominantly localized within the maternal decidua (Figure 1C and 1D).

Leveraging our spatial transcriptomic dataset, we further identified multinucleated syncytiotrophoblast cells and trophoblast giant cells (TGCs) across multiple developmental stages in situ (Figure 1C and 1D), a feat previously unattainable with conventional single-cell or single-nucleus RNA-seq methods^22,23. Additionally, we captured the spatial dynamics of two distinct decidual stromal cell (DSC) populations: angiogenic DSCs and nourishing DSCs, each exhibiting unique molecular signatures and functions²⁴. From E9.5, angiogenic DSCs migrate toward the upper decidua, near the myometrium, to promote angiogenesis, stabilizing in number after E14.5. Conversely, nourishing DSCs, which are involved in energy homeostasis and hormone metabolism^24,31, gradually decline and shift toward the fetal side of the placenta (Figure 1C and 1D).

In summary, by integrating spatial transcriptomics and snRNA-seq datasets, we decoded spatially resolved transcriptomes at single-cell resolution and determined cell type composition during mouse placentation. Crucially, the dynamic spatial changes of cell types over time illuminated key developmental milestones during placentation.

Molecular and spatial characterization of glycogen trophoblast cell subtypes

We next examined the spatiotemporal development of trophoblast cells, the primary cell types of the placenta. Reclustering the snRNA-seq data revealed two distinct glycogen cell (GC) subclusters (Figure 2A and Supplementary Table 2). Mapping their spatial distributions using Stereo-seq data showed that one subcluster (GC-1) was predominantly in the junctional zone (JZ), while the other (GC-2) was confined to the maternal decidua (Figure 2B).

Figure 2.

To dissect the molecular differences between these GC subclusters, we performed differential gene expression analysis (Figure 2C and Figure S3A; Supplementary Table 3). Aldh1a3, a known marker of glycogen cells and their precursors³², was highly expressed in GC-1, whereas Prl7b1, overlapping with Pcdh12 expression in the maternal decidua³³, was more abundant in GC-2. These patterns were corroborated by their spatial distributions (Figure S3B), and were further confirmed by RIBOseq analysis³⁴, which provided spatially resolved measurements showing predominant Aldh1a3 and Prl7b1 protein synthesis in the E14.5 and E18.5 placental sections (Figure S3C). Gene ontology analysis further revealed that GC-1 was enriched in processes such as chromatin remodeling and GTPase-mediated signal transduction, while GC-2 was associated with epithelial cell migration and vascular development (Figure 2C; Supplementary Table 4). Notably, genes shared between GC-1 and GC-2 were enriched in glycogen metabolism pathways (Figure 2C; Supplementary Table 4).

To explore the transcriptional dynamics underlying GC fate, we conducted SCENIC analysis to assess transcription factor (TF) expression and motif enrichment. This approach identified key regulons with differential activity between the subclusters, including established TFs (Tfap2c, Tcf12, Prdm1, Fosl2)^35,36,37,38 as well as novel ones (Nfib, Elf4, Etv6, Anxa11 in GC-1; Creb5 in GC-2). Spatial transcriptomics validated these regulons in E14.5 placenta sections (Figure 2D and S3D). Furthermore, using the CellChat algorithm³⁹, we uncovered distinct cellular communication networks: GC-2 showed signaling interactions (e.g., Sema3e-Plxnd1) indicative of an angiogenic role⁴⁰, as well as pathways involving Lgals9 and receptors such as Ptprc, Ighm, Havcr2, and Cd44, suggesting immunomodulatory functions. Additionally, Ccl27a, which is involved in immune cell recruitment, was predominantly expressed in GC-2, while both subclusters secreted growth factors like Igf2 (Figure S4E and S4F).

Temporal trajectory analysis of the GC lineage revealed that GC-1 emerges early in the JZ, peaks briefly, and then declines, whereas GC-2 appears after E13.5, becomes restricted to the maternal decidua, and increases steadily until E18.5 (Figure 2E and S4G). Overall GC numbers rise from E12.5 to E16.5 before markedly decreasing by E18.5 (Figure 2E and S4G), consistent with reports of extensive GC lysis during the perinatal period^15,41. Trajectory reconstruction using Monocle3⁴² and CytoTRACE⁴³ further delineated a developmental pathway following a JZ progenitor – GC precursor – GC-1 – GC-2 sequence (Figure 2F, 2G, 2H, 2I, and S4A).

In summary, our findings reveal previously uncharacterized, transitioning GC subtypes with distinct spatial distributions, molecular profiles, and regulatory networks, and uncovered transcriptional transitions accompanying their migration from the junctional zone to the maternal decidua from E12.5 onward.

