Main

The amnion is an extra-embryonic structure essential for the development of reptilian, avian and mammalian embryos as it encases the embryo, providing both mechanical and biochemical support. In humans, the amnion originates from a subset of pluripotent epiblast cells specified soon after implantation. Before implantation, the blastocyst consists of an outer extra-embryonic trophoblast layer and the inner cell mass that will differentiate into epiblast and extra-embryonic hypoblast by days 6–7 post-fertilization1,2. Upon implantation, epiblast cells polarize to form a rosette structure that undergoes lumenogenesis, creating the amniotic cavity as cells exit naive pluripotency3,4,5. Epiblast cells in contact with the hypoblast form the epiblast disc, which will develop into the embryo proper, while those in contact with the trophectoderm will form the amnion (Fig. 1a). The amnion not only physically protects the embryo but also secretes essential hormones and cytokines that support embryonic development6,7.

Fig. 1: scRNA-seq analysis of human amnion during the first trimester.
figure 1

a, A diagram illustrating the development of the human amnion. Trophectoderm is marked in grey, amnion in orange, extra-embryonic mesoderm in red, hypoblast or yolk sac endoderm in green, epiblast/embryo body in burgundy or pink and umbilical cord in purple. b, Images of human embryos representing different stages. CS16–CS19 were the collected sample, and CS22 is a representative image taken from the same centre. Arrows point to the amniotic membrane, and triangles mark the yolk sac. Scale bar, 0.5 cm. c, UMAP displaying the identified cell types within the analysed samples. d, A bubble plot showing selected top marker gene expression across cell types. e, Immunofluorescence (IF) staining of amnion sections, with protein markers labelled at the top of the image and amnion stages labelled at the bottom. Different fluorescent markers highlight the localization of the proteins in the tissue. The region inside the white box is magnified on the right. White arrows indicate double-positive cells. Scale bars, 50 µm. CD45–VIM co-staining is representative of four independent experiments that yielded similar results. CD45–E-Cadherin co-staining is representative of two independent experiments. VIM–E-Cadherin and N-Cadherin–KRT18 are representative of one experiment. scRNA-seq analyses depicted in this figure are generated from human amnion samples of the following developmental stages: CS16 (n = 1), CS17 (n = 1), CS19 (n = 1) and CS22 (n = 1).

Source data

Full size image

The human amnion is composed of two primary cell types—epithelial cells and mesenchymal cells—separated by a thick basement membrane8,9. Amniotic epithelial cells, which line the amniotic cavity, are responsible for producing the amniotic fluid, whereas the amniotic mesenchymal cells, embedded within the extracellular matrix, contribute to the structural scaffold of the avascular foetal membranes10. These membranes define the intrauterine cavity and protect the foetus during gestation11.

The amnion undergoes extensive growth, repair and remodelling throughout pregnancy to align with embryonic development. These processes are closely associated with epithelial-to-mesenchymal transitions (EMT) and mesenchymal-to-epithelial transitions (MET)11,12,13. In addition, EMT in the amnion has been reported to influence the immune properties of amniotic epithelial cells, often associated with localized inflammation and facilitating tissue remodelling14.

Beyond its fundamental role in pregnancy, the human amnion serves as a valuable source of stem cells with multilineage differentiation potential. The stem cells derived from amnion can be utilized for cell-based therapies and regenerative medicine applications8,15,16,17,18. The unique properties of the amnion, including low immunogenicity19, anti-inflammatory20 and antimicrobial properties, make it an attractive candidate for various therapeutic applications21, such as wound healing, treatment of ocular surface disorders and tissue engineering22.

Despite growing interest, a comprehensive understanding of human amnion development remains limited. Single-cell RNA sequencing (scRNA-seq) of human pregastrulation embryos23 and of primate gastrulating embryos24,25 has provided transcriptional snapshots at specific stages of early development, offering preliminary insights into amnion development. Advances in stem cell technology, including the generation of stem cell-derived embryo-like models25,26,27 and stem cell-derived amnion-like cells28,29, have further clarified the developmental pathway of amnion specification. RNA sequencing (RNA-seq) of amnion tissue collected from pregnant women at term has confirmed the presence of multiple cell types within the fully developed amnion, including fibroblasts, epithelial cells, immunocytes and various intermediate cell types30. However, characterizing amnion from the early stages of human pregnancy remains challenging. In this study, we used scRNA-seq to profile various cell types present in human amnion during the first trimester of human pregnancy to gain insight into their interactions and potential functional contributions.

Results

Cell composition of human amnion in the first trimester

To explore the dynamics of transcriptional changes during amnion development, we collected seven human amnion samples representing 5–9 weeks of pregnancy and corresponding to Carnegie stages (CS) 16, 17, 19, 22 and 23, respectively. CS16 embryos have developed limb buds, the otic vesicle, early eye structures and the primitive heart tube, along with forming somites and the neural tube. CS17 embryos have developed hand rays, cartilage, ribs, intercostal muscles, mammary glands and the thymus. By CS19, embryos have developed the cerebral aqueduct, middle cerebral artery, renal artery and tibia. By CS22, the embryonic brain has developed nerve cell clusters and bundles of nerve fibres, and ossification has begun in the clavicle and long bones (Fig. 1a,b).

