An endometrial immune atlas across the female menstrual cycle reveals lymphoid aggregates compromise pregnancy

This study was approved by the Medical Ethics Committee of Shenzhen Zhongshan Obstetrics & Gynecology Hospital (formerly Shenzhen Zhongshan Urology Hospital), Approval No. SZZSECHU-2020010 and conducted in accordance with the Declaration of Helsinki. Human tissues involved in this study were obtained from five centers: four in China (Shenzhen Zhongshan Obstetrics & Gynecology Hospital, Guangdong Women and Children Hospital, Nanjing Drum Tower Hospital, and The First Affiliated Hospital of Guangxi Medical University) and one in Germany, Universitätsklinikum Düsseldorf (in technical and scientific cooperation with the University Hospital Jena, Germany, local ethical approval “2021-2333-Material”). For healthy controls: Endometrium was obtained from women aged 18–40 years with regular menstrual cycling (3–5 d every 28–30 d) in their natural menstrual cycle (no hormone stimulation) before an endometrial biopsy was performed. Body mass index 18–25 kg/m 2 (inclusive), normal karyotype and negative serological for human immunodeficiency virus, hepatitis B virus, hepatitis C virus and syphilis. Women with the following conditions were excluded from tissue collection: uterine pathology (endometritis, endometriosis, adenomyosis, intrauterine adhesion, uterine scar with diverticulum or longer menstrual bleeding; bacterial, fungal or viral infection), polycystic ovary syndrome, diminished ovarian reserve, history of autoimmune or thyroid-related disease, history of abortions ≥ 2, history of biochemical pregnancies ≥ 3. For CE: Endometrium was obtained from women aged 18–40 years in their natural menstrual cycle (no hormone stimulation) before an endometrial biopsy was performed. Body mass index 18–25 kg/m 2 (inclusive), normal karyotype and a confirmed diagnosis of chronic endometritis based on CD138 immunohistochemistry. Women with the following conditions were excluded from tissue collection: uterine pathology (endometriosis, adenomyosis, intrauterine adhesion, uterine scar with diverticulum or longer menstrual bleeding), polycystic ovary syndrome, diminished ovarian reserve or history of thyroid-related disease. All participants were women (female), and sex was determined based on self‑report and medical records. All participants provided written informed consent prior to inclusion in the study. The menstrual stage of the endometrial samples was precisely determined through a multimodal approach combining: date of last menstrual period, standard histological assessment, transvaginal B-mode ultrasonography, and serum hormone profiling (including estradiol and progesterone levels). Chronic endometritis (CE) was diagnosed through histological identification of CD138⁺ plasma cells in endometrial stroma using anti-CD138 immunohistochemistry (Gene Tech, GT245102 , GR106, 1:150 dilution). For each patient, 30 randomly selected high-power fields (HPFs; 400 × magnification, 0.0625 mm 2 per field) were examined under a light microscope (Nikon, Melville, NY). Based on established criteria 21 , the presence of ≥5 CD138⁺ cells in at least one of the 30 HPFs was considered diagnostic for CE. Cases with <5 CD138⁺ cells in all 30 fields were classified as non-CE. All slides were independently evaluated by two experienced pathologists who were completely blinded to clinical grouping information. Any discrepant diagnoses were resolved through consensus discussion or adjudication by a senior pathologist. Any patient who had received antibiotic therapy within the three months preceding the endometrial sampling was explicitly excluded from both the CE and control cohorts. In the IVF cohort, Endometrial samples were categorized based on IVF outcomes: clinical pregnancy vs. implantation failure, and live birth vs. pregnancy loss. Clinical pregnancy was defined as the detection of fetal heartbeat by transvaginal ultrasound at 6–7 weeks post-transfer; implantation failure was defined as the absence of gestational sac or fetal heartbeat at the corresponding time points. Live birth was defined as delivery of a viable infant at ≥28 weeks of gestation; pregnancy loss included spontaneous abortion or missed abortion after confirmed clinical pregnancy. Fresh endometrial samples were obtained from 53 healthy donors and 34 patients with histologically confirmed CE, along with early pregnancy decidual tissues (decidua basalis and decidua parietalis) from 5 women undergoing elective termination (gestational age: 6–8 weeks). All samples were immediately rinsed in RPMI-1640 medium (Gibco) to remove blood and mucus. After draining excess medium, tissues were minced into small fragments and digested in RPMI-1640 supplemented with 2 mg/mL collagenase type IV (17104-019, Invitrogen) and 0.0125 mg/mL DNase I (D4527, Sigma) for 20 min at 37 °C. The digested suspension was filtered through a 70-μm strainer to obtain a single-cell suspension. Red blood cells were lysed with RBC lysis buffer for 10 min at room temperature, followed by a wash with PBS. Freshly isolated endometrial single-cell suspensions were immediately processed for immune cell sorting. Single-cell suspensions were stained with the following fluorophore-conjugated antibodies for 30 min at 4 °C: Viability Dye-e506 (eBioscience, 65-0866-14, 1:200), anti-CD45-FITC (Biolegend, 304006, 1:200), anti-CD3-BUV805 (BD, 612893, 1:200), anti-CD19-Pacific Blue (Biolegend, 302232, 1:200), anti-CD56-BV605 (Biolegend, 362537, 1:200), anti-CD11c-PE (Biolegend, 371504, 1:200), anti-CD16-BV650 (Biolegend, 302041, 1:200). After staining, cells were washed, centrifuged at 500 g for 5 min at 4 °C, resuspended in FACS buffer (0.5% BSA in PBS), and filtered through a 200-mesh filter before analysis using a BD FACSAria™ Fusion for sorting or downstream analysis. CD45 + immune cells were sorted into 384-well capture plates, each containing 2 μL of lysis solution with barcoded ploy(T) reverse-transcription (RT) primers. Four empty wells (O1, O2, P1, P2) were kept in each 384-well plate as the “no-cell control”. 384-well capture plates with single cells were centrifuged at 500 g for 30 s and stored at −80 °C for the following transcriptome library preparation. Single-cell transcriptome libraries were prepared using the Massively Parallel Single-Cell RNA-seq method (MARS-seq) 37 . In brief, the mRNA from every single cell was barcoded and reverse transcribed into cDNA and pooled together. Subsequently, the pooled single-strand cDNA was converted into double-stranded DNA, which was then linearly amplified into RNA by T7 RNA polymerase-mediated in vitro transcription. The resulting RNA was fragmented and converted into a sequencing-ready library by tagging the samples with pool barcodes and Illumina sequences during ligation, reverse transcription, and PCR. Finally, the quality and concentration of the library were assessed by real-time PCR and Qubit 2.0. All human endometrial tissues or mouse uteri were fixed in 4% PFA overnight at 4 °C, then dehydrated, and embedded in paraffin. Paraffin-embedded tissues were sectioned at 4-μm thickness onto slides. All staining was performed on slides using the PANO 6-plex IHC kit (Panovue, Beijing, China, 10234100100) and Bond Polymer Refine Detection (DS9800-CN), by the automatic dyeing machine (BOND RX, Leica). The primary antibodies applied for human tissues are as follows: anti-CD45 (Biolynx, BX50068, 1:200), anti-CD56 (Gene Tech, GT200529 , 1:100), anti-CD11c (CST, 45581S, 1:200), anti-CD16 (CST, 24326S, 1:300), anti-HLA-G (CST, 79769S, 1:1000), anti-Ki67 (Gene Tech, GT209429 , 1:400), anti-CD20 (Abcam, ab9475, 1:400), anti-CD4 (ZSGB-BIO, ZM-0418, 1:400), anti-CD8 (Novocastra, CD8-4B11-L-CE, 1:400), anti-CD138 (Gene Tech, GT245102 , 1:150), anti-CXCR5 (abcam, ab46218, 1:200), anti-IgG (HUABIO, EM170407 , 1:200), anti-IgA (Abcam, ab193189, 1:200) and anti-IgM (Abcam, ab134159, 1:200). The primary antibodies applied for mouse tissues are as follows: anti-CD45 (Abcam, ab10558, 1:100), anti-CD3 (Abcam, ab16669, 1:200), anti-CD19 (Abcam, ab245235, 1:200) and anti-Ki67 (Abcam, ab16667, 1:400). Nuclei were stained with DAPI after all the antigens of interest had been labeled. Whole tissue sections were scanned in their entirety using an Olympus Slide View VS200 multispectral scanner, and the entire section area was analyzed by HALO software (Version 3.3, Indica Labs). Automated cell segmentation and phenotyping: Acquired multispectral images were imported into HALO software. Cell segmentation was performed automatically using the software’s built‑in nuclear detection algorithm based on DAPI counterstain. Markers including CD4, CD8, CD20, Ki67, and others were used to classify each cell as positive or negative based on manually defined fluorescence intensity thresholds. Using these optimal