Trajectory analysis reveals regulators of GC lineage linked to embryonic lethality

Embryonic lethality in mice is often linked to impaired placental function¹¹. To investigate this connection, we compiled a comprehensive list of 151 mouse mutant lines associated with embryonic death and placental development defects. This list includes 103 knockout lines from the Deciphering the Mechanisms of Developmental Disorders (DMDD) program (https://dmdd.org.uk) and an additional 48 mutant lines from other studies¹¹. We categorized these genes by the timing of lethality and reviewed the literature for evidence of placental abnormalities (Table 1).

Table 1.

To further elucidate placental development, we performed trajectory analysis of trophoblast cells using Monocle3 and CytoTRACE^42,43, which revealed five principal trophoblast lineage progression trajectories, including the GC lineage (Figure S4A and S4B). Along these trajectories, we identified genes with temporally regulated expression patterns along pseudotime that likely drive key stages of placental development (Figure S4C)⁴⁴.

We identified Ano6, a member of the Anoctamin chloride channel family, as a crucial factor in GC lineage development and placental function (Figure 3A, 3B, S5A, and S5B), with previous studies showing that Ano6-deficient mice exhibit impaired placental development and perinatal lethality⁴⁵. RT-PCR analysis indicates that Ano6 is uniquely expressed and upregulated in the placenta (Figure S6A). SnRNA sequencing data further reveal that Ano6 is specifically highly expressed in GCs at various developmental stages (Figure S6B and S6C), a finding that is also corroborated by RIBOseq experiments, which provided spatially resolved measurements showing predominant Ano6 protein synthesis in the junctional zone (JZ) at embryonic day 13.5 (E13.5) (Figure 3C).

Figure 3.

Spatial transcriptomic profiling of placental defects in an embryonic-lethal mutant

To uncover the mechanisms behind placental defects linked to embryonic lethality, we generated Ano6 knockout (KO) mice by deleting exon 3 of the Ano6-208 transcript, resulting in a frame-shift mutation (Figure 3D, S6D and S6E). Due to the low postnatal survival rate of homozygous KO mice (3.03%) (Table 2), we bred heterozygous pairs to obtain KO offspring. Compared with wildtype (WT) and heterozygous (HET), KO placentas were smaller and lighter, and displayed prominent large white plaques on the fetal side, suggesting defects in vascular formation (Figure 3E and 3F), consistent with the previous report⁴⁵. The phenotypic differences among WT, HET and KO placentas became apparent at embryonic day 18.5 (E18.5) (Figure S6F and S6G). Specifically, KO placentas exhibited labyrinth defects characterized by enlarged cavities, reduced area and cellularity, and impaired vascular structure (Figure S6H-M), consistent with previously reported phenotypes⁴⁵.

Table 2.

To probe molecular differences in situ, we combined Stereo-seq with snRNA-seq analysis of WT, HET, and KO samples at E18.5 (Figure S7A and S7B). This integrated approach identified various cell clusters, including trophoblasts, immune cells, stromal cells, and endothelial cells (Figure 3G, 3H, 3I, and S7C). Notably, KO placentas at E18.5 contained markedly increased numbers of glycogen trophoblast cells (Figure 3J and S7D), indicating potential abnormalities in their development or function. While total macrophage numbers were similar across genotypes (Figure S7D), macrophages were enriched in the labyrinth layer of KO placentas (Figure 3K, 3L, and S7E), as confirmed by F4/80 immunostaining (Figure 3M and 3N).

In summary, in addition to phenotypes consistent with previous reports, our knockout placentas at E18.5 showed increased glycogen trophoblast cells, and abnormal macrophage accumulation in the labyrinth, suggesting that impaired glycogen metabolism and aberrant immune responses may underlie placental dysfunction and perinatal lethality.

GC persistence and impaired glycogen catabolism in Ano6-null placentas

The increased number of glycogen cells in KO placentas was confirmed by Periodic acid-Schiff (PAS) and hematoxylin and eosin (H&E) staining (Figure 4A, 4B, S8A, and S8B). Under normal physiological conditions, glycogen trophoblast cells (GCs) expand nearly 300-fold by E16.5 but decline markedly by E18.5 as they undergo lysis/apoptosis and form glycogen-filled lacunae near vascular sinuses, a process thought to supply energy at term or influence parturition³⁰. Since GCs serve as glycogen reservoirs, we next used transmission electron microscopy (TEM) to visualize glycogen granules (Figure 4C). Quantitative analysis showed that KO GCs contained significantly more glycogen granules than those in WT and HET placentas (Figure 4D and 4E), indicating impaired glycogen metabolism. Consistently, total placental glycogen content was significantly elevated in KO placentas compared with WT and HET controls (Figure 4F). Adjustment for placental weight yielded similar results, excluding placental size as the explanation for this increase (Figure 4G). Time-course analysis of total placental glycogen from E12.5 to E18.5, using both quantitative measurement and PAS staining of placental sections, further demonstrated that this phenotype emerged only at E18.5, when KO placentas exhibited significantly higher glycogen content than WT and HET placentas (Figure S8C, S8D and S8E).