We prepared single-cell suspensions from four of these samples (CS16, CS17, CS19 and CS22) and performed scRNA-seq using the 10x Genomics Chromium system (Extended Data Fig. 1a). Cells with fewer than 500 or more than 8,000 genes expressed were excluded. In addition, we excluded cells with more than 20% mitochondrial reads to remove dead cells. Cell doublets were removed by Souporcell31 analysis. In addition, amnion tissue can be contaminated during dissection with maternal cells such as maternal blood cells and blood vessels, we used Souporcell to analyse and remove cells that could be of maternal origin (Extended Data Fig. 1b). We also scored the cells and excluded those with marker gene expression characteristic of yolk sac, chorion, blood vessels and erythroid cells (Extended Data Fig. 1c,d). In total, 14,027 single cells passed quality control and were included in our analysis. Data from the four stages were integrated using the ‘IntegrateData’ function in Seurat432.

Unsupervised clustering utilizing the Seurat package revealed ten distinct cell clusters defined by their transcriptional signatures (Extended Data Fig. 2a). We identified a total of six major cell types among the ten clusters based on their marker genes (Fig. 1c,d and Supplementary Table 1). These include amnion epithelial cells (AECs, clusters 1 and 5, marked by GABRP and KRT18), amnion mesenchymal cells (AMCs, cluster 0, marked by MGP and VIM), fibroblasts (cluster 3, marked by COL6A1 and COL5A1), macrophages (cluster 9, marked by MRC1 and CD36) and two clusters of actively proliferating cells (marked by CDK1 and TOP2A), which were defined as amnion mesenchymal stem cells (AMSCs, clusters 2, 6 and 8) and amnion epithelial stem cells (AESCs, clusters 4 and 7) based on the expression of lineage-specific genes (Fig. 1d and Extended Data Fig. 2b). Their stem cell characteristics were also demonstrated by subsequent pseudotime analysis. The immunofluorescence staining of sectioned tissues confirmed the presence of epithelial cells (expressing E-Cadherin and KRT18), mesenchymal cells (expressing VIM), fibroblasts (expressing N-Cadherin), and macrophages (expressing CD45) in human CS19 and CS23 amnion tissues (Fig. 1e and Extended Data Fig. 3a–e).

Cell subtypes and lineage trajectories in amnion development

To provide a more comprehensive and detailed depiction of the amnion’s cellular composition, we used three-dimensional (3D) Uniform Manifold Approximation and Projection (UMAP) plots (Fig. 2a and Extended Data Fig. 4a). This approach allowed us to further subdivide the AESCs into two distinct groups, labelled as AESCs_1 (cluster 7) and AESCs_2 (cluster 4). Moreover, we identified a population of intermediate-state cells (cluster 5) that express both epithelial and mesenchymal marker genes (Fig. 2b and Extended Data Fig. 4b). In addition, our analyses revealed a group of cells with high expression of ectoderm markers such as SOX2, TUBB3 and NR2F1 (Fig. 2b and Extended Data Fig. 4c), which we labelled as amnion ectodermal cells (Amnion-Ect, AM-Ect). These ectoderm markers were also found to be expressed in the bulk RNA-seq of first and second trimester of human amnion samples33 (Extended Data Fig. 4d) and in the scRNA-seq of CS8-11 cynomolgus monkey amnion cells34 (Extended Data Fig. 4e,f). Immunofluorescence staining suggests the presence of amnion ectodermal cells (expressing SOX2 and TUBB3) in the CS16 and CS19 amnion section (Fig. 2c and Extended Data Fig. 4g).

Fig. 2: Cell subtypes and lineage trajectories in human amnion.
figure 2

a, A 3D UMAP representation showing different cell subtypes in the amnion. b, A violin plot showing marker gene expression across subtypes. c, Top: immunostaining of GABRP and SOX2 in the CS16 amnion section, with triangles marking the ectodermal cells. Representative image from two independent experiments. Bottom: immunostaining of E-Cadherin and TUBB3 in the CS19 amnion section. Scale bars, 50 µm. Representative image from four independent experiments. d, RNA velocity analysis indicating development tendencies of epithelial and mesenchymal cells, based on the integrated analysis of four independent biological samples. e, Pseudotime and trajectory plots showing the epithelial–macrophage trajectory (top) and mesenchymal trajectory (bottom). f, Cell subtypes arranged along the Destiny pseudotime in two trajectories: epithelial–macrophage lineage (left) and mesenchymal lineage (right). g, The expression of selected differentially expressed genes (DEGs) during lineage progression in amnion: macrophage (top), epithelial (middle) and mesenchymal (bottom). scRNA-seq analyses depicted in this figure are generated from human amnion samples of the following developmental stages: CS16 (n = 1), CS17 (n = 1), CS19 (n = 1) and CS22 (n = 1).

