We collected 8 proliferative endometrial samples from women with normal pregnancies, 6 proliferative endometrial samples from patients with recurrent miscarriages as defined elsewhere [35, 36], and 5 mid secretory-phase endometrial samples, each from women with normal pregnancies and patients with recurrent miscarriages. In addition, we collected decidua and villus samples (6–8 weeks (w)) from 9 women with normal pregnancies, and 7 patients with recurrent miscarriages, at the Department of Gynaecology and Obstetrics, Union Hospital, Huazhong University of Science and Technology, from 2021 to 2022. After signing the informed consent with the patient, the endometrium of different periods was collected, cleaned with normal saline three times and sterile gauze, stored in 1.5 mL EP tube and marked, then stored in the –80 °C refrigerator for follow-up experiments

GSE111976 single cell RNA-seq of human endometrium [37] was acquired from National Center for Biotechnology Information (NCBI)’s Gene Expression Omnibus (GEO) [38]. The data were filtered by the percentage of mitochondria. Nonlinear dimensionality reduction was performed using the t-Distribution Stochastic Neighbour Embedding (t-SNE), the FindAllMarkers function was used to identify significant clusters [39], and cell types were annotated based on specific biomarkers [40, 41], DCN, UCA1, FOXJ1, PTPRC, RNASE1, ACTA2, and LYZ was respectively one of biomarkers of Stromal fibroblast, Unciliated Epithelium, Ciliated cell, Lymphocyte, Endothelium, Smooth muscle cell and Macrophage.

We analyzed ciliated cells independently by the subset function and performed a statistical analysis of the proportions of ciliated cells, unciliated epithelium, and stromal fibroblasts according to the staging of the menstrual cycle, and the count function in Rstudio Version 4.2.1 was used to calculate the ratio of the number of cells to the total number of cells. In addition, we found that 2 of the 10 samples had large differences in cell numbers, and we removed the samples numbered 58 and 41. We performed a staging analysis according to the number of days in the menstrual cycle and counted the proportion of ciliated cells by using Rstudio and the function of count.

Differentially expressed genes (DEGs) were identified by EdgeR packages Version 3.38.4 (https://bioconductor.org/packages/release/bioc/html/edgeR.html), which works on a table of integer read counts, the counts represent the total number of reads aligning to each gene. EdgeR was concerned with differential expression analysis and relative changes in expression levels between conditions rather than with the quantification of expression levels and not directly with estimating absolute expression levels. The differential expression threshold was set as a p value 0.05. In volcano plots (VlnPlot), up-regulated DEGs, down-regulated DEGs, and stable DEGs were marked separately with ‘red’, ‘blue’, and ‘grey’. Then, Gene Set Enrichment Analysis (GSEA) [42] was used to perform gene ontology (GO) [43, 44] and Kyoto Encyclopedia of Genes and Genomes (KEGG) [45] pathway analyses. GO is an internationally standardized classification system of gene functions that provides a dynamically updated, standardized vocabulary to comprehensively describe the properties of genes and gene products in an organism. KEGG is a database that systematically analyzes gene functions, genomic information and functional information, including metabolic pathways database, hierarchical classification database, gene database, and genome database. Gseaplot2 graphs of the main functions of the proliferative, early secretory, early-mid secretory, mid secretory, and late secretory phases were constructed. CellPhoneDB [46] is a publicly available repository of curated receptors, ligands and their interactions, and subunit architectures are included for both ligands and receptors, accurately representing heteromeric complexes. We used this function to find key ligands.

Circular cnetplots and gseaplot2 graphs of significant biological processes (BPs) of the early and mid secretory stages were constructed by using Rstudio Version 4.2.1. Then, PPI networks [47] of the early secretory phase and mid secretory phase were structured using search tool for recurring instances of neighbouring genes (STRING Version 11.0 (https://string-db.org/)) [48]. A combined score $>$0.4 (medium confidence), DEGs selected with $|$logFC$|$ $>$0.5, and the most significant modules were identified by Cytoscape software Version 3.7.1 (https://cytoscape.org/) [49] and Molecular Complex Detection (MCODE Version 1.6.1 (https://apps.cytoscape.org/apps/mcode)) [50].