Figure 4.

In summary, at E18.5, KO placentas showed increased GCs with excessive undegraded glycogen granules, indicating disrupted glycogen catabolism.

Defective GC glycogen degradation reduces fetal energy supply and compromises survival

To confirm defective glycogen breakdown in KO placentas, we measured key intermediates of glycogenolysis in both placental and fetal liver tissues by LC-MS analysis, including glucose-1-phosphate (G1P), glucose-6-phosphate (G6P) and glucose (Figure 5A and 5B). KO placentas showed markedly reduced levels of these metabolites (Figure 5C), and a similar reduction was also observed in fetal liver (Figure 5D), supporting the idea that impaired glycogen degradation in placental GCs compromises fetal energy supply and contributes to perinatal lethality. During late gestation, when energy demand is high^15,20, the sharp decrease in G6P and glucose suggests insufficient metabolic fuel to sustain placental and fetal development^14,30,46. To test this hypothesis, pregnant dams received daily oral glucose gavage from E13.5 to E18.5 (Figure 5E). Glucose supplementation significantly improved KO fetal survival (from 3.03% to 10.8%) and restored key metabolites, particularly energetically favorable metabolites G6P and glucose, in both placenta and fetal liver (Figure 5F, 5G, 5H, and Table 3).

Figure 5.

Table 3.

To investigate the cause of abnormal glycogen metabolism in KO placentas, we examined GC differentiation and the expression of glycogenolytic enzymes¹³. GC marker genes, Pchd12, Aldh1a3 and Gjb3, showed comparable expression among WT, HET, and KO placentas from E13.5-E18.5 (Figure S9A), indicating intact GC differentiation. Likewise, key glycogenolytic enzymes were expressed at similar levels across genotypes (Figure S9B and S9C). These findings suggest that the metabolic defects in KO placentas are not due to impaired GC differentiation or transcriptional downregulation of glycogenolytic enzymes, but likely reflect a functional impairment in glycogen utilization.

Together, these results demonstrate that loss of Ano6 blocks glycogen breakdown in placental GCs, resulting in metabolic insufficiency that underlies placental dysfunction and perinatal lethality, while maternal glucose supplementation can partially rescue this defect by restoring energetically favorable metabolites.

Secondary macrophage accumulation in the labyrinth of Ano6-null placentas

In addition to the GCs-related defects identified above, previous studies have reported that Ano6 deficiency also causes defective trophoblast syncytialization in the SynT-2 layer, leading to maternofetal exchange deficiency, labyrinth malformation, and perinatal lethality⁴⁵. These findings suggest that multiple pathological processes may converge to cause placental insufficiency in Ano6-null mice.

To further leverage the strength of spatial transcriptomics in capturing in situ microenvironmental changes that are often missed by dissociative single-cell approaches, we next explored cell type distributions within the labyrinth. Interestingly, our spatial data revealed a pronounced accumulation of macrophages specifically in the labyrinth region of KO placentas, which aligns with the known labyrinthine structural defects and highlights the added value of our spatial atlas in detecting region-specific cellular alterations.

Differential gene expression analysis of the labyrinth layer identified 521 genes upregulated and 253 genes downregulated in KO placentas (adjusted p<0.05, |log2FC|>1) (Figure S10A; Supplementary Table 5). Upregulated genes were enriched for innate immune response and inflammatory pathways, including Toll-like receptors (Tlr7, Tlr13)⁴⁷, chemokine ligands (Ccl4, Ccl6), and immunomodulatory molecules (Sirpa, Ccr5, Spp1, Ptafr, Trem2) (Figure S10A, S10B and Figure S11)^{48,49,50,51,52,53,54}, whereas several solute carrier transporters (Slc7a2, Slc22a4, Slc28a3) were downregulated (Figure S10A, 10C, and Figure S12). Gene ontology analysis confirmed enrichment of terms related to TNF production, leukocyte migration, and tissue remodeling among the upregulated genes (Figure S10D and Figure S11).

Focusing on macrophages, we found 140 genes significantly upregulated in labyrinthine macrophages compared with maternal-side macrophages (adjusted p<0.05, |log2FC|>1) (Figure 6A; Supplementary Table 6), including tissue remodeling genes (Spp1, Mmp12, Ctss, Gpnmb) (Figure 6B and 6C)^{55,56,57,58,59}. Immunostaining showed that F4/80⁺ macrophages co-expressed TGFβ (Figure 6D and S13A), indicating an activated state potentially involved in local immune regulation and tissue repair⁶⁰. Their spatial distribution is closely associated with areas of vascular abnormalities marked by CD31 (Figure S13B), suggesting that these macrophages may represent a secondary response to tissue injury and vascular defects in the KO labyrinth (Figure 6E)⁶¹.