Source data

Full size image

To investigate the developmental trajectories within amnion cells, we utilized three pseudotime analysis methods. RNA velocity35 revealed two primary trajectories: epithelial and mesenchymal. The epithelial trajectory further branched into two distinct paths: one transitioning from AESCs_1 to AESCs_2 and the other from AESCs_1 to AECs and then to macrophages. The mesenchymal trajectory delineated a progression from AMSCs to AMCs and finally to fibroblasts (Fig. 2d). Similarly to the RNA velocity results, trajectory and pseudotime analyses using Monocle336 and Destiny37 also revealed the same two developmental trajectories (Fig. 2e,f).

Based on the UMAP and pseudotime analyses, AESCs_2 and intermediate cells may serve as bridges between epithelial and mesenchymal lineages, suggesting that an EMT occurs in the amnion. Consequently, we examined the expression of transcription factors related to EMT in different cell subtypes. Our analyses indicated that most EMT-related transcription factors were highly expressed in the mesenchymal lineage (Extended Data Fig. 5a). Specifically, expression of SNAI1 and SNAI2 was detected in the AESCs_2 cells, while expression of SNAI2, ZEB1 and TWIST1 was observed in the intermediate cells. Interestingly, we discovered high SNAI1 expression only in the AESCs_2 cells that were closest to the mesenchymal lineage (Extended Data Fig. 5b), suggesting that AESCs_2 might be transitioning into mesenchymal cells through EMT.

Our analyses also identified transcripts that change expression in accordance with pseudotime (Supplementary Table 2). Specifically, along the epithelial–macrophage trajectory, we observed a progressive increase in the expression levels of epithelial and macrophage marker genes, such as GABRP, IGFBP3, MRC1 and CD36. By contrast, mesenchymal marker genes such as MGP and VIM are systematically downregulated (Fig. 2g). Interestingly, this trend was reversed in the mesenchymal trajectory, where mesenchymal marker genes showed an increase in expression, highlighting the distinct and dynamic cellular behaviours in different developmental paths.

Intercellular communication in amnion development

To investigate intercellular communication among amniotic cells, we utilized CellChat38, which uncovered numerous potential interactions between various cell populations (Fig. 3a,b and Extended Data Fig. 6a). We examined the expression of receptors and ligands within cells to identify the roles of different cell types in the interaction network. We found that macrophages and intermediate cells exhibited an outgoing profile, primarily expressing ligands, whereas amnion ectodermal cells displayed an incoming profile, predominantly expressing receptors. Other cell types demonstrated a combination of both outgoing and incoming signalling capabilities (Fig. 3c).

Fig. 3: Cellular communication and secretion patterns within the amnion.
figure 3

a, The total number of inferred signalling interactions among different cell types. b, Overall interaction strength representing the cumulative communication probability between cell types. c, Cell roles in secreting and receiving signals. d, Classification of cells into three distinct secretion patterns based on gene expression profiles. e, A heatmap showing the expression of ligands across different cell types. scRNA-seq analyses depicted in this figure are generated from human amnion samples of the following developmental stages: CS16 (n = 1), CS17 (n = 1), CS19 (n = 1) and CS22 (n = 1).

Source data

Full size image

We classified cells into three patterns based on their expression of genes encoding secreted signalling ligands (Fig. 3d,e and Supplementary Table 3). The epithelial pattern included AESCs_1, AESCs_2, AECs and intermediate cells, which expressed ligands associated with bone morphogenetic protein (BMP), Wingless/Integrated (WNT), platelet-derived growth factor (PDGF) and growth differentiation factor (GDF) signalling. The mesenchymal pattern consisted of AMSCs, AMCs and fibroblasts, which expressed ligands associated with midkine (MDK), non-canonical Wnt (ncWNT), hepatocyte growth factor (HGF) and insulin-like growth factor (IGF). Finally, the macrophage pattern showed expression of secreted phosphoprotein 1 (SPP1) and transforming growth factor beta (TFG-β).

Our analyses identified several growth factors that are linked to specific growth and developmental stages of amniotic cells, illustrating a complex intercellular communication network within the amnion. By analysing ligand and receptor expression, we were able to identify the likely signalling and responding cells across different cell populations (Fig. 4a–c). Specifically, our analyses indicated that within the BMP signalling pathway, which is known to be critical for amnion development28,39,40, cells of the epithelial lineage function primarily as recipients of BMP signals (Fig. 4d). Interestingly, we found that BMP4 was primarily expressed by amnion mesenchymal lineage, whereas BMP7 was predominantly expressed by the epithelial cells themselves (Fig. 4e and Extended Data Fig. 7a). The receptors for these proteins, BMPR1A, ACVR2A, ACVR2B and BMPR2, were mainly expressed in the epithelial lineage cells (Extended Data Fig. 7b). In addition, we also identified potential crosstalk between the MDK and WNT signalling pathways in amnion cells (Extended Data Fig. 7c–f). In our in vitro human stem cell differentiation experiments, BMP4 treatment of induced pluripotent stem (iPS) cells resulted in the upregulation of both early and late amnion markers (Extended Data Fig. 8a).