thresholds, whole‑slide analysis was performed, yielding automated counts and phenotypes of all single cells across the entire section. Quantification of lymphoid aggregates: Lymphoid aggregates were defined as organized clusters of immune cells containing T cells and/or B cells. For inclusion in quantitative analyses, aggregates had to meet both criteria: (1) containing ≥20 contiguous T and/or B cells, and (2) exhibiting an immune cell density of ≥30 cells/3000 μm². The identification and quantification of lymphoid aggregates were performed manually by two independent observers blinded to the experimental groups. Observers visually scanned entire tissue sections and manually selected all qualifying regions. The total number and area of lymphoid aggregates were recorded and normalized to the total analyzed tissue area per sample. Decidual samples were obtained from women undergoing elective termination of first-trimester pregnancies (6-8 weeks gestation). Immediately after collection, tissues were fixed in 4% PFA for 16 h at 4 °C with gentle agitation. Tissues were then dehydrated through a graded ethanol series, cleared in xylene, and embedded in paraffin using standard protocols. Paraffin-embedded tissues were sectioned at 10-μm thickness using a rotary microtome (Leica) and mounted on charged slides. Multiplex immunohistochemistry was performed using the PANO 6-plex IHC kit (Panovue, 10234100100) according to the manufacturer’s instructions. Automated staining was conducted on a BOND RX system (Leica Biosystems) with Bond Polymer Refine Detection (DS9800-CN), including appropriate antigen retrieval steps. Imaging was performed on a Nikon AX. Z stacks and 3D reconstructions were generated using Imaris Imaging software (Version 10.0). Bcl6 fl/fl and CD4 cre mice on the C57BL6J background were kindly provided by Dr. Nengming Xiao and subsequently bred and maintained at the Laboratory Animal Center of Xiamen University. Tfh cell–specific knockout (cKO) mice were generated by Bcl6 fl/fl mice crossed with CD4 cre mice. All animals were housed under specific pathogen-free (SPF) conditions on a 12-h light/dark cycle at 22–24 °C with 50–60% humidity, with ad libitum access to standard chow and water. All animal procedures were conducted in accordance with institutional guidelines and approved by the Animal Welfare Committee of the Research Organization at Xiamen University (XMULAC20250115). Mice were euthanized via CO₂ inhalation for at least 5 min until respiratory arrest was observed. Death was confirmed by toe pinch, and cervical dislocation was performed as a secondary method of euthanasia prior to tissue collection. To synchronize the estrous cycle, soiled bedding from cages of adult male mice (8 weeks, housed for 1 week) was introduced into female (8 weeks) cages 2 days prior to intrauterine injection (Day −2). After 48 h, 50 µL of either phosphate-buffered saline (PBS) or bacterial mixture (3 × 10 9 CFU/ml) containing a precisely formulated 2:1:1 ratio of E.coli (Nanjing Lezhen Biotechnology, BZW23058), S.aureus (Nanjing Lezhen Biotechnology, BZW23512), and S.pyogenes (Nanjing Lezhen Biotechnology, BZW2311E) was injected into both uterine horns of female mice under sterile conditions. Experimental and control animals were housed in separate cages under the same environmental conditions to avoid cross-contamination. Lymphoid aggregates were assessed at predefined time points following infection. To evaluate the effect of lymphoid aggregates on reproductive function, B6 male mice were randomly introduced to female mice from infected and control groups at 28 days post-infection. Mating was confirmed by the presence of a vaginal copulation plug, with the observation day designated as gestational day 0.5 (E0.5). On GD 13.5, pregnant females were humanely euthanized, and uterine horns were dissected to assess pregnancy status. Resorption rates and viable fetal counts were recorded to determine litter size and reproductive success. Lymphoid aggregate scoring and CD138 status were evaluated as two independent, parallel predictors of pregnancy outcomes, without cross-referencing to create a joint classification or treating either as a gold standard. Clinical pregnancy (confirmed by gestational sac on transvaginal ultrasound) and live birth served as the definitive gold-standard endpoints. Classical receiver operating characteristic (ROC) curve analysis was performed, and the area under the curve (AUC) was calculated for each biomarker separately to assess and compare their predictive performance for clinical pregnancy and live birth. MARS-seq libraries, pooled at equimolar concentrations, were sequenced using an Illumina NextSeq 500 sequencer. We used statistics on empty-well spurious UMI detection to ensure that the batches we used for analysis showed a low level of cross single-cell contamination (<3%). MARS-seq reads were processed as previously described 37 , 38 . Reads were mapped to the human reference genome hg38 using HISAT 39 (version 0.1.6); reads with multiple mapping positions were excluded. Reads were associated with genes if they were mapped to an exon, using Homo sapiens Ensembl90 for reference. Exons of different genes that shared genomic position on the same strand were to be considered a single gene with a concatenated gene symbol. To analyze the MARS-seq data from all the samples, we employed the MetaCell package 25 as previously described 40 . First, we excluded cells with fewer than 400 unique molecular identifiers (UMIs), cells exhibiting a total percentage of mitochondrial UMIs over 50%, and cells with a hemoglobin gene count greater than 1. Then, LincRNA, immunoglobulin genes, genes associated with the cell cycle, etc., were forbidden from being feature genes in the MetaCell algorithm (Supplementary Data  9 ). Gene features with high variance to mean were selected using the variation-to-mean parameter Tvm > 0.08. From the gene features, we excluded forbidden genes. The feature gene selection strategy described above retained a total of 1747 feature genes from 128898 immune cells for the computation of the metacells (MCs) and was used for clustering. Finally, we annotated the MCs based on their log enrichment scores (log footprint [lfp] values). We annotated immune cells using canonical cell markers (Fig.  S1C ). We examined the batch effect by comparing the multiple plates from the same sample processed in the same batch or across different batches and cells from the same population (Fig.  S1B ). The results confirmed that our method did not exhibit a discernible batch effect, as demonstrated in previous publications using the MARS-seq platform 40 , 41 . The same workflow was applied to sub-clustering each main lineage. The 2D projection is computed using a force-directed layout algorithm on a balanced similarity graph. By applying a force-directed layout algorithm to the metacell graph, the coordinates for each metacell are generated, and the single cells are positioned by averaging the metacell coordinates of their filtered neighbor cells in the original balanced graph 42 . To understand the differential composition of immune populations, we compared the frequencies of main lineages and subclusters over the menstrual cycle (Fig.  2 ) or between difference groups (Fig.  3 ). Meanwhile, the ratio of observed cell numbers to random expectation calculated by chi-square test (RO/E) was used to adjust cell-sampling biases for each patient (Figs.  3 D–F, 3 K, L, 3O, P , S 3 A, S 3 D–G, S 3J ). RO/E > 1 (above the dashed line) indicates enrichment. Differential gene expression analyses between different groups were performed using the ‘Findmarkers’ function from the Seurat R package. To account for the cell number variations between groups, the same number of cells was randomly sampled from each group. Hotspot (v0.9.1) was used to identify functional gene modules that illustrate heterogeneity within NK subsets. Briefly, we used the “danb” model and selected the top 1000 genes with the highest autocorrelation z-score for module identification. Modules were then identified using the “create_modules” function with the parameters: min_gene_threshold = 15 and fdr_threshold = 0.05. Module scores were calculated by using the “calculate_module_scores” function. Pearson similarity coefficients were used to evaluate the transcriptional similarity between modules. Data were presented as mean (±SEM) of more than three independent experiments. The Wilcoxon test was used to compare differences between two groups of samples. The Kruskal-Wallis test was used to compare multiple groups if the grouping variable contained more than two levels. All statistical analyses were performed using R software version 4.1.1. Further information on research design is available in the  Nature Portfolio Reporting Summary linked to this article.