Figure 6.

Discussion

The placenta is a complex organ whose proper development is essential for sustaining pregnancy and fetal growth^1,3,62,63,64. However, how the spatial organization and molecular dynamics of placental cells support this function, and how their disruption contributes to embryonic lethality, remain incompletely understood. Here, we constructed a spatiotemporal transcriptomic atlas (STAMP) of the mouse placenta from E9.5 to E18.5, which delineates cell type composition, spatial organization, and developmental trajectories at single-cell resolution. This atlas revealed a progressive shift of glycogen trophoblast cells (GCs) from the junctional zone (JZ) to the maternal decidua, and uncovered previously unrecognized GC subclusters with distinct molecular profiles and regulatory networks. These findings provide a comprehensive resource for investigating placental ontogeny and pathological perturbations.

Importantly, our study provides direct functional evidence linking placental glycogen metabolism to fetal viability. In a lethal Ano6-null model, GCs abnormally persisted at E18.5 and accumulated excessive glycogen granules, while placental and fetal liver tissues exhibited markedly reduced levels of glycogen breakdown intermediates (G1P, G6P, glucose). These results indicate that failure to mobilize placental glycogen deprives the fetus of sufficient metabolic fuel during late gestation, when energy demand peaks^14,16. Consistently, maternal glucose supplementation restored metabolite levels and significantly improved fetal survival, supporting the notion that GCs serve as an auxiliary energy reservoir to sustain fetal growth under high metabolic demand. This work fills a major knowledge gap by functionally validating the long-suspected role of placental glycogen as a readily mobilizable energy source during late gestation^14,16.

In addition, our spatial data revealed transcriptional state transitions accompanying GC migration from the JZ to the decidua. GC-1, marked by Aldh1a3, arises in the JZ and peaks transiently, while GC-2, marked by Prl7b1, emerges after E13.5 in the decidua and expands until E18.5. This spatial shift, validated by RIBO-seq³⁴, aligns with classical histological observations of vacuolated, PAS-positive GCs infiltrating the decidua^15,33,65,66, and highlights the dynamic nature of GC lineage specification. These insights underscore how spatial transcriptomics can capture developmental trajectories and cellular plasticity that are challenging to resolve using dissociative single-cell methods alone.

Beyond the GC-centered metabolic defects, we also observed a secondary phenotype of macrophage accumulation in the labyrinth of Ano6-deficient placentas. This feature aligns with previously reported labyrinthine structural defects resulting from impaired trophoblast syncytialization, and likely represents a response to local tissue injury rather than a primary pathogenic mechanism⁴⁵. While not directly implicated in fetal lethality, this finding illustrates the added value of our spatial atlas for uncovering region-specific microenvironmental changes that are invisible to conventional single-cell approaches.

Collectively, our study establishes a spatiotemporal transcriptomic framework for placental development and demonstrates that GC-mediated glycogen metabolism is indispensable for fetal survival. By functionally linking placental glycogen utilization to fetal viability, this work advances our understanding of placental physiology and provides a reference for identifying pathogenic mechanisms underlying pregnancy complications.

Materials and methods

Key resources table

Resource availability

Lead contact

Further information and requests for the resources and reagents may be directed to the corresponding author Xiaodong Liu (liuxiaodong@westlake.edu.cn).

Materials availability

This study did not generate new unique reagents. All the reagents in this study were included in the key resources table.

Experimental model and study participant details

Animals

All strains of mice concerned in the study were in the C57BL/6J genetic background. The Ano6^+/- mouse was generated by Cyagen Biosciences. The knockout (KO) mouse for the Ano6 gene was generated using a CRISPR/Cas9-mediated strategy. Two guide RNAs (gRNA-B1 and gRNA-B2) were designed to target sequences flanking exon 3 of Ano6. Cas9-induced double-strand breaks at these sites resulted in the deletion of the intervening exon, creating a frameshift and functional knockout of the gene. The specific gRNA sequences used were: gRNA-B1: CGGTGACTGCAGTGTTAAGGTTGG; gRNA-B2: CAGAAATCAGCCAAACGAAGCAGG. The Ano6 mouse mating strategy was crossing heterozygous male mice with heterozygous female mice in order to generate wildtype, heterozygous and homozygous offspring. All animal experiments were carried out with permission of the ethics committee on laboratory animal welfare of Westlake University (AP#22-032-2-LXD-7). All mice were housed in an SPF facility under a 12/12 hours light/dark cycle. Animals were mated at 16:00–17:00, and the vaginal plugs were checked by visual inspection the next morning (8:00–9:00).