Fig. 4: Key signalling pathways in distinct amnion cell patterns.
figure 4

a–c, A chord diagram showing ligand–receptor interactions in epithelial (a), mesenchymal (b) and macrophage (c) patterns. Distinct cell types are represented by different colours. d, A heatmap showing interactions in the BMP signalling pathway. Commun prob., communication probability. e, A bubble plot showing the significant interactions in the BMP signalling pathway. f–i, Heatmaps showing interactions in PDGF (f), IL6 (g), TGF-β (h) and SPP1 (i) signalling pathways. scRNA-seq analyses depicted in this figure are generated from human amnion samples of the following developmental stages: CS16 (n = 1), CS17 (n = 1), CS19 (n = 1) and CS22 (n = 1).

Full size image

Some signalling pathways were also related to the characteristics of the amnion, including angiogenesis (PDGF, Fig. 4f), anti-inflammatory effects (IL6, Fig. 4g), EMT promotion (TGF-β, Fig. 4h) and immunosuppression (SPP1, Fig. 4i). The immunomodulatory properties of the amnion were particularly striking as we found that amnion cells expressed various immunosuppressive factors, such as macrophage migration inhibitory factor (MIF) (Extended Data Fig. 8b) and SPP1 (Extended Data Fig. 8c), which play crucial roles in inhibiting immune responses41,42,43,44. Immunostaining of CS16 amnion sections confirmed the expression of MIF, SPP1 and its receptor CD4441 in the amnion (Extended Data Fig. 8d). The expression of immunosuppressive factors provides a possible explanation for the amnion’s capacity to inhibit immune responses. Overall, these pathways highlight the complexity of cellular communication and may play an integral role in coordinating cellular interactions during amnion development.

Analysis of amnion development in vivo and in vitro

To investigate the developmental dynamics of amnion across different species and experimental conditions, we combined published scRNA-seq datasets from human CS724 (Extended Data Fig. 9a) with data from monkey CS8–1134 (Extended Data Fig. 9b). In addition, we incorporated three sets of in vitro-derived amnion-like cells generated from stem cells25,28,39 (Extended Data Fig. 9c–e) and included data from our own study on human amnion samples CS16–22.

By leveraging a combined UMAP analysis, we identified a clear developmental progression of amnion formation (Fig. 5a). To further delineate this trajectory, we isolated amnion cells from these datasets and performed a diffusion map analysis (Fig. 5b). Interestingly, our findings revealed that the different in vitro amnion models correspond different developmental stages in vivo. Specifically, the amnion models derived from human pluripotent stem (hPS) cells using either in a microfluidic device28 or 3D biomimetic culturing39 resembled earlier amnion stages (around CS7). By contrast, amnion-like cells derived from two-dimensional (2D) hPS cell cultures25 more closely reflected later stages of amnion development (CS11–16) (Fig. 5c). This divergence highlights variations in developmental timing across different in vitro models, emphasizing the need for careful selection of model systems to accurately recapitulate in vivo amnion development.

Fig. 5: Combined analysis of amnion data from human, monkey and in vitro stem cell-derived embryo models.
figure 5

a, Combined UMAP of amnion data from human, monkey and in vitro stem cell-derived models. NNE, non-neural ectoderm; AM, amnion; ExE.Meso, extra-embryonic mesoderm; ECT, ectoderm; EPI, epiblast; Mes, mesenchyme; SE1, surface ectoderm1; SE2, surface ectoderm2; VE, visceral endoderm; ESC, embryonic stem cell; AMLC, amnion-like cell; MeLC, mesoderm-like cell; PGCLC, primordial germ cell-like cell. Data were merged from the scRNA-seq data generated in this study and five published datasets. b, A diffusion map illustrating the distribution of various amnion and amnion-like cell populations based on the integrated dataset comprising our data and five published datasets. Arrows indicate the developmental process. AME-E, amnion early like cell; AME-L, amnion late-like cell; AP3, hPS cells that were primed for 3 days; AP8, hPS cells that were primed for 8 days. c, Pseudotime plots of amnion and amnion-like cells derived from this study and five published datasets, with dataset origins indicated on the side. Pseudotime_dm; pseudotime computed using diffusion maps (dm). d, A heatmap showing the expression of each gene module across amnion and amnion-like cells from our data and five published datasets. Gene module numbers are shown on the left. Specific markers and transcription factors (TFs) are shown on the right. e–g, GO enrichment analysis of early amnion (e), AMCs (f) and AECs (g). Data sources include previously published datasets from human CS7 (ref. 24), monkey CS8–11 (ref. 34), and three sets of in vitro-derived amnion-like cells25,28,39.