To characterize the endometrial immune landscape across the natural menstrual cycle and define alterations associated with chronic endometritis (CE), we conducted scRNA-seq on CD45 + immune cells from endometrial biopsies of 53 healthy controls and 34 CE patients. After quality control, we obtained 147,468 immune cells spanning 8 phases, including menstrual phase (endometrial tissue shed during menstruation, day1–3 of menstrual cycle), early-, mid-, and late-proliferative phase (day4–7, day8–11, day12–14 of menstrual cycle), early-, mid-, and late-secretory phase (day15–18, day19–22, day23–25 of menstrual cycle); premenstrual phase (day 26–28 of menstrual cycle), covering the entire menstrual cycle in humans (see “Methods”, Figs.  1a , S 1a–c ; Supplementary Data  1 and Supplementary Data  2 ). Fig. 1 Single-cell immune landscape of the human endometrium. a Schematic overview of the study design (Created in BioRender. Zhang, Y. (2026) https://BioRender.com/llv2e84 ): Endometrial biopsies were obtained from healthy donors and patients with chronic endometritis (CE). CD45 + immune cells were isolated for MARS-seq-based single-cell RNA sequencing and downstream bioinformatic analysis. Flow cytometry and multiplex immunofluorescence (mIF) were used to validate the immune features in both healthy and CE endometrium. A CE mouse model and multi-center clinical cohort were further employed for mechanistic investigation, clinical validation, and translational application. b Two-dimensional (2D) projection of 147,468 single endometrial immune cells annotated as six main lineages distinguished. c Normalized expression of selected marker genes across the metacell model for the main lineages. Each bar represents one metacell, colored as in panel ( b ). d Subclustering of T cells, NK/ILC, myeloid cells (Mono./MAC/DC) and B cells. A total of 38 lineages were annotated and marked by a color code. Source data are provided as a Source Data file. a Schematic overview of the study design (Created in BioRender. Zhang, Y. (2026) https://BioRender.com/llv2e84 ): Endometrial biopsies were obtained from healthy donors and patients with chronic endometritis (CE). CD45 + immune cells were isolated for MARS-seq-based single-cell RNA sequencing and downstream bioinformatic analysis. Flow cytometry and multiplex immunofluorescence (mIF) were used to validate the immune features in both healthy and CE endometrium. A CE mouse model and multi-center clinical cohort were further employed for mechanistic investigation, clinical validation, and translational application. b Two-dimensional (2D) projection of 147,468 single endometrial immune cells annotated as six main lineages distinguished. c Normalized expression of selected marker genes across the metacell model for the main lineages. Each bar represents one metacell, colored as in panel ( b ). d Subclustering of T cells, NK/ILC, myeloid cells (Mono./MAC/DC) and B cells. A total of 38 lineages were annotated and marked by a color code. Source data are provided as a Source Data file. We used the MetaCell algorithm 25 , 26 to identify homogeneous and robust groups of cells (metacells) from scRNA-seq data, resulting in a detailed map of 334 metacells categorized into eight broad lineages, including NK cells, T cells, monocytes, macrophages, and dendritic cells (collectively referred to as Mono./MAC/DC), B cells, innate lymphoid cells (ILC) and mast cells (Fig.  1b, c ). We further subdivided NK/ILC, T cells, Mono./MAC/DC, and B cells into several subtypes or cellular states, resulting in 38 subtypes/states that represent the transcriptional diversity of the human endometrial immune microenvironment during the menstrual cycle (Figs.  1d , S 1d ; Supplementary Data  3 and Supplementary Data  4 ). Endometrial immune cell populations exhibited distinct temporal dynamics across the menstrual cycle in 53 healthy women (Fig.  2a–c ). NK and T cells were the dominant subsets, together comprising approximately 80% of all endometrial immune cells. Throughout the cycle, NK cell proportions demonstrated a progressive increase, while T cell proportions displayed a concomitant decrease (Fig.  2b ). In contrast, the monocyte, macrophage, and dendritic cell (Mono./Mac/DC) compartment showed significant expansion specifically during the secretory phase. Minor populations, including B cells (~5% of immune cells), innate lymphoid cells (ILCs) and mast cells, exhibited distinct and more variable fluctuation patterns (Fig.  2b ). Fig. 2 Temporal dynamics of endometrial immune cells across the menstrual cycle. a Spearman correlation of immune cell composition across eight menstrual phases in 53 healthy women. Sample sizes are indicated in the legend and apply to a , b , d – f and the corresponding Supplementary Figures. Two-sided Wilcoxon tests. Center lines indicate median; boxes, interquartile range (IQR); whiskers, 1.5 × IQR (applies to a , b , d – f and the corresponding Supplementary Figures ). b Compositional changes of major immune lineages across the menstrual cycle. Left, relative proportions; right, individual lineage frequencies. Each data point represents one participant, Kruskal–Wallis tests (applies to b , d – f and the corresponding Supplementary Figures ). c Compositional changes of NK/ILC, T, myeloid, and B subtypes across the menstrual cycle in 53 healthy women. d – f Quantification of subtype frequency changes across the menstrual cycle. g Temporal dynamics of NK subsets and CD11c + NK-to-CD103 + NK ratio. Top: scRNA-seq analyzed with LOESS regression. Bottom: multiplex immunofluorescence (mIF) validation in endometrial and decidual sections stained for CD45, CD56, CD11c, and CD16 ( n  = 5 per phase). Decidual tissues (decidua parietalis (decP) and decidua basalis (decB)) were collected from elective terminations at 5–7 weeks of gestation. Data are representative of three independent experiments. h Heatmap showing mean expressing levels for HLA-E, HLA-G and HLA-C receptors in NK subsets. i Spatial localization of NK subsets and extravillous trophoblasts (EVTs) in first-trimester decidua. Decidual tissues were collected from elective terminations at 5–7 weeks of gestation ( n  = 5). Regions were categorized as HLA-G⁺⁺, HLA-G⁺, or HLA-G − based on EVT density, outlined with white dashed circles. Quantification shows the density (cells/mm²) of CD11c⁺ NK, CD103⁺ NK and their ratio in each region (mean ± SEM). Data are representative of three independent experiments. j Representative 2D and 3D orthogonal reconstructions of CD11c⁺ NK–HLA-G⁺ EVT interactions in first-trimester decB. All images represent single z‑slice captures. White dashed lines outline interacting cells; arrows indicate contact sites. Bar graph showing the percentage of CD11c⁺ NK and CD103 + NK cells interacting with EVTs ( n  = 5). Two-tailed Mann–Whitney test (mean ± SEM). Data are representative of three independent experiments. k Heatmap of hotspot -derived gene module (GM) scores across menstrual cycle stages. Data were analyzed at the cell level. Source data are provided as a Source Data file. a Spearman correlation of immune cell composition across eight menstrual phases in 53 healthy women. Sample sizes are indicated in the legend and apply to a , b , d – f and the corresponding Supplementary Figures. Two-sided Wilcoxon tests. Center lines indicate median; boxes, interquartile range (IQR); whiskers, 1.5 × IQR (applies to a , b , d – f and the corresponding Supplementary Figures ). b Compositional changes of major immune lineages across the menstrual cycle. Left, relative proportions; right, individual lineage frequencies. Each data point represents one participant, Kruskal–Wallis tests (applies to b , d – f and the corresponding Supplementary Figures ). c Compositional changes of NK/ILC, T, myeloid, and B subtypes across the menstrual cycle in 53 healthy women. d – f Quantification of subtype frequency changes across the menstrual cycle. g Temporal dynamics of NK subsets and CD11c + NK-to-CD103 + NK ratio. Top: scRNA-seq analyzed with LOESS regression. Bottom: multiplex immunofluorescence (mIF) validation in endometrial and decidual sections stained for CD45, CD56, CD11c, and CD16 ( n  = 5 per phase). Decidual tissues (decidua parietalis (decP) and decidua basalis (decB)) were collected from elective terminations at 5–7 weeks of gestation. Data are representative of three independent experiments. h Heatmap showing mean expressing levels for HLA-E, HLA-G and HLA-C receptors in NK subsets. i Spatial localization of NK subsets and extravillous trophoblasts (EVTs) in first-trimester decidua. Decidual tissues were collected from elective terminations at 5–7 weeks of gestation ( n  = 5). Regions were categorized as HLA-G⁺⁺, HLA-G⁺, or HLA-G − based on EVT density, outlined with white dashed circles. Quantification shows the density (cells/mm²) of CD11c⁺ NK, CD103⁺ NK and their ratio in each region (mean ± SEM). Data are representative of three independent experiments. j Representative 2D and 3D orthogonal reconstructions of CD11c⁺ NK–HLA-G⁺ EVT interactions in first-trimester decB. All images represent single z‑slice captures. White dashed lines outline interacting cells; arrows indicate contact sites. Bar graph showing the percentage of CD11c⁺ NK and CD103 + NK cells interacting with EVTs ( n  = 5). Two-tailed Mann–Whitney test (mean ± SEM). Data are representative of three independent experiments. k Heatmap of hotspot -derived gene module (GM) scores across menstrual cycle stages. Data were analyzed at the cell level. Source data are provided as a Source Data file. As the dominant immune populations in the endometrium, NK/ILC cells can be further categorized into six distinct subtypes (Fig.  