Method details

Placenta dissection

Pregnant mice were euthanized, and their placentas were quickly dissected from the uterus. Cold DPBS was used to wash the placentas three times to remove blood. Yolk sac DNA was extracted to confirm genotypes (Vazyme, PD101-01). Surface water was carefully wiped away before the tissue was embedded in OCT (SAKURA, 4583), ensuring that air bubbles were avoided. The frozen blocks were then stored at −80°C. For paraffin embedding, the placentas were fixed overnight in 4% PFA at 4°C, dehydrated through a series of ethanol washes, cleaned three times in xylene, and finally embedded in paraffin (Leica, 39601006) overnight.

Stereo-seq experiment

Frozen sections with a thickness of 10 μm were attached to the surface of a 1 cm x 1 cm stereo-seq chip. The tissues were fixed with pre-chilled methanol at −20°C for 30 minutes, then stained with Qubit ssDNA (Invitrogen, Q10212). Tissue permeabilization was performed using STOmics PR Enzyme (1000028500) for 5 minutes. Reverse transcription was conducted at 42°C using the STOmics Gene Expression kit. Bead-purified cDNA was then amplified by PCR, followed by another round of bead purification. For library construction, the cDNA was fragmented and amplified. The products were subsequently filtered twice, first with 0.6X and then with 0.2X beads.

Nuclei extraction and snRNA-seq

We prepared 200 mg of frozen tissue and transferred it into a 1.5 ml microcentrifuge tube, taking care not to thaw the tissue before lysis. We added 300 µl of NP40 lysis buffer, minced the tissue using scissors, then added an additional 1 ml of NP40 lysis buffer. We incubated the mixture on ice for 5-8 minutes to allow for lysis. After lysis, the suspension was passed through a 30 µm filter into a 5 ml tube and centrifuged at 4°C at 500 g for 5 minutes. Next, we added 1 ml of PBS containing 1% BSA and 1 U/µl RNase inhibitor, mixed gently by pipetting, and conducted microscopic quality checks. We assessed the integrity of the nuclear membrane and background using Trypan blue under a microscope and applied AOPI stain to evaluate nuclear concentration, viability, and aggregation of nuclei. After confirming satisfactory quality, we proceeded with labeling for downstream analysis as soon as possible.

Tissue sectioning

All frozen blocks were equilibrated in the box for at least 30 minutes before sectioning. The frozen embedded tissue was sectioned sagittally until reaching the umbilical cord, with each section being 10 µm thick. Paraffin-embedded sections were cut at a thickness of 5 µm. For the histological examination of the placenta, at least three sections, positioned 70 µm apart, were analyzed per placenta.

Hematoxylin-Eosin (H&E) staining

Frozen slices were allowed to recover for 20 minutes at room temperature before being fixed in 4% PFA for 3 minutes. Tissue sections were stained with hematoxylin for 6 minutes and with eosin for 2 minutes. Dehydration was carried out through an alcohol gradient of 75%, 85%, 95%, and 100%, followed by two xylene washes, each lasting 1 minute. Paraffin slices were incubated in hematoxylin for 6 minutes, rinsed in water, and then stained in eosin for 30 seconds. Spatial transcriptomic sections and all validation analyses (H&E, IF, and RIBOseq) were prepared from adjacent or same-block placental regions to ensure anatomical comparability across datasets.

Immunofluorescence staining

Placentas were dissected and post-fixed in 4% PFA/PBS for 30 minutes, then permeabilized in 0.1% Triton X-100/PBS for 30 minutes. Blocking was performed with 5% donkey serum and 1% BSA for 30 minutes at room temperature. Sections were incubated with diluted primary antibodies overnight at 4°C, washed three times with PBS, and then incubated with secondary antibodies for 1 hour at room temperature. Primary and secondary antibodies were used at the following concentrations: chicken anti-rat MCT1 (sigma, AB 1286-1, 1:200), rabbit anti-rat MCT4 (Millipore, AB3314P, 1:100), rat anti-mouse CD31 (BD Pharmingen, 557355, 1:200), rabbit anti-mouse F4/80 (CST, 30325, 1:200), rabbit anti-mouse F4/80 (HUABIO, HA721520, 1:100), goat anti-rabbit IgG H&L (Abcam, Alexa Fluor 488, 1:500), goat anti-chicken IgY H&L (Abcam, Alexa Fluor 555, 1:500).

Periodic Acid–Schiff (PAS) staining

PAS staining was performed using a PAS staining kit (Cat #G1280, Solarbio) following the manufacturer’s instructions. Paraffin sections were dewaxed to distilled water. The sections were treated with an oxidant for 10 minutes and subsequently rinsed in running tap water for 5 minutes. Samples were then stained with Schiff reagent for 15 minutes and rinsed again for 5 minutes. The sections were counterstained with Mayer’s hematoxylin solution for 3 minutes. Following conventional dehydration and clearing in xylene, the sections were mounted with neutral gum. Quantification of the PAS-positive area relative to the total placental area for each section was performed using ImageJ software.

