Stalling of the endometrial decidual reaction determines the recurrence risk of miscarriage

Abstract

In every menstrual cycle, progesterone acting on estrogen-primed endometrium elicits an inflammatory decidual reaction, rendering it poised for embryo implantation and transformation into the decidua of pregnancy. Here, we show that the sequential functions of the decidual reaction-implantation and decidualization-pivot on the time-sensitive loss of progesterone-resistant DIO2⁺ stromal cells that form a specialized implantation niche and reciprocal expansion of progesterone-dependent PLA2G2A⁺ predecidual cells. Simultaneously, uterine natural killer (uNK) cell proliferation results in the accumulation of immunotolerant subsets. Examination of endometrial biopsies from 924 women revealed that the recurrence risk of miscarriage closely aligns with the incidence of a weakened or stalled decidual reaction, more so than poor uNK cell expansion. Analysis of paired biopsies obtained in different cycles and modeling in assembloids intimated that prior miscarriages disrupt intercycle endometrial homeostasis and calibration of the decidual reaction. Our findings show that erosion of the decidual reaction following a miscarriage drives the recurrence risk irrespective of maternal age.

Fig. 1.. DIO2⁺ and PLA2G2A⁺ mark functionally distinct stromal subsets.

(A) Depiction of endometrial sampling in six subjects across two menstrual cycles. For each sample (n = 12), the ratio of SCARA5 and DIO2 transcripts measured by reverse transcription quantitative polymerase chain reaction was normalized for the day of the biopsy relative to the LH surge; samples were deemed progesterone resistant [SCARA5/DIO2 < 25th percentile (25th %ile), highlighted in red] or progesterone responsive (SCARA5/DIO2 ≥ 25th percentile). (B) Volcano plot showing gene expression variance between progesterone-resistant and progesterone-responsive endometrial biopsies. Transcripts of interest are highlighted. FDR, false discovery rate. (C) Representative images of DIO2 and PLA2G2A transcripts in midluteal endometrium by chromogenic (left) and fluorescent (right) smISH. LE, luminal epithelium. (D) Uniform manifold approximation and projection (UMAP) projections of scRNA-seq data showing the cellular composition of luteal phase endometrium (n = 12). The color key in the inset depicts days post-LH surge; asterisks denote sample numbers for each day. (E) Feature plots showing relative expression of DIO2 and PLA2G2A in different cell types. The color scale represents log-transformed average gene expression. (F) RNA velocity stream mapped onto a UMAP plot of stromal subsets from a midluteal biopsy. (G) GSEA enrichment plot. The dot size corresponds to the enrichment score; the color key denotes Benjamini-Hochberg–adjusted P values. TNFα, tumor necrosis factor–α; NF-κB, nuclear factor κB; UV, ultraviolet. (H) Heatmap of normalized TE-derived transcript levels encompassing DNA transposons (DNA), long interspersed nuclear elements (LINEs), long terminal repeats retrotransposons (LTRs), and short interspersed nuclear elements (SINEs). (I) Heatmap of z score–scaled genes in source, DIO2⁺, PLA2G2A⁺, and other stromal cells.

Fig. 2.. Spatiotemporal dynamics of uNK cells.

(A) Representative image of CD56 (neural cell adhesion molecule 1) and pan-KIR immunofluorescence in midluteal endometrium. CD56 and pan-KIR immunoreactivity in different regions is shown separately in boxes i to iii. The insert shows a region at higher magnification. Scale bars, 50 μm. Nuclei are stained with 4′,6-diamidino-2-phenylindole (DAPI). (B) Flow cytometry density plot showing different uNK subsets based on KIR and CD39 expression. (C) Bar graph depicting the temporal changes in the relative abundance of uNK subsets in 55 endometrial samples, as determined by flow cytometry. (D) Schematic of the experimental design. The uNK subsets were purified by FACS using pan-KIR and CD39 antibodies. (E) Bar graphs showing median fold change (FC) in SA-β-GAL activity in 10 biological repeat experiments. Individual data points are also shown. Different letters indicate statistically significant differences at P < 0.01, Friedman test with Dunn’s multiple comparisons test. (F) UMAP projections of scRNA-seq data showing lymphoid populations and uNK subsets. (G) Bar graphs depicting the temporal changes in the relative abundance of uNK subsets, as determined by scRNA-seq. (H) Violin plots (left) and feature plots (right) of ITGAD and CD160 expression with P values based on Wilcoxon rank sum test; a.u., arbitrary units. (I) Relative expression of stromal and uNK subset marker genes and their ratios in 779 endometrial biopsies obtained 6 to 10 days after the LH surge. The solid line indicates median expression; the dotted lines mark the boundaries of the upper and lower quartiles. Ratios were fitted against a gamma or a log-normal distribution depending on LH + day. (J) Top: Spearman’s correlation between ITGAD/CD160 and PLA2G2A/DIO2 percentiles. Bottom: Box plots comparing ITGAD/CD160 percentiles and PLA2G2A/DIO2 percentiles grouped in quartile bins. Different letters above the whiskers indicate significance between groups at P < 0.05, Kruskal-Wallis test with Dunn’s multiple comparisons test.