All tissues were fixed in 4% paraformaldehyde, deparaffinized after sectioning, and incubated with 1% peroxidase to block endogenous peroxidase activity. Endogenous peroxidase activity was blocked by soaking in 1% H${}_2$O${}_2$ in PBS for 30 minutes. Antigen retrieval was achieved through incubation in a microwave oven at 100 ${}^{\circ }$C for 30 minutes. After blocking with 3% bovine serum albumin (BSA, ST025-20g, Beyotime, Shanghai, China) for 1 hour at room temperature, sections were incubated with anti-human GJA1 (dilution 1:1000; Proteintech, 26980-1-AP, Wuhan, Hubei, China) and SPP1 rabbit monoclonal antibodies (dilution 1:1000; Proteintech, KHC0782, Wuhan, Hubei, China) overnight at 4 ${}^{\circ }$C. After rinsing three times with phosphate buffered saline (PBS), sections were incubated with the secondary antibody (dilution 1:5000; Proteintech, PR30009, Wuhan, Hubei, China) for 1 hour and then stained with 3,30-diaminobenzidine tetrahydrochloride (DAB, P0203-1/2, Beyotime, Shanghai, China).

The data were filtered and described in descending order of their characteristics. A violin diagram demonstrates the features, count, and percent of RNA (Fig. 1A) of each sample. The t-SNE diagram demonstrates nonlinear dimensionality reduction clustering (Fig. 1B). Then, after annotating the cells with markers specific to each cell type (Fig. 1C), an annotated clustered t-SNE plot (Fig. 1D) was constructed. Lastly, an heatmap showing the expression signature of the top 10 marker genes for each 7 cell type was created (Fig. 1E).

Fig. 1.

Basic data processing. Violin Diagram (VlnPlot) (A). Integrated distribution t-distributed stochastic neighbor embedding (t-SNE) plot of 10 sample data in GSE111976 dataset, with each color-coded to indicate different samples (B). Dot plot showing average expression for marker genes in each cell types (C). Annotated clustered t-SNE plots for the different marker genes (D). Heatmap showing the expression signature of the top 10 marker genes for each 7 cell type (E).

We annotated t-SNE clustering plots for 10 samples of single-cell sequencing data (Fig. 2A) and determined the proportions of ciliated cells, unciliated epithelium, and stromal fibroblasts according to the staging. The proportion of ciliated cells peaked at the early to mid secretory phases (Fig. 2B). After removing 2 samples (58 and 41) with large differences in cell counts and performing new staging of remaining samples according to the days of the menstrual cycle, ciliated cell ratios were plotted, and found to peak in the mid secretory phase, suggesting that ciliated cells may play a key role in the mid secretory phase (Fig. 2C).

Fig. 2.

The stage and ciliary cell proportion of the sample. t-SNE plots for 10 samples (A). After analyzing the cell type annotation, we got seven cell types per sample, and calculated the proportions of ciliated cells, unciliated epithelium, and stromal fibroblasts according to the original staging (B). The proportion of ciliated cells according to the new staging after removing samples 58 and 41 (C).

Volcano plots and GSEA GO bubble diagrams were generated by comparing ciliated cells to stromal fibroblasts mainly in the early mid secretory phase to the early secretory phase (Fig. 3A) and the mid secretory phase to the early-mid secretory phase (Fig. 3B). A gseaplot2 of meaningful BP was drawn. Compared to the proliferative phase, in the early secretory phase the fertilization, and sperm-egg recognition processes were downregulated, while ciliated cell assembly and epithelial cilium movement, which was involved in extracellular fluid movement, were upregulated in ciliated cells. In contrast, ciliated cell assembly and cilium movement in fluid were downregulated in stromal fibroblasts (Fig. 4A). In the early mid secretory phase, in contrast to the early secretory phase, embryo implantation and decidualization processes were upregulated, cilium assembly and epithelial cilium movement involved in extracellular fluid movement were downregulated in ciliated cells. In stromal fibroblasts, embryo implantation, decidualization, female pregnancy, cilium movement, and sperm-egg recognition were upregulated (Fig. 4B). In the mid- secretory phase, in contrast to the early mid secretory phase, embryo implantation, decidualization, pregnancy, and embryonic morphogenesis were upregulated, while epithelial cilium movement involved in extracellular fluid movement was downregulated in ciliated cells. All processes were downregulated in stromal fibroblasts (Fig. 4C). In the late secretory phase, in contrast to the mid secretory phase, the function of epithelial cilium movement involved in extracellular fluid movement, cilium movement, and embryonic appendage morphogenesis in ciliated cells were all upregulated, while embryonic processes involved in female pregnancy and protein localization to the cilium were downregulated in stromal fibroblasts (Fig. 4D). CellPhoneDB analysis of ligand/receptor expression revealed interactions between unciliated epithelium and ciliated epithelium (Fig. 5), and NOTCH1_JAG1, CD46_JAG1, and MIF_TNFRSF14 may be the most important ligands/receptors.