S2a ): CD11c + NK, CD103 + NK, CD16 + NK, ISG15 + NK, ILC2, and ILC3. Among these, CD11c + NK and CD103 + NK cells constituted the majority of the NK lineage (Figs.  1d and 2c ). Due to its extremely low abundance, the ISG15 + NK subset was excluded from further analysis. Using flow cytometry and multiplex immunofluorescence (mIF) with antibodies against CD45, CD56, CD11c, and CD16, we successfully validated the three primary NK subsets: CD11c + NK, CD103 + NK, and CD16 + NK (Fig.  S2b, c ). Analysis across the menstrual cycle revealed distinct temporal dynamics: The proportion of CD103 + NK cells increased during the proliferative phase but declined sharply in the secretory phase. Conversely, CD11c + NK cells remained stable during proliferation but increased dramatically during the secretory phase. The other NK/ILC subsets (CD16 + NK, ILC2, ILC3) maintained relatively low and stable proportions throughout the cycle (Figs.  2d and S 2d ). The 13 T cell subtypes exhibited distinct dynamic changes throughout the menstrual cycle (Figs.  2e and S2e). Among them, CCL5 + CD8 + T cells represented the predominant T cell population in the human endometrium, with their proliferating subset showing a significant increase during the secretory phase (Fig.  2e ). The CD4 + T cell compartment was primarily composed of memory phenotype cells (TM CD4), the proportion of which declined in the secretory phase compared to the proliferative phase (Fig.  2e ). In contrast, effector CD4 + T cell subsets, including CD4 + T follicular helper-like (Tfh-like) cells and Th17-like cells, remained at consistently low and stable levels across the menstrual cycle in healthy women (Fig.  2e ). Among B cells, naïve and memory B cells constituted the majority (~95%), while germinal center B-like (GC B-like) cells, CD83 + B cells, and plasma cells were present in minor proportions (Fig.  2f ). Plasma cells, though scarce, were predominantly detected in the proliferative phase, whereas other B cell subsets maintained stable frequencies throughout the menstrual cycle (Fig.  2f ). In addition, innate immune cells including monocytes, macrophages, and dendritic cells exhibited high heterogeneity and displayed phase-specific compositional changes (Fig.  S2f ). To further verify the temporal dynamics of the three primary NK subsets, we performed in situ quantitative analysis. CD16 + NK cells maintained relatively low and stable proportions throughout the cycle. In contrast, we observed a progressive decline in CD103 + NK cell frequency from the proliferative phase through the secretory phase and into pregnancy, while CD11c + NK cells showed a continuous increase during these stages (Fig.  2g ). These opposing dynamic patterns suggest distinct functional specializations between these NK cell subsets. CD16⁺ NK cells were transcriptionally similar to peripheral CD56 dim NK cells (Fig.  S2g ), whereas CD103⁺ NK cells were enriched for cytokines, including CCL5 and IL-16 (Fig.  S2h ), indicating a potential immunoregulatory role. In contrast, CD11c⁺ NK cells expressed receptors for HLA-G, HLA-C, and HLA-E (LILRB1, KIRs, KLRCs), (Fig.  2h ), ligands presented by extravillous trophoblasts (EVTs), a specialized placental cell population that invades maternal decidua, remodels spiral arteries, and mediates maternal immune tolerance during embryo implantation 27 , 28 . This receptor-ligand pairing indicates functional adaptation of CD11c⁺ NK cells for EVT recognition. In consistency, multiplex immunofluorescence on early pregnancy decidua (parietalis and basalis) revealed selective enrichment of CD11c⁺ NK cells in HLA-G⁺ EVT-rich regions (Fig.  2i ). High-resolution 2D and 3D imaging further confirmed direct cell–cell contacts between CD11c⁺ NK cells and HLA-G⁺ EVTs in the decidua basalis (Fig.  2j ). Analysis of NK cell gene modules throughout the menstrual cycle also revealed: (1) progressive downregulation of CD103 + NK-associated modules, and (2) upregulation of CD11c + NK-enriched modules during the secretory phase, particularly genes involved in EVT interaction (LILRB1, KIRs, ENTPD1) (Figs.  2k , S2i , and Supplementary Data  5 ). These gene expression shifts align with the observed temporal changes in subset abundance and further support a functional transition of the endometrial NK cell compartment in preparation for implantation. By constructing a high-resolution immune cell atlas of the human endometrium across the menstrual cycle, we identified CD11c + NK cells as a specialized functional subset that exhibits dynamic accumulation during the secretory phase and pregnancy, and demonstrated their unique ability to physically interact with EVTs. With the immune cell atlas of the menstrual cycle in healthy women as a reference, we were able to characterize endometrial immune alterations in 34 CE patients. Comparative analysis revealed profound reorganization of the endometrial immune landscape in CE patients (Fig.  3a–c ), marked by significant expansions of CD3⁺ T, CD19⁺ B and CPA3⁺ mast cells, concurrent with reduction of NK cells and ILCs (Fig.  3d ). Notably, these disruptions were consistent across both proliferative and secretory phases, suggesting persistent pathological changes independent of menstrual stages (Figs.  3e and S 3a ). Fig. 3 Perturbation of immune homeostasis in the endometrium of CE patients. a Reference map and 2D density plots of endometrial immune cells in healthy controls and CE patients. b Two-dimensional (2D) scatter plots with kernel density estimation (KDE) illustrating the distribution differences of major immune lineages between healthy and CE groups. Statistical significance was assessed using a two-sided Kolmogorov–Smirnov test (applies to a , g , k and the corresponding Supplementary Figures ). c Radar plots showing the fold changes of major immune lineages in CE relative to healthy controls across the entire cohort, proliferative and secretory phase. d – f Quantification of immune cell composition in healthy controls and CE patients: d Relative abundance of major immune lineages across the entire cohort (healthy controls: n  = 47, CE patients: n  = 34); e Relative abundance of NK, T, and B cells in the proliferative (healthy controls: n  = 19, CE patients: n  = 19) and secretory (healthy controls: n  = 28, CE patients: n  = 15) phases; f Relative abundance of CD11c⁺ NK and CD103⁺ NK cells across the entire cohort (healthy controls: n  = 47, CE patients: n  = 34). The ratio of observed cell numbers to random expectation calculated by chi-square test (R O/E ) was used to adjust cell-sampling biases for each patient; values are shown as log 2 (RO/E); log 2 (RO/E + 1) > 1 indicates enrichment; error bars: mean ± SEM; two-sided Wilcoxon test (applies to d-f, i-j, l-m and the corresponding Supplementary Figures ). g , h Reference map, 2D density plots and KDE plots of B cells in healthy donors and CE patients. i , j Quantification of Germinal Center-B-like (GC-B-like), plasma cells across the entire cohort (i, healthy controls: n  = 47, CE patients: n  = 34) and separately during the proliferative (healthy controls: n  = 19, CE patients: n  = 19) and secretory (healthy controls: n  = 28, CE patients: n  = 15) phase ( j ). k Reference map, 2D density plots and KDE plots of T cells in healthy and