Quantitative PCR (qPCR)

Total RNA was extracted using the FastPure Cell/Tissue Total RNA Isolation Kit V2 (Vazyme RC112). Each RNA sample was reverse-transcribed with the GoScript Reverse Transcription system (A5001, Promega) using Oligo(dT) and random primers. Relative quantitation was determined using the BioRad CFX Touch and calculated by the comparative Ct method (2−ΔΔCt), with the expression of β-Actin serving as the control.

Ribosome-bound mRNA mapping (RIBOmap)

10 μm thick placenta slices were mounted on confocal dishes. Tissues were fixed in 4% PFA for 15 minutes and permeabilized with cold methanol at −20°C for 1 hour. Yeast tRNAs were added to quench the tissues. Splint primers, padlock, and primer probes were added to the tissue and hybridized for 12 hours at 40°C. After washing twice in PBSTR and once with a high-salt washing buffer for 20 minutes each at 37°C, T4 ligase was added to the solution and incubated for 2 hours at room temperature. Rolling circle amplification of the DNA circle by phi29 polymerase was performed at 34°C for 30 minutes and at 30°C for 2 hours. Tissues were washed twice with 0.1% PBST. After incubation in modification solution for one hour at room temperature, tissues were monomerized with 4% acrylamide-0.2% bis-acrylamide for 15 minutes. Coverslips were then placed directly onto the polymerization solution and incubated under nitrogen flow for 1 hour, after which the coverslip was discarded. Tissues were digested in PK buffer mix (0.2 mg/ml PK final) for 1 hour at 37°C. Placentas were stripped for 10 minutes in 60% formamide twice. After 3 hours of incubation at room temperature, the fluorescent oligos were visualized by confocal fluorescence microscopy using LAS X software.

Transmission electron microscopy (TEM)

Placental tissues were rapidly collected and immersed in a solution of 2.5% glutaraldehyde and 2% paraformaldehyde in 0.1 M phosphate buffer (PB) at 4°C overnight for fixation. Post-fixation, the samples were washed three times with PB buffer and subsequently fixed with 2% osmium tetroxide for 60 minutes on ice. This was followed by a further fixation in 2% osmium tetroxide containing 2.5% potassium ferrocyanide for an additional 60 minutes on ice. The samples were dehydrated through a graded ethanol series (30%, 50%, 70%, 85%, 95%, and twice in 100%) for 10 minutes each step. They were then infiltrated and embedded in EPON12 resin. After sectioning into ultra thin slices (70 nm) and staining, the sections were imaged using a Thermo Fisher Talos 120 electron microscope operated at 80 kV. Quantification of glycogen granules per μm² in glycogen cells was performed using ImageJ software.

Placental glycogen content quantification

Whole placentas were immediately weighed on a scale. Each sample was treated with 0.5 mL of 30% KOH saturated with Na₂SO₄. The tubes were then placed in a boiling water bath for 30 minutes until a homogeneous solution formed. After removal from the bath, the tubes were cooled on ice. To precipitate the glycogen, 95% ethanol was added. The samples were left to stand on ice for 30 minutes, followed by centrifugation at 840 g for 30 minutes. The glycogen precipitates were dissolved in 3 mL of distilled water. To each mL of sample, 1 mL of 5% phenol solution was added, followed by the rapid addition of 5 mL of 98% H₂SO₄. The tubes were allowed to stand for 10 minutes, shaken, and then incubated at 30°C for 20 minutes. Blanks were prepared using 1 mL of distilled water. Absorbance was measured at 490 nm. Quantification of glycogen content and the percentage of glycogen weight relative to total placental weight were performed using GraphPad Prism software.

Quantification and statistical analysis

Data Processing of snRNA-seq Data

Single-nucleus RNA sequencing data processing utilized Cell Ranger (version 7.2.0, 10x Genomics) for quality control, alignment to the Mus musculus genome (refdata-gex-mm10-2020-A), and transcript quantification. The integration step employed the Cell Ranger aggr pipeline for sequencing depth normalization. Post-integration, cells with fewer than 500 genes detected or over 10% mitochondrial expression were excluded. The remaining data underwent normalization with Seurat’s NormalizeData function (version 5.0.3).

Batch Correction and UMAP Nonlinear Dimensionality Reduction

The dataset underwent batch correction and dimensional reduction using IntegrateLayers with FastMNNIntegration. Neighborhood identification uses FindNeighbors (reduction = ’integrated.mnn’, dims = 1:30), followed by cluster identification with FindClusters (resolution = seq(0.1, 0.2, 0.1)). Dimensional reduction concluded with RunUMAP (dims = 1:30, reduction = ’integrated.mnn’), visualizing cell clusters in two dimensions.