Source data

Full size image

Building on this integrative analysis, we performed differential expression and gene module analyses along the amnion developmental trajectory. Genes were categorized into distinct modules based on their expression patterns, and their average expression was visualized in a heatmap (Fig. 5d and Supplementary Table 4). We identified several transcription factors, including POU5F1, HAND1 and ISL1, as being enriched in the early stages of amnion development. Gene Ontology (GO) annotations for these genes were predominantly associated with pathways such as ‘embryonic organ development’ and ‘signalling pathways regulating pluripotency of stem cells’ (Fig. 5e). By contrast, late-stage AMCs predominantly express mesenchymal markers such as COL15A1 and MGP, along with transcription factors related to EMT, including GATA6, TWIST1 and ZEB1. GO annotations for these genes were enriched in ‘extracellular matrix’ and ‘mesenchyme development’ (Fig. 5f). In late-stage AECs, gene expression was enriched for epithelial markers such as GABRP and IGFBP5, as well as transcription factors TFAP2A and TFAP2B. GO annotations for these genes were associated with ‘extracellular matrix’ and ‘cell adhesion molecule binding’ pathways (Fig. 5g). This comprehensive genomic analysis illustrates the dynamic changes that occur throughout amnion development.

Discussion

Our single-cell analysis of the human amnion has revealed a dynamic cellular landscape, developmental trajectories and intercellular interactions between different amnion cell types. Within the first trimester amnion, we identified six major cell types and nine cell subtypes spanning epithelial, mesenchymal and macrophage lineages. In addition, we also discovered a population of amnion ectodermal cells expressing some neural-related genes.

Although traditionally considered a non-neuronal tissue45,46,47, the amnion has been reported to contain mesenchymal cells with neural progenitor-like characteristic48,49,50 and neurotransmitter metabolism capabilities51,52,53. Furthermore, neural-related genes such as SOX9, ID4 and STMN2 have been detected in human amnion samples from both the first and second trimester33. Our results further show that SOX2-positive ectodermal cells also express neural markers such as TUBB3 and NR2F1, suggesting that these neural progenitor-like cells in the amnion are probably ectodermal cells. However, higher-resolution immunofluorescence imaging would be required to confirm the presence of SOX2-positive cells and to refine our understanding of their morphology and spatial arrangement.

In our study, the entire amnion was collected without specific positional selection, and potential regional variations within the tissue were not specifically addressed. This limitation may contribute to the observed variability. Future studies using spatial transcriptomics or other positional mapping techniques could provide a deeper insight into the spatial heterogeneity and regional distinctions within the amnion.

Our study further demonstrates the potential developmental pathways of epithelial, mesenchymal and macrophage lineages in the human amnion. The observed EMT and epithelial-to-immune transitions (EIT) suggest dynamic cellular remodelling and immune modulation as the amnion develops during the first trimester of human pregnancy. These findings align with previous studies indicating the presence of EMT and EIT in the amnion and their importance in amniotic membrane remodelling11,13,30,54. However, further research will be necessary to provide deeper insights into these processes and their functional implications.

Beyond working as a protective barrier, the amnion is a major source of several growth factors crucial for embryogenesis, including EGF, FGF, PDGF and VEGF55,56, which are involved in angiogenesis, tissue repair and immunomodulation. However, the specific cell types that secrete these growth factors remained unclear. Leveraging scRNA-seq data, we identified three distinct secretion patterns within the amnion: epithelial (MIF, WNT, BMP, GDF, PDGF and activin), mesenchymal (MDK, ncWNT, HGF, IGF and IL6) and macrophage (SPP1, TGF-β and CCL). Previous studies have shown that BMP4 promotes amnion development28,39, and our recent work confirmed that BMP4 is essential for the epiblast differentiation into amnion in a stem cell-derived human embryo model40. Consistent with these results, we show here that BMP4 treatment of iPS cells led to the upregulation of both early and late amnion marker expression. Our findings reveal that immune cells within the amnion express TGF-β, a cytokine known to promote EMT11. This result suggests a potential mechanism in which macrophages may facilitate EMT via TGF-β secretion, thereby enhancing tissue repair12. Moreover, we observed that amnion cells express immunosuppressive factors such as SPP141, MIF42,43,44 and TGF-β57,58, which are known to inhibit immune responses. These findings further support the immunosuppressive properties of the amnion when used as a clinical biomaterial, highlighting its potential to modulate the immune environment and reduce inflammation during tissue repair. Clinically, this understanding can be leveraged to enhance wound healing, reduce inflammation and promote tissue regeneration in applications such as treating burns, chronic wounds and organ injuries, demonstrating the amnion’s promise in regenerative medicine.