CE groups. l , m Quantification of T follicular helper-like (Tfh-like) and Th17-like cells across the entire cohort ( l , healthy controls: n  = 47, CE patients: n  = 34), and separately during the proliferative (healthy controls: n  = 19, CE patients: n  = 19) and secretory (healthy controls: n  = 28, CE patients: n  = 15) phases ( m ). Source data are provided as a Source Data file. a Reference map and 2D density plots of endometrial immune cells in healthy controls and CE patients. b Two-dimensional (2D) scatter plots with kernel density estimation (KDE) illustrating the distribution differences of major immune lineages between healthy and CE groups. Statistical significance was assessed using a two-sided Kolmogorov–Smirnov test (applies to a , g , k and the corresponding Supplementary Figures ). c Radar plots showing the fold changes of major immune lineages in CE relative to healthy controls across the entire cohort, proliferative and secretory phase. d – f Quantification of immune cell composition in healthy controls and CE patients: d Relative abundance of major immune lineages across the entire cohort (healthy controls: n  = 47, CE patients: n  = 34); e Relative abundance of NK, T, and B cells in the proliferative (healthy controls: n  = 19, CE patients: n  = 19) and secretory (healthy controls: n  = 28, CE patients: n  = 15) phases; f Relative abundance of CD11c⁺ NK and CD103⁺ NK cells across the entire cohort (healthy controls: n  = 47, CE patients: n  = 34). The ratio of observed cell numbers to random expectation calculated by chi-square test (R O/E ) was used to adjust cell-sampling biases for each patient; values are shown as log 2 (RO/E); log 2 (RO/E + 1) > 1 indicates enrichment; error bars: mean ± SEM; two-sided Wilcoxon test (applies to d-f, i-j, l-m and the corresponding Supplementary Figures ). g , h Reference map, 2D density plots and KDE plots of B cells in healthy donors and CE patients. i , j Quantification of Germinal Center-B-like (GC-B-like), plasma cells across the entire cohort (i, healthy controls: n  = 47, CE patients: n  = 34) and separately during the proliferative (healthy controls: n  = 19, CE patients: n  = 19) and secretory (healthy controls: n  = 28, CE patients: n  = 15) phase ( j ). k Reference map, 2D density plots and KDE plots of T cells in healthy and CE groups. l , m Quantification of T follicular helper-like (Tfh-like) and Th17-like cells across the entire cohort ( l , healthy controls: n  = 47, CE patients: n  = 34), and separately during the proliferative (healthy controls: n  = 19, CE patients: n  = 19) and secretory (healthy controls: n  = 28, CE patients: n  = 15) phases ( m ). Source data are provided as a Source Data file. In-depth profiling identified three major immune perturbations. First, CD11c⁺ NK cells, implicated in EVT interaction, were consistently and significantly reduced across both phases in CE patients compared to healthy controls (Figs.  3f and S 3b–g ). Second, the B cell compartment showed extensive remodeling, with robust expansion of GC B-like cells, plasma cells, and memory B cells (Figs.  3g, h ; 3i, j and S 3h ). Third, the T cell landscape exhibited dramatic reorganization (Fig.  3k ), most notably in Tfh-like cells, which showed the greatest increase at both phases, accompanied by elevated Th17-like cells, TM CD4, CCL5 + CD8 and ISG15 + CD8 cell populations in CE patients (Figs.  3l-m and S 3i ). In addition, myeloid subsets exhibited distinct compositional alterations in CE patients compared to healthy individuals (Fig.  S3j–l ). The coordinated expansion of proinflammatory lymphocyte subsets, particularly GC B-like cells, plasma cells, Tfh-like cells, and Th17-like populations, demonstrates marked activation of adaptive immunity in the CE endometrium. When combined with the concomitant depletion of CD11c + NK cells and dysregulation of myeloid compartments, these findings reveal a fundamental breakdown of endometrial immune homeostasis in CE patients. Among all immune lineages, B cells exhibited the most pronounced alterations in the CE endometrium. Multiplex flow cytometry confirmed a significant increase in CD19 + B cells frequency during both proliferative and secretory phases compared to healthy controls (Figs.  4a, b and S4a ). Multiplex immunofluorescence further revealed a marked elevation in absolute B cell numbers in CE patients compared to controls (2.95 vs . 19.29 cells/mm 2 ). CD4 + and CD8 + T cell numbers also increased substantially (35.03 vs . 199.1 cells/mm 2 ; 78.16 vs . 114.0 cells/mm 2 ; Fig.  4c, d ). Notably, T and B cells exhibited an aggregate state rather than being dispersed in the endometrium of CE patients (Fig.  4c ). Fig. 4 Proliferation and differentiation of B cells in the endometrium of CE patients. a A representative FACS plot of CD19 + B cells in endometrial samples. b Frequency of CD19 + B cells in proliferative and secretory endometrium from healthy donors (proliferative, n  = 7; secretory, n  = 7) and CE patients (proliferative, n  = 6; secretory, n  = 6). Two-tailed Mann–Whitney test; mean + SD. c Representative images of CD20 + B cells, CD4 + T cells, and CD8 + T cells in the proliferative endometrium from healthy donors and CE patients. Dotted lines delineate areas of T and B cell aggregation. d Quantification of cell densities (cells/mm 2 ) for each subset ( n  = 5 per group). Each data point represents one participant; center lines indicate median; boxes, interquartile range (IQR); whiskers extend to the minimum and maximum values, with outliers falling beyond; two-tailed Mann–Whitney (applies to d , f , g ). e The Metacell projection showing proliferating B cells. Circle size represents the proportion of proliferating cells within each metacell, and colors indicate B cell subsets as in Fig.  1d . f Representative images and quantification of proliferating B cells in the proliferative endometrium ( n  = 5 per group). Arrowheads indicate Ki67 + CD20 + cells; dotted lines outline aggregation zones. Results are shown as both cell density (cells/mm²) and the percentage within CD20 + B cells. g Representative images and quantification of CD138 + plasma cells in the proliferative endometrium from healthy donors and CE patients ( n  = 5 per group). h A UMAP projection of endometrial plasma cells from healthy donors and CE patients. Pi chart showing the distribution of plasma cells between the two groups. i Expression of immunoglobulin isotype genes (IgG, IgA, IgM) projected onto the plasma cell metacell map of CE samples. Pi chart showing the relative abundance of each isotype. j Representative images of IgG, IgA and IgM in CE endometrium. Pi chart showing the relative abundance of each isotype. All validation data are representative of three independent experiments. Source data are provided as a Source Data file. a A representative FACS plot of CD19 + B cells in endometrial samples. b Frequency of CD19 + B cells in proliferative and secretory endometrium from healthy donors (proliferative, n  = 7; secretory, n  = 7) and CE patients (proliferative, n  = 6; secretory, n  = 6). Two-tailed Mann–Whitney test; mean + SD. c Representative images of CD20 + B cells, CD4 + T cells, and CD8 + T cells in the proliferative endometrium from healthy donors and CE patients. Dotted lines delineate areas of T and B cell aggregation. d Quantification of cell densities (cells/mm 2 ) for each subset ( n  = 5 per group). Each data point represents one participant; center lines indicate median; boxes, interquartile range (IQR); whiskers extend to the minimum and maximum values, with outliers falling beyond; two-tailed Mann–Whitney (applies to d , f , g ). e The Metacell projection showing proliferating B cells. Circle size represents the proportion of proliferating cells within each metacell, and colors indicate B cell subsets as in Fig.  1d . f Representative images and quantification of proliferating B cells in the proliferative endometrium ( n  = 5 per group). Arrowheads indicate Ki67 + CD20 + cells; dotted lines outline aggregation zones. Results are shown as both cell density (cells/mm²) and the percentage within CD20 + B cells. g Representative images and quantification of CD138 + plasma cells in the proliferative endometrium from healthy donors and CE patients ( n  = 5 per group). h A UMAP projection of endometrial plasma cells from healthy donors and CE patients. Pi chart showing the distribution of plasma cells between the two groups. i Expression of immunoglobulin isotype genes (IgG, IgA, IgM) projected onto the plasma cell metacell map of CE samples. Pi chart showing the relative abundance of each isotype. j Representative images of IgG, IgA and IgM in CE endometrium. Pi chart showing the relative abundance of each isotype. All validation data are representative of three independent experiments. Source data are provided as a Source Data file. The expansion of B cells in CE patients can be attributed to local proliferation or recruitment from the periphery, or both. Proliferation analysis revealed a substantial fraction of actively cycling B cells, particularly within GC B-like cell subsets (Fig.  4e ). Furthermore, spatial analyses demonstrated that proliferating (Ki67 + ) B cells were significantly upregulated and these cells account for about 24.04% of all B cells in CE patients (Fig.  4f ). Notably, Ki67 + B cells were preferentially localized within B cell aggregates (Fig.  4f ). These data indicated that the increase of B cells in CE patients is due to, at least partially, the local proliferation and differentiation of GC B-like cells. Consistent with our scRNA-seq findings, immunostaining analysis also confirmed a significant increase of CD138 + plasma cells in CE patients (Fig.  4g, h ). These plasma cells showed predominant expression of IgG heavy chains (IGHG) compared to other isotypes (IGHA, IGHE, IGHD, IGHM) (Figs.  4i and S 4b ). This immunoglobulin class-switching pattern was validated by multiplex immunofluorescence, which demonstrated that IgG + plasma cells, indicative of mature class-switch recombination, constituted the dominant plasma cell population in CE patients (Fig.  4j ). Together, our findings demonstrate an enhanced B cell differentiation cascade in CE patients, progressing from naïve B cells through proliferating and germinal center B cell stages to ultimately IgG-secreting plasma cells. The predominance of class-switched IgG + plasma cells, coupled with significant Tfh-like cell expansion, strongly suggests enhanced Tfh-dependent B cell activation in CE pathogenesis. In consistency with our scRNA-seq data, multiplex immunostaining analysis further validated that the number of CXCR5 + CD4 + Tfh-like cells significantly increased in CE patients (Fig.  5a ). We next correlated lineage-normalized cell-type frequencies in CE patients to investigate intercellular interactions. Tfh-like cells showed high correlation with plasma cells, GC B-like and memory B cells (Fig.  5b ), suggesting cell-cell interactions among these subsets. Tfh-like cells expressed classical feature genes of Tfh cells, including BCL6, CXCR5, CXCL13, PDCD1, ICOS, IL21 and TOX2 (Fig.  5c ). Notably, multiple ligand-receptor pairs associated with the differentiation and recruitment of B cells were highly expressed between Tfh-like cells and B subsets, including CXCL13-CXCR5, IL21-IL21R, CD40LG-CD40 and IL6ST-EBI3 (Fig.  5d ). The coordinated upregulation of these molecular pathways strongly suggested the functional coupling between Tfh-like cells and B cell subsets in CE pathogenesis. Fig. 5 Tfh-dependent B cell response and lymphoid aggregate formation in CE patients. a Representative immunofluorescence images and quantification of T follicular helper (Tfh) cells in proliferative endometrium from healthy donors and CE patients ( n  = 7 per group). Dotted lines outline CD4 + T-cell aggregates. Each data point represents one participant; center lines indicate median; boxes, interquartile range (IQR); whiskers extend to the minimum and maximum values, with outliers falling beyond; two-tailed Mann–Whitney (applies to a , g , h ). Data are representative of three independent experiments. b Spearman correlation matrix displaying pairwise correlations among lineage-normalized immune cell frequencies in CE patients. c Expression of representative Tfh-associated marker genes across T subsets. d Heatmap of ligand–receptor interactions between T subsets and B subsets. e , f Representative immunofluorescence images of proliferative endometrium from CE patients stained for anti-CD20, anti-CD4, anti-CD8, anti-Ki67, anti-CD138, and anti-CXCR5. Dotted lines delineate B and T cell zones, as well as germinal center-like regions within aggregates. Magnified views in e correspond to the region indicated in Supplementary Fig.  5a . g , h Quantification of lymphoid aggregates in healthy controls and CE patients during the proliferative (healthy, n  = 18; CE, n  = 11) and secretory ( n  = 5 per group) phases. Data are representative of three independent experiments. Source data are provided as a Source Data file. a Representative immunofluorescence images and quantification of T follicular helper (Tfh) cells in proliferative endometrium from healthy donors and CE patients ( n  = 7 per group). Dotted lines outline CD4 + T-cell aggregates. Each data point represents one participant; center lines indicate median; boxes, interquartile range (IQR); whiskers extend to the minimum and maximum values, with outliers falling beyond; two-tailed Mann–Whitney (applies to a , g , h ). Data are representative of three independent experiments. b Spearman correlation matrix displaying pairwise correlations among lineage-normalized immune cell frequencies in CE patients. c Expression of representative Tfh-associated marker genes across T subsets. d Heatmap of ligand–receptor interactions between T subsets and B subsets. e , f Representative immunofluorescence images of proliferative endometrium from CE patients stained for anti-CD20, anti-CD4, anti-CD8, anti-Ki67, anti-CD138, and anti-CXCR5. Dotted lines delineate B and T cell zones, as well as germinal center-like regions within aggregates. Magnified views in e correspond to the region indicated in Supplementary Fig.  5a . g , h Quantification of lymphoid aggregates in healthy controls and CE patients during the proliferative (healthy, n  = 18; CE, n  = 11) and secretory ( n  = 5 per group) phases. Data are representative of three independent experiments. Source data are provided as a Source Data file. Our findings of expanded Tfh-like cells, GC B-like cells, and IgG + plasma cells in CE patients, along with their spatial organization into lymphoid aggregates, suggested the formation of tertiary lymphoid structures (TLS) in the endometrium with CE. TLS are ectopic lymphoid formations that develop in non-lymphoid tissues during chronic inflammation, such as in tumors and autoimmune diseases 29 – 31 . We therefore performed multiplex immunostaining on endometrial tissues from CE patients and confirmed the presence of TLS-like lymphoid aggregates composed of Tfh-like cells, Ki67 + proliferating B cells and plasma cells (Fig.  5e, f ). Quantitative analysis demonstrated a significant increase in the frequency and area of lymphoid aggregates in CE patients versus healthy individuals in proliferative and secretory phases (Figs.  5g, h and S5a, b ). These data demonstrated enhanced Tfh-dependent B cell responses and the formation of lymphoid aggregates in CE patients. Chronic endometritis is predominantly caused by bacterial infection in clinical practice. This is supported by the well-established efficacy of antibiotic therapy, with treatment achieving clinical resolution in approximately 90% of CE cases. To further investigate the functional consequence of lymphoid aggregates on pregnancy outcome, we developed a CE mouse model through intrauterine infusion of clinically relevant bacterial strains (Fig.  6a ). This mouse model recapitulated the immunopathological features of CE patients, showing increased formation of TLS-like lymphoid aggregates in the infected mouse uterus compared to sham controls at 28 days post-infection (Fig.  6b, c ). Fig. 6 Lymphoid aggregates are associated with compromised pregnancy outcomes in a mouse CE model. a An experimental timeline of the CE mouse model and pregnancy outcome assessment (Created in BioRender. Zhang, Y. (2026) https://BioRender.com/llv2e84) . b , c Representative images and quantification of lymphoid aggregates in wild-type mice (WT), Bcl6 fl/fl ; CD4-Cre conditional knockout mice, and corresponding sham mice ( n  = 8 per group). Dotted lines outline organized lymphoid aggregates. Two-tailed Mann–Whitney test; mean + SEM. d Representative images of dissected uteri and fetuses at gestational day (gd) 13.5 from the CE model and control groups. e Quantification of litter size and fetal resorption rate in WT-sham ( n  = 7), cKO-sham ( n  = 8), WT-model ( n  = 6), and cKO-model ( n  = 7) mice. Two-tailed