Identification of Differentially Expressed Genes

Differential gene expression was determined using FindMarkers, focusing on genes with only positive expression in specific clusters, detected in at least 25% of cells (min.pct = 0.25) and showing significant expression differences (logfc.threshold = 0.25).

SCENIC Analysis

Seurat objects converted to loom format enabled SCENIC analysis. GRNBoost2 from Arboreto inferred gene regulatory networks. The workflow involved runSCENIC_1_coexNetwork2modules, runSCENIC_2_createRegulons, and runSCENIC_3_scoreCells, assessing regulon activity across cell types with cisTarget databases set to mm10, and visualized through heatmaps.

Pseudotime Analyses and Characterization of GC Subtypes

Monocle 3 was used to transform Seurat objects into cell datasets and cluster cells (resolution = 1e-3). Trajectory analysis with learn_graph (use_partition = TRUE) identified developmental pathways, with order_cells setting the JZP cluster as the starting point. Visualization focused on pseudotime and cluster trajectories.

Cell-Cell Communication Analyses

Using CellChat, cell-cell communication networks were analyzed, utilizing the CellChatDB mouse database. IdentifyOverExpressedGenes and identifyOverExpressedInteractions pinpointed crucial genes and interactions. computeCommunProb (type = “triMean”) and filterCommunication (min.cells = 10) refined the communication data.

Cell Type Integration

Data integration across sources standardized Seurat objects to the “RNA” assay using IntegrateLayers and “FastMNNIntegration”. Neighborhoods were defined with FindNeighbors (dims = 1:30), and clusters identified with FindClusters. Visualization was achieved with RunUMAP, harmonizing and analyzing integrated datasets.

Ontology Annotation

GO enrichment analysis with clusterProfiler and org.Mm.eg.db mapped mouse genes to GO terms. Gene symbols converted to Entrez IDs with bitr preceded enrichGO analysis for clusters (snRNA-seq: pvalueCutoff = 0.01, pAdjustMethod = “fdr’’, minGSSize = 10, maxGSSize = 500, qvalueCutoff = 0.01; Stereo-seq: pvalueCutoff = 0.05, pAdjustMethod = “BH’’, minGSSize = 10, maxGSSize = 500, qvalueCutoff = 0.2).

Stereo-seq raw data processing

Single-end Stereo-seq fastq files were generated using a MGI DNBSEQ-Tx sequencer, which contained CID (coordinate identity), MID (molecular identity, UMI) and cDNA sequences. The retained reads were aligned to the mouse reference genome (mm10) (https://ftp.ensembl.org/pub/release-93/fasta/mus_musculus/dna/Mus_musculus.GRCm38.dna.primary_assembly.fa.gz) by STAR. Only mapped reads with a mapping quality score (MAPQ) > 10 were then annotated and calculated using handleBam (available at https://github.com/BGIResearch/handleBam). UMIs with identical CID and gene locus were collapsed, allowing 1 base pair of mismatch to correct for sequencing and PCR errors. Exonic reads were then used to construct an expression profile matrix comprising CIDs.

Image-based Cell Segmentation of Stereo-seq data

We utilized ssDNA staining image from the same section to segment cells by projecting the ssDNA staining image onto the Stereo-seq chip image. Firstly, a grayscale map was generated from the Stereo-seq data, with each pixel representing one DNB. Manual registration was conducted on both the grayscale map and the ssDNA staining image to align the pixels of the staining image with the DNB coordinates. Cell segmentation was then carried out using the scikit-image python package (v0.19.3)^67,68. Global threshold was used to filter the background noise of the registered image, and a mask was generated for segmentation. Gaussian-weighted local threshold was calculated to represent the position of the cell with block size of 45 and offset of 0.03. Cells with overlapped regions were segmented using the exact Euclidean distance transformation (with a minimal distance of 15). We then expanded the labels representing different cells in the label image by 5 pixels without causing overlaps, via expand_labels function. For each segmented cell, UMIs from all DNBs within the corresponding segmentation were aggregated per gene and subsequently summed to generate a cell-by-gene matrix for downstream analysis.

Spatially-constrained clustering of Stereo-seq data and region annotation

Cells with UMI counts below 200 (100 for E9.5, E15.5) were excluded from further analysis. The remaining raw count matrices of the placenta samples were normalized by SCTransform function in Seurat (v4.3.0)⁶⁹. Spatial information was taken into account for unsupervised clustering. The centroid of each cell was computed based on its spatial coordinates. The spatial k-nearest neighbor graph was constructed using Squidpy (v1.2.3)⁷⁰. This graph was then combined with the k-nearest neighbor graph based on transcriptomic data (k₂ is by default set to be 30). The combined graph was used as input for Leiden clustering. We annotated the placental region according to the anatomical structure, combined with spatially-constrained clustering and HE section partitioning.