Fig. 3.

Functional analysis of differential genes. Volcano plots, gene set enrichment analysis (GSEA) gene ontology (GO) bubble diagrams of early mid secretory phase compared to early secretory phase (A), and mid secretory phase compared to early mid secretory phase (B). log2 Fold Chang $>$1 and p 0.05 was statistically significant, up-regulated genes were log2 Fold chang $>$1 and p 0.05, while down-regulated genes were log2 Fold Chang –1 and p 0.05.

Fig. 4.

Functional analysis of ciliated cells and stromal fibroblasts at different periods. Gseaplot2 plots of early secretory phase compared to proliferative phase (A). Early mid secretory phase compared to early secretory phase (B). Mid secretory phase compared to early-mid secretory phase (C). Late secretory phase compared to mid secretory phase in both ciliated cells and stromal fibroblasts (D).

Fig. 5.

CellphoneDB analysis of ligands/receptors between unciliated epithelium and ciliated epithelium. The darker the red, the more significant the up-regulation.

We performed a functional analysis of ciliated cells and stromal fibroblasts and constructed cnetplots and gseaplot2 graphs of the pregnancy, embryo implantation, and decidualization. In comparative analyses of the early mid secretory phase to the early secretory phase (Fig. 6A) and the mid secretory phase to the early mid secretory phase (Fig. 6B), we found that the three BPs were all upregulated. This was followed by mapping the PPI networks and MCODEs for the three biological processes to identify the hub genes, with all seed genes being marked with yellow (Fig. 7). According to our preliminary data, we found that GJA1, SPP1 were common gene of pregnancy, embryo implantation, and decidualization, NOTCH1_JAG1 might be the main communication ligand (Fig. 5). There’s a protein interaction between the GJA1, SPP1, NOTCH1 and JAG1, so they might play key roles in pregnancy.

Fig. 6.

Major biological processes and their involvement in genes. Cnetplot and gseaplot2 of pregnancy, embryo implantation, and decidualization in early mid secretory phase compared to early secretory phase (A). Mid secretory phase compared to early-mid secretory phase in ciliated cells and stromal fibroblasts (B).

Fig. 7.

Common DEGs identified by PPI networks and molecular complex detection (MCODEs). DEGs, differentially expressed genes; PPI, protein–protein interaction; GJA1, connexin 43; SPP1, secreted phosphoprotein 1; CTSB, Cathepsin B; STC1, Stanniocalcin‑1; IDO1, Indoleamine 2,3-dioxygenase 1; TIMP1, Tissue inhibitor of metalloproteinases-1; VEGFA, Vascular endothelial growth factor A; ADM, Acinar-to-ductal metaplasia; JAG1, Jagged-1; CD46, Membrane cofactor protein; NOTCH1, Neurogenic locus notch homolog protein 1; AGO2, Argonaute-2; TMED2, Transmembrane emp24 domain-containing protein 2.

The results of immunohistochemistry showed that the staining intensity of GJA1 and SPP1 were significantly higher in the proliferative endometrium (Fig. 8A) and fetal villus (Fig. 8B) in normal pregnancy as compared to that of recurrent miscarriage, but no significant difference in decidua was noted (Fig. 8C).

Fig. 8.

Immunohistochemistry of proliferating endometrium (A), villus (B), and decidua (C) between normal and recurrent miscarriage. *** represented p 0.001, **** represented p 0.0001. GJA1, connexin 43; SPP1, secreted phosphoprotein 1; ns, no significance.