Our comparative analysis of amnion RNA-seq data from human, non-human primate and in vitro amnion models indicates that stem cell-derived amnion models correspond to various stages of in vivo amnion development. Specifically, the amnion models derived from hPS cells using a microfluidic device and 3D biomimetic culture28,39 appear to reflect earlier developmental stages compared with amnion-like cells derived from 2D culturing of hPS cells25. This difference might be linked to their respective culturing methods: microfluidic and 3D biomimetic culture systems more closely replicate the complex, dynamic conditions of the embryonic environment by providing a 3D, fluid-based context that supports cell–cell and cell–matrix interactions. This supports more accurate tissue development and spatial organization, resembling early stages of amnion development. By contrast, 2D culture systems tend to mimic a more mature structure of the amnion membrane by promoting flattened, layered cell growth, which may better reflect the architecture of foetal membranes. In addition, our study identified TFAP2A and TFAP2B as key transcription factors enriched in the development of AECs, while GATA6, HAND2 and SOX6 were highly expressed in AMCs. The expression patterns of these transcription factors suggest their involvement in regulatory pathways governing amnion development. However, their precise roles remain to be explored in future studies.

In conclusion, our findings provide comprehensive insights into the complex cellular architecture of amnion, highlighting its potential roles in embryonic development and tissue repair. This cellular map serves as a valuable resource for future functional studies on amnion development, in vitro amnion models and potential therapeutic applications.

Methods

Human amnion collection

Human amnion tissue samples were collected from healthy pregnant donors after obtaining informed consent and following institutional ethical guidelines. All procedures were approved by the MRC-Wellcome Trust Human Developmental Biology Resource (HDBR) under ethical approval from the London – Fulham Research Ethics Committee (reference: 08/H0712/34+5, IRAS Project ID: 134561). Sample collection followed HDBR standard operating procedures and documentation, including: Patient Information Sheet and Consent Form, version 16; SOP – Recruitment of Donors, version 8; SOP – Collection of Consented Material, version 7; HDBR Background and Protocol, version 10. Tissue samples were obtained from elective caesarean sections or vaginal deliveries, with no known maternal or foetal complications. Detailed covariate information such as age, genotype or medical history of the donors was not available. When the amnion was collected, the yolk sac was readily identifiable as a distinct vascular sac, separate from the embryo, and could be separated from the amnion. The entire amnion was collected without any specific positional preference.

Tissue processing

All tissues for sequencing were collected in HypoThermosol FRS preservation solution (H4416-100ML Merck) and stored at 4 °C until processing. Tissue dissociation was conducted within 24 h of tissue retrieval.

Tissues were cut into segments of less than 1 mm3 and washed with RPMI 10% FBS 1% penicillin–streptomycin medium before being digested with trypsin–EDTA 0.25% phenol red (25200072, Thermo Fisher Scientific) for 10–15 min at 37 °C with intermittent shaking. The digested tissue was passed through a 100-µm filter and the cells were collected by centrifugation (500g for 5 min at 4 °C). Cells were washed with PBS and resuspended in PBS 0.04% BSA before cell counting. In the case of the CS22 amnion sample, after digestion and washing, a reddish cell pellet was observed and red blood cell lysis buffer (eBioscience, 00-4333-57) was used for optimal lysis of erythrocytes in the single-cell suspension.

10x Genomics Chromium GEX (gene expression) library preparation and sequencing

For the scRNA-seq experiments, cells were loaded according to the manufacturer’s protocol for the Chromium Next GEM Single Cell 5 v2 (dual index) kit for the CS17 amnion and Chromium Next GEM Single Cell 3 v3.1 (dual index) kit for the CS16, CS19 and CS22 amnion from 10x Genomics to attain 7,000 cells per reaction. Library preparation was carried out according to the manufacturer’s protocol. Libraries were sequenced, aiming at a minimum coverage of 20,000 raw reads per cell, on the Illumina HiSeq4000 or Novaseq 6000 systems using the following sequencing format: read 1, 26 cycles; i7 index, 8 cycles, i5 index, 0 cycles; read 2, 98 cycles.

10x Genomics data preprocessing

Cell Ranger software from 10x Genomics was used for data preprocessing. Raw sequencing data were organized, with the requirement that sequencing reads be demultiplexed into FASTQ format files for each sample. The tool ‘cellranger mkfastq’ was used to demultiplex raw base call (BCL) files generated by Illumina sequencers into FASTQ files. Reads were mapped to the human reference genome hg38 and counted with GRCh38-3.0.0 annotation using ‘cellranger count’. The data preprocessing workflow was streamlined and standardized to maintain consistency across samples.