Mann–Whitney test, mean + SEM. Data are representative of three independent experiments. Source data are provided as a Source Data file. a An experimental timeline of the CE mouse model and pregnancy outcome assessment (Created in BioRender. Zhang, Y. (2026) https://BioRender.com/llv2e84) . b , c Representative images and quantification of lymphoid aggregates in wild-type mice (WT), Bcl6 fl/fl ; CD4-Cre conditional knockout mice, and corresponding sham mice ( n  = 8 per group). Dotted lines outline organized lymphoid aggregates. Two-tailed Mann–Whitney test; mean + SEM. d Representative images of dissected uteri and fetuses at gestational day (gd) 13.5 from the CE model and control groups. e Quantification of litter size and fetal resorption rate in WT-sham ( n  = 7), cKO-sham ( n  = 8), WT-model ( n  = 6), and cKO-model ( n  = 7) mice. Two-tailed Mann–Whitney test, mean + SEM. Data are representative of three independent experiments. Source data are provided as a Source Data file. We then mated both the CE model and the sham control female mice with healthy male mice. Bacterial-induced CE resulted in a reduced number of live fetuses and total fetuses, and severe fetal lethality with a median fetal resorption rate reaching 60%, 15-fold of that observed in the sham controls (Fig.  6d, e ), suggesting the association of lymphoid aggregates with adverse pregnancy outcomes. Next, we sought to examine whether the compromised pregnancy outcomes in CE mice can be rescued by depleting lymphoid aggregates. Our findings in CE patients suggest a critical role of Tfh-like cells in lymphoid aggregate formation. Since BCL6 is the master transcriptional regulator of Tfh cell development 32 , 33 , we thus employed the Bcl6 fl/fl ;CD4-cre mouse strain, in which Tfh cells can be genetically depleted. Indeed, mice deficient of Tfh cells showed significantly reduced lymphocyte aggregate formations under CE conditions (Fig.  6b ). Importantly, Tfh cell depletion partially rescued pregnancy outcomes, with fetal resorption rates decreasing from 60% to 20% and viable fetuses increasing by two-fold (Fig.  6c ). Collectively, in this mouse model, the genetic depletion of Tfh cells reduced lymphoid aggregate formation and partially rescued pregnancy outcomes, providing experimental evidence for a causal contribution of Tfh-dependent lymphoid aggregates to compromised establishment and maintenance of pregnancy in mice. To further examine whether lymphoid aggregates are associated with pregnancy outcomes in humans, we took advantage of the clinical practice of in vitro fertilization (IVF). During the IVF process, endometrial biopsies were collected for clinical examination before embryo transfer, allowing us to investigate the association between endometrial immune features and subsequent pregnancy outcomes (Fig.  7a ). We thus designed a retrospective multicenter clinical study, collaborating with five IVF centers across the world (Shenzhen, Nanjing, Guangzhou, Nanning and Düsseldorf), with 229 patients included (Figs.  7a , S6a, b ; Supplementary Data  6 and Supplementary Data  7 ). Fig. 7 Endometrial lymphoid aggregates are associated with compromised pregnancy outcomes in human. a Study design (Created in BioRender. Zhang, Y. (2026) https://BioRender.com/llv2e84) . A total of 229 infertile patients undergoing endometrial biopsy during the proliferative phase before IVF-ET were included based on predefined criteria (see “Methods” and Supplementary Data  6 ). Biopsy specimens were simultaneously assessed for both lymphoid aggregates abundance and CD138 + plasma cells by immunofluorescence. Histopathological assessment was independently performed by two blinded pathologists, with discrepancies resolved by a third senior pathologist. Finally, patients were stratified according to clinical pregnancy outcomes and live birth results, enabling comparative analysis of the correlation between lymphoid aggregates abundance, CD138 status, and reproductive outcomes. b Quantification of endometrial lymphoid aggregates abundance in relation to pregnancy outcomes. Each data point represents one patient. Two-tailed unpaired t-test, mean ± SD. c Individual-level data of three patients highlighted in panel B (green box), who exhibited more lymphoid aggregates, achieved embryo implantation, but failed to achieve live birth. d Receiver operating characteristic (ROC) curves evaluating the predictive performance of lymphoid aggregates abundance for clinical pregnancy and live birth. Area under the curve (AUC) values are indicated. e Density of CD138⁺cells in patients with successful or failed embryo implantation and live birth (Two-tailed unpaired t-test, mean ± SD). ROC curves evaluating the predictive performance of CD138 + plasma cells for clinical pregnancy and live birth. Source data are provided as a Source Data file. a Study design (Created in BioRender. Zhang, Y. (2026) https://BioRender.com/llv2e84) . A total of 229 infertile patients undergoing endometrial biopsy during the proliferative phase before IVF-ET were included based on predefined criteria (see “Methods” and Supplementary Data  6 ). Biopsy specimens were simultaneously assessed for both lymphoid aggregates abundance and CD138 + plasma cells by immunofluorescence. Histopathological assessment was independently performed by two blinded pathologists, with discrepancies resolved by a third senior pathologist. Finally, patients were stratified according to clinical pregnancy outcomes and live birth results, enabling comparative analysis of the correlation between lymphoid aggregates abundance, CD138 status, and reproductive outcomes. b Quantification of endometrial lymphoid aggregates abundance in relation to pregnancy outcomes. Each data point represents one patient. Two-tailed unpaired t-test, mean ± SD. c Individual-level data of three patients highlighted in panel B (green box), who exhibited more lymphoid aggregates, achieved embryo implantation, but failed to achieve live birth. d Receiver operating characteristic (ROC) curves evaluating the predictive performance of lymphoid aggregates abundance for clinical pregnancy and live birth. Area under the curve (AUC) values are indicated. e Density of CD138⁺cells in patients with successful or failed embryo implantation and live birth (Two-tailed unpaired t-test, mean ± SD). ROC curves evaluating the predictive performance of CD138 + plasma cells for clinical pregnancy and live birth. Source data are provided as a Source Data file. Endometrial samples were categorized based on IVF outcomes: clinical pregnancy vs. implantation failure, and live birth vs. pregnancy loss. We then compared the presence and extent of endometrial lymphoid aggregates between these outcome groups. Our analysis revealed a strong inverse correlation between the frequency and area of lymphoid aggregates and IVF success. Women who experienced implantation failure or did not achieve live birth showed significantly higher numbers and larger areas of lymphoid aggregates compared to those with successful outcomes (Fig.  7b ). This trend was consistently observed across all five participating centers (Fig.  S6c ), supporting the robustness of our findings. Notably, while several patients exhibiting enriched endometrial lymphoid aggregates achieved initial embryo implantation, they nevertheless experienced subsequent pregnancy loss (Fig.  7b, c ). Besides, univariable and multivariable logistic regression analyses showed that after adjusting for common potential confounders, including female age, female BMI, female hormonal levels, the number, quality, and other characteristics of the transplanted embryos, endometrial lymphocyte aggregates remained an independent predictor of pregnancy outcome (Supplementary Data  8 ). Furthermore, we concurrently evaluated CD138 + plasma cell levels in endometrial biopsy tissues, and receiver Operating Characteristic (ROC) analyses showed that the lymphocyte aggregate level was a better predictor biomarker for adverse pregnancy outcome than CD138 + plasma cell number, which is generally considered a risk factor for pregnancy failure (Fig.  7d, e ). In summary, these human data demonstrate a strong association between endometrial lymphoid aggregates and adverse pregnancy outcomes. While our mouse model suggests a potential causal role for Tfh-dependent lymphoid aggregates in pregnancy compromise, the translational relevance of these mechanistic findings to human disease requires further prospective validation.