Single cell level annotation of Stereo-seq data

Cells with UMI counts below 200 (100 for E9.5, E15.5) were excluded from further analysis. The remaining raw count data were normalized by the SCTransform function in Seurat (v4.3.0) to mitigate the effects of sequencing depth. Subsequently, high-quality single-nucleus RNA sequencing (snRNA-seq) data from the corresponding developmental stage were employed as a reference to annotate the Stereo-seq data using the TACCO (v0.3.0) framework²⁶. Specifically, we mapped the Stereo-seq data using snRNA-seq data as a reference and the adjusted snRNA-seq data cell type proportions as the annotation prior distribution via the tacco.tools.annotate function. The annotation process assigned each spatial single cell the cell type with the highest score, indicating the most probable identity.

Acknowledgements

The authors thank all members of the Liu Lab for their constructive advice and comments on the study. We are grateful to the Imaging Platform at Westlake University for the use of microscopes and to the Histochemical Platform for providing technological support in tissue handling. We would like to thank the Microscopy Core Facility (MCF) of Westlake University for our electron microscopy work and the technical assistance from Yilin Sun and Guicun Fang. We also thank the Laboratory Animal Research Center (LARC) at Westlake University for animal care and operational assistance. Furthermore, we would like to thank CHI BIOTECH CO., LTD. for single nuclei RNA sequencing support. This work was supported by the National Natural Science Foundation of China (32370784, 22DAA01467), the National Key R&D Program of China (2022YFA1105700, 2022YFC3400400) and the Westlake Education Foundation. Additionally, we thank the China National GeneBank for providing sequencing services for this project.

Additional Information

Supplementary figure

Figure S1.

Figure S2.

Figure S3.

Figure S4.

Figure S5.

Figure S6.

Figure S7.

Figure S8.

Figure S9.

Figure S10.

Figure S11.

Figure S12.

Figure S13.

Data availability

The data that support the findings of this study have been deposited into CNGB Sequence Archive (CNSA) of China National GeneBank DataBase (CNGBdb) with accession number CNP0004934. The snRNA-seq data and scStereo-seq data are accessible on our interactive data portal at CNSA https://db.cngb.org/stomics/stamp/. All the code in this study can be found in the GitHub repository: https://github.com/xdliulab-AI/Placenta_Project.

Additional information

Author contributions

X.L. conceived the project. X.W., L.L. and X.L. supervised the project. Y.F., S.J., performed all experiments and prepared samples for Stereo-seq and snRNA-seq with help from Y.M., Y.Y., S.W., and X.L. Y.F., X.Z., Y.L. analyzed the data with support from Y.J., T.X., J.S., L.M., X.M., Y.Y., W.P., and L.T. C.Y. and S.S. performed the RIBOmap experiment. Y.F., J.T., M.P.M., S.C., X.W., L.L., and X.L. wrote and revised the manuscript with input from all authors. All authors approved of and contributed to the final version of the manuscript.

Funding

MOST | National Natural Science Foundation of China (NSFC) (32370784)

• Xiaodong Liu

MOST | National Natural Science Foundation of China (NSFC) (22DAA01467)

• Xiaodong Liu

MOST | National Key Research and Development Program of China (NKPs) (2022YFA1105700)

• Xiaodong Liu

MOST | National Key Research and Development Program of China (NKPs) (2022YFC3400400)

• Longqi Liu

Additional files

Tables 1-3 in PDF format.

Supplementary Table 1. DEGs in all cell types, related to Fig. 1. All differential expressed genes (DEG) of all placental and maternal cells.

Supplementary Table 2. DEGs in trophoblast cells, related to Fig. 2. Sub-clustering of trophoblast cells, and all differential expressed genes (DEG) of all trophoblast subtypes.

Supplementary Table 3. DEGs in GC subclusters, related to Fig. 2. All differential expressed genes (DEG) between two GC subtypes, GC-1 and GC-2.

Supplementary Table 4. GO enrichment analysis of GC subclusters, related to Fig. 2. GO enrichment of DEGs in two GC subtypes, GC-1 and GC-2.

Supplementary Table 5. DEGs in the labyrinth of KO placenta and GO enrichment analysis, related to Fig. 5. All differential expressed genes (DEG) and their enriched GO terms in the labyrinth of KO placenta compared to WT placenta (adjusted p < 0.05, |log₂FC| > 1).

Supplementary Table 6. DEGs in the labyrinthine macrophages of KO placenta and GO analysis, related to Fig. 5. All differential expressed genes (DEG) and their enriched GO terms in the labyrinthine macrophages of KO placenta compared to WT placenta (adjusted p < 0.05, |log₂FC| > 1).

Supplementary Table 7. Primers and immunostaining antibodies. Information regarding the PCR and qRT-PCR primers and antibodies used for immunostaining in this study.