Quality control

Doublets and maternal cells were removed by the Souporcell software, using the default parameter with ‘--clusters=4’; only ‘singlet’ cells were kept for analysis. Souporcell clusters were shown in the UMAP; cells from a single genotype that clustered into the same Seurat cluster were identified as maternal cells and excluded from the analysis.

To eliminate contamination, we used the ‘AddModuleScore’ function in Seurat4 package to assign scores to cells identified as erythrocytes (markers: HBZ, HBE1, HBG2, HBG1, HBA1, HBA2, HBM, ALAS2, HBB, GYPB, GYPC and GYPA), chorion (markers: CGA, CGB3, GCM1, CGB5, CGB7 and CGB8), yolk sac (markers: AFP, CER1, HHEX, FOXA2 and SPINK1) and blood vessels (markers: CD34, PECAM1, CLDN5, CDH5, ESAM, FLT1 and OGN). Only cells with scores <0 were kept for analysis.

We processed scRNA-seq data using a Seurat4. Initially, each dataset underwent quality control, filtering out cells on the basis of gene expression metrics, specifically retaining cells with gene counts (nFeature_RNA) between 500 and 8,000 and mitochondrial gene content (percent.mt) below 20%.

Data integration

We normalized the data using the ‘SCTransform’ method from the Seurat package, using ‘glmGamPoi’ for normalization while regressing out the mitochondrial gene content and cell cycle genes. To mitigate batch effects and integrate data across different sources, we selected 5,000 integration features using Seurat’s ‘SelectIntegrationFeatures’ function. ‘PrepSCTIntegration’ function prepared the datasets for integration, and ‘FindIntegrationAnchors’ was used to identify integration anchors, using SCT normalization. ‘IntegrateData’ function, with SCT normalization, was used to integrate the datasets.

Dimension reduction and clustering

Dimensionality reduction and clustering were performed on the integrated scRNA-seq dataset. Principal component analysis was applied using the ‘RunPCA’ function. We used the ‘RunUMAP’ function to generate a UMAP representation, using the first 40 principal components (dims = 1:40) and specifying ‘umap-learn’ as the method. We constructed a shared nearest-neighbour graph using ‘FindNeighbors’, again focusing on the first 40 dimensions. The ‘FindClusters’ function was applied with a resolution parameter set to 0.4 to detect distinct cell clusters within the data.

Identification of cluster-specific marker genes and cell type annotation

We normalized the data using the ‘NormalizeData’ function and proceeded to scale the data across all genes using the ‘ScaleData’ function. To identify markers for each cluster, we used the ‘FindAllMarkers’ function, focusing only on positive markers (only.pos = TRUE), and setting the minimum percentage of cells expressing the gene (min.pct) at 0.1 and the log fold change threshold (logfc.threshold) at 0.25. The resulting markers were grouped by their respective clusters and sorted to highlight the top markers based on average log fold change.

GO enrichment

The Metascape (http://metascape.org) was utilized for comprehensive gene list annotation and enrichment analysis. GO enrichment analysis was performed across three main categories: biological process (BP), cellular component (CC) and molecular function (MF). For enrichment, we set ‘Min Overlap’ = 3, ‘P Value Cutoff’ = 0.01 and ‘Min Enrichment’ = 1.5.

Trajectory and pseudotime analysis

Trajectory and pseudotime analysis was performed by Monocle336. In brief, a CellDataSet (cds) was prepared from a Seurat object. To integrate UMAP coordinates from Seurat into Monocle3, we extracted the UMAP embedding from cds and aligned it with the Seurat UMAP embedding. We initiated trajectory analysis with ‘learn_graph’ and ordered cells based on UMAP trajectories using ‘order_cells’. For identifying genes associated with cell trajectories, we used ‘graph_test’ on the cds, specifying the ‘neighbor_graph’ as ‘principal_graph’ to highlight genes strongly related to the developmental pathways.

RNA velocity

The BAM file was generated using the default parameters of Cell Ranger. For molecule counting, GRCh38 genome annotations from Cell Ranger’s prebuilt references were utilized, categorizing the molecules into ‘spliced’, ‘unspliced’ and ‘ambiguous’. Repeat annotation files were acquired from the UCSC Genome Browser. AnnData objects (h5ad) were built with ScanPy and ScVelo35. RNA velocity analysis was performed by the scVelo package in Python. In brief, we filtered and normalized the data using ‘scv.pp.filter_and_normalize’, setting a minimum threshold of 30 shared counts and selecting the top 2,000 highly variable genes to focus on the most informative aspects of the dataset. We calculated the neighbourhood moments with ‘scv.pp.moments’, using 30 principal components and 30 nearest neighbours, to capture the local structure and variability within the data. We estimated the RNA velocity using the stochastic model through ‘scv.tl.velocity’. We constructed the velocity graph using ‘scv.tl.velocity_graph’ to visualize.