This study establishes a high-resolution immune cell atlas of 53 healthy endometria across eight precisely timed menstrual stages. Our analysis reveals a progressive expansion of CD11c⁺ NK cells, which we identified as interacting with extravillous trophoblasts (EVTs), from the proliferative phase through the secretory phase and into early pregnancy. At the transcriptomic level, these cells exhibit profound similarity to canonical dNK1 cells 34 (Fig.  S7a ), highlighting their pivotal role in orchestrating the uterine environment for successful implantation. This atlas provides a comprehensive reference map of endometrial immune dynamics across the menstrual cycle in health, serving as an essential baseline for investigating immune-related disorders, including chronic endometritis (CE), endometriosis, endometrial cancers and others. The exceptional rarity of longitudinal samples spanning all eight phases of the menstrual cycle elevates this resource as a foundational tool for reproductive immunology research in the future. This atlas is publicly accessible for the community via our website ( http://119.8.233.169:5010/ ). Clinical observations from the assisted reproduction field suggest CE is associated with adverse pregnancy outcome, but the precise immune characteristics of CE remain poorly defined 22 , 35 . Integrating scRNA-seq, multiplex immunofluorescence, and flow cytometry on CE patient endometrial samples, we identified three hallmark perturbations: 1) significant reduction of CD11c + NK cells; 2) hyperactivated Tfh-driven germinal center reactions; and 3) formation of lymphoid aggregates. Notably, diagnostic criteria for CE vary across institutions, with some centers still requiring complete CD138 negativity for non-CE. Our healthy control cohort used a threshold of <5 CD138⁺ cells/HPFs. To address potential confounding from low-level CD138 infiltration, we performed a restricted analysis using only strictly CD138-negative samples. All key immunological findings remained robust, confirming that the expansion of GC-like B cells, Tfh-like cells, and plasma cells, alongside the reduction of CD11c⁺ NK cells, represents a genuine pathological signature of CE (Fig.  S7b, c ). These results suggest that chronic endometrial inflammation is not characterized solely by the presence of plasma cells (traditionally defined by CD138 immunostaining 36 ), but rather reflects a coordinated immune reorganization involving local germinal center reaction and reduction of EVT-interacting CD11c + NK cells. Although CE is diagnosed by the presence of endometrial plasma cells in the clinics, the source of plasma cells remains elusive. Our results demonstrate that CE-associated plasma cells primarily originate from local germinal center reactions rather than peripheral circulation. Notably, we found that the presence of endometrial lymphoid aggregates was strongly associated with adverse pregnancy outcomes, with a predictive value superior to that of plasma cell counts. This is likely due to the temporal delay in the generation or resolution of plasma cells during chronic inflammation, while endometrial lymphoid aggregates are more of a real-time indicator for chronic inflammatory processes. Therefore, lymphoid aggregates might be a better diagnostic marker for CE than plasma cells, yet large-scale randomized controlled trials (RCTs) are required before their clinical application. The longstanding lack of a suitable murine model has impeded CE research. Leveraging clinical insights from our data, we developed an experimental mouse model that recapitulates several key features of human CE. In this model, we have established a link between Tfh-dependent lymphoid aggregates and impaired pregnancy outcomes. It thereby serves as an experimental platform for the mechanistic dissection of the cellular and molecular pathways driving CE-associated reproductive failure in the future. However, it is essential to distinguish the levels of evidence presented in this study. Our human IVF cohort analyses establish a robust associative relationship between endometrial lymphoid aggregates and adverse reproductive outcomes, with lymphoid aggregates serving as an independent predictor after adjustment for established clinical confounders. Causal claims regarding the mechanistic necessity of lymphoid aggregates for compromised pregnancy are explicitly limited to our BCL6-deficient mouse model, which, while informative, differs from human endometrial biology in fundamental respects, including the absence of menstruation, distinct uterine immune cell composition, and divergent decidualization and implantation mechanisms. Thus, while the mouse model offers valuable proof-of-concept for the Tfh-dependent pathogenic mechanism, direct extrapolation to human disease should be made with caution, and prospective clinical studies are warranted to validate the translational relevance of these findings. Our findings indicate that Tfh-dependent lymphoid aggregates may compromise pregnancy outcomes through multiple potential biological pathways. (1) Local Autoimmunity: The GC-like B cells within the aggregates may produce antibodies targeting endometrial or trophoblast antigens, potentially disrupting implantation and placental development. (2) T Cell-Mediated Inflammation: The Tfh-like cells likely sustain a pro-inflammatory microenvironment (e.g., via IL-21, CXCL13) that could be harmful to the semi-allogeneic embryo, impairing trophoblast invasion and survival. (3) Breach of Immune Tolerance: The formation of lymphoid aggregates itself may cause a fundamental breakdown in the unique immune tolerance required for embryo acceptance. 4) Reduction of CD11c + NK cells: Lymphoid aggregates may also lead to pregnancy loss by inducing the reduction of EVT-interacting CD11c + NK cells, which are critical for the invasion of EVTs during embryo implantation and growth. 5) Altered Tissue Remodeling: The persistent inflammatory niche may dysregulate critical processes for decidualization and vascular remodeling, essential for a successful pregnancy. 6) Structural barrier: These lymphoid aggregates may physically constrain embryo implantation by occupying spatial niches normally invaded by EVTs, creating a structural barrier to proper placentation. These potential mechanisms may act independently or synergistically to contribute to reproductive failure. Further studies are warranted to investigate the exact cellular and molecular mechanisms by which lymphoid aggregates drive pregnancy failure. Collectively, our study provides a temporally resolved single-cell immune landscape of human endometrium across the whole menstrual cycle. Using CE as a clinical model, we show that disruption of this cyclical immune homeostasis is linked to adverse pregnancy outcomes. We identify Tfh-dependent lymphoid aggregates and depletion of CD11c⁺ NK cells as potential contributors to both implantation failure and early pregnancy loss, offering mechanistic insight into CE-related immune dysregulation. We further propose lymphoid aggregates as a robust biomarker for clinical risk stratification. Together, our findings redefine CE as a T–B cell–driven immunopathology and highlight lymphoid aggregates as a clinically applicable biomarker of reproductive failure.

The human endometrium is a specialized mucosal tissue with unique immune properties, which not only protects against microbial challenges but also promotes embryo implantation and growth 1 – 3 . Its immune landscape is highly dynamic, undergoing cyclical remodeling throughout the menstrual cycle 4 . In each cycle, the endometrium needs to prepare an immune-privileged niche to support the implantation of the allograft embryo in the secretory phase, known as the “implantation window“ 5 – 7 . Such functional complexities and temporal dynamics have posed a great challenge for the field to investigate the characteristics and functions of the endometrial immune microenvironment in humans. Recent scRNA-seq studies have begun to unravel the intricate cellular heterogeneity of the human endometrium 8 – 10 . However, a comprehensive atlas of immune cell dynamics across the entire menstrual cycle remains unavailable. Certain immune cell types, such as NK cells, are known to expand during the secretory phase 11 , contribute to fetal development 12 , 13 and antimicrobial defense 14 . Nevertheless, whether this is a functional transition of NK cells and other endometrial immune cells during the menstrual cycle in preparation for embryo implantation and growth remains an open question. Endometrial immune perturbations are implicated in diverse reproductive disorders 15 – 19 . However, inadequate characterization of the physiological endometrial immune landscape across the menstrual cycle fundamentally limits our understanding of its pathological alterations in reproductive disorders and the underlying immunological mechanisms. Furthermore, elucidating the relationship between immune dysregulation and pregnancy outcomes remains critical for improving clinical management. Chronic endometritis (CE), characterized histologically by the presence of plasma cells within the endometrium, is a chronic inflammatory condition linked to adverse pregnancy outcomes 20 – 22 . However, little is known about the exact immunological alterations that occur in CE and how they affect pregnancy outcomes 23 , 24 . In this study, we use single-cell profiling to establish a reference immune atlas of the cycling human endometrium, leveraging this framework to dissect the immunological basis of chronic endometritis. We demonstrate that lymphoid aggregates compromise pregnancy outcomes and serve as a prognostic indicator of pregnancy failure, validated functionally in a murine model and clinically in a prospective multicenter cohort. Our work provides a resource for understanding endometrial immunity and offers a mechanistic rationale for targeted intervention in immune-mediated reproductive disorders.

Source data Source data