Cell–cell communication

We used the CellChat38 R package to analyse cell–cell communication networks based on the scRNA-seq dataset. A CellChat object was created by loading the normalized gene expression data into the CellChat environment, followed by assigning cell identities based on their respective metadata. We used the human ligand–receptor interaction database (CellChatDB.human) for our analysis, focusing specifically on ‘Secreted Signaling’ pathways to tailor our investigation towards secretome-mediated interactions. We computed the communication probabilities to infer the cellular communication networks. Communications between cell groups with fewer than ten cells were filtered out.

Diffusion map

Diffusion map analysis was performed using the Destiny package37, considering the 2,000 most variable protein-coding genes. ‘Diffusion map 1’ was set as pseudotime.

Human, monkey and in vitro model data comparison

Human CS7 embryo data were downloaded from Array Express (E-MTAB-9388). Cells annotated as epiblast, amnion, hypoblast and non-neural ectoderm were selected for the analysis. Monkey CS8_CS11 data were obtained from Gene Expression Omnibus (GEO) under accession number GSE193007. Cells annotated as amnion, ectoderm, epiblast, extra-embryonic mesoderm, mesenchyme, surface ectoderm1, surface ectoderm2 and visceral endoderm were selected for the analysis. Stem cell-derived model data were downloaded from GSE179309, GSE205611 and GSE134571. Those datasets were integrated by the Seurat function ‘merge’.

Immunostaining

Human amnion sections were fixed using 4% paraformaldehyde and permeabilized with 0.1% Triton X-100. Non-specific binding was blocked using 10% donkey serum. The sections were incubated with primary antibodies (Supplementary Table 5), followed by washing and incubation with fluorophore-conjugated secondary antibodies. After additional washes, nuclear counterstaining was performed using DAPI. The samples were mounted with anti-fade medium and visualized under a confocal microscope.

Human iPS cell culture

The human iPS cell line (WTC-11) was generously provided by Dr Bruce R. Conklin (Gladstone Institute of Cardiovascular Disease, UCSF). Human naive iPS cells were cultured on Matrigel-coated plates in 4CL medium59 (1:1 mix of Neurobasal medium (Gibco, 21103049) and Advanced DMEM/F12 (Gibco, 12634028) supplemented with N2 (Gibco, 17502048) and B27 (Gibco, 17504044), sodium pyruvate (Corning, 25000CL), non-essential amino acids (Corning, 25025CL), GlutaMAX (Gibco, 35050061), penicillin–streptomycin (HyClone, SV30010), 10 nM DZNep (Selleck, S7120), 5 nM TSA (Vetec, V900931), 1 µM PD0325901 (Axon, 1408), 5 µM IWR-1 (Sigma, I0161), 20 ng ml−1 human LIF (Peprotech, 300-05), 20 ng ml−1 activin A (Peprotech, 120-14E), 50 µg ml−1 l-ascorbic acid (Sigma, A8960) and 0.2% (v/v) Matrigel, Cells in 4CL were cultured at 37 °C, 5% O2 and 5% CO2. To induce specific signalling pathways, we supplemented the medium with 20 ng ml−1 MDK (PeproTech, 450-16), 100 nM RA (STEMCELL Technologies, 72262) and 20 ng ml−1 BMP4(PeproTech, 120-05ET). These cells were maintained in culture and collected for analysis on the fifth day.

Quantitative PCR (qPCR)

Total RNA was extracted using the Direct-zol RNA Purification Kit, Miniprep (Zymo Research, R2051), according to the manufacturer’s protocol. RNA concentration and purity were assessed using a Nanodrop spectrophotometer. For qPCR, 10 ng of RNA per reaction was used with the Luna Universal One-Step RT-qPCR Kit (New England Biolabs, E3005L). Reactions were carried out under the following cycling conditions: reverse transcription at 55 °C for 10 min, initial denaturation at 95 °C for 1 min, followed by 40 cycles of 95 °C for 10 s and 60 °C for 30 s. Relative gene expression was calculated using the ΔΔCt method, with GAPDH as the internal control. Primers were synthesized by Integrated DNA Technologies (IDT). The sequences are presented in Supplementary Table 5.

Statistics and reproducibility

Immunofluorescence staining experiments were repeated independently with consistent results. No statistical method was used to predetermine sample size. For scRNA-seq experiments, the number of samples was determined by tissue availability, which also guided collection and processing. Randomization was not applied during data collection or processing.

Putative maternal cells were excluded on the basis of the expression of maternal markers and genotype mismatches. This exclusion was performed before downstream analyses. Investigators were not blinded to group allocation during experiments or outcome assessment.

Statistical analyses were conducted using GraphPad Prism (v8.2.0). Data distribution was assumed to be normal, although this was not formally tested. For scRNA-seq analyses, differentially expressed marker genes were identified using Seurat (v4.3.0) with a two-sided Wilcoxon rank-sum test. Details regarding sample sizes, statistical tests and P values are provided in the main text, figures, figure legends and supplementary tables.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.