Fig. 3.. Regulation of stromal and uNK subsets.

(A) Schematic representation of the spatial analysis of normalized PLA2G2A/DIO2 and ITGAD/CD160 ratios (percentiles) in endometrial biopsies. (B and D) Representative analysis of two independent samples with normalized gene ratios measured across sequential (~0.5 cm) regions. (C and E) Spearman’s correlation of normalized gene ratios at the opposite ends of the samples (A/Z regions) from 264 subjects. (F) Spearman’s correlation between normalized PLA2G2A/DIO2 or ITGAD/CD160 ratios (percentiles) and circulating progesterone, estradiol, and TSH levels normalized to the day of sampling post-LH surge (percentiles). Paired endometrial and peripheral blood samples were obtained from 315 subjects. (G) Schematic of experimental design. Freshly isolated endometrial samples were divided, and randomly selected sections were incubated for 3 hours in additive-free media (AFM) supplemented or not with 50 pM T3 before gene expression analysis using digital droplet polymerase chain reaction (ddPCR). (H) Bar graphs showing median FC in gene expression/gene ratios of 11 and 10 biological repeat experiments for stromal and uNK subsets, respectively. Individual data points are also shown; P values are based on the Wilcoxon matched-pairs signed-rank test. (I) Heatmap showing the temporal changes in the number of predicted receptor-ligand interactions between uNK, stromal, or epithelial subsets and trophectoderm cells from day 6/7 human blastocysts. (J) Schematic of the experimental design. Freshly isolated endometrial samples were divided, and sections were incubated for 3 hours in pooled spent medium, diluted 1:1 in AFM, of IVF blastocysts that resulted in a clinical pregnancy (P) or not [not pregnant (NP)]. Pooled embryo-free droplets served as the control (C) group. (K) Bar graphs showing median FC in gene expression/gene ratios of five biological repeat experiments. Individual data points are also shown; P values are based on the Friedman test with Dunn’s multiple comparisons test.

Fig. 4.. The recurrence risk of miscarriage.

(A and B) Dot plots (left) showing the frequency of endometrial samples (percentage) stratified by the number of previous pregnancy losses for (A) PLA2G2A/DIO2 percentile quartile bins and (B) ITGAD/CD160 percentile quartile bins. The number of subjects in each clinical group is also shown. The frequency of samples in the lowest and highest quartile bins is also enumerated in bar graphs (right) for (A) normalized stromal subset ratios and (B) normalized uNK subset ratios. Different letters above the bars indicate significance between groups at P < 0.05, chi-square test with Bonferroni correction. (C) Median number of prior miscarriages in subjects with a stalled decidual reaction (normalized stromal subset ratios < 25th percentile), poor uNK cells expansion (normalized uNK subset ratios < 25th percentile), or both. Statistical analysis is based on one-way analysis of variance (ANOVA) and Tukey’s multiple comparisons test. (D) Schematic of the cohort. (E) Forest plots showing ORs and 95% CIs for miscarriage risk and live birth rates following assessment in a nonconception cycle of the decidual reaction (PLA2G2A/DIO2 percentile) and uNK cell expansion (ITGAD/CD160 percentile). Subjects were grouped in quartile bins. Data are shown for all subjects (left) and upon exclusion of confirmed aneuploid pregnancy losses (right); n.s., not significant.

Fig. 5.. Analysis of intercycle variability.

(A) Schematic of paired endometrial sample collection. (B) Contingency tables showing the distribution (percentage), grouped in quartile bins, of normalized stromal and uNK subset ratios (PLA2G2A/DIO2 percentile and ITGAD/CD160 percentile, respectively) in B samples for each A sample quartile bin. The number (n) of paired biopsies in each A sample quartile bin is shown. The colored squares in the contingency tables indicate statistical significance (P < 0.05), as determined by Fisher’s exact test for enriched (red key) and depleted (blue key) associations. (C) Contingency tables, stratified by the number of prior losses, of normalized stromal and uNK subset ratios (percentiles) grouped in quartile bins in paired A and B samples. The color key represents −log₁₀ Bonferroni-adjusted P value, as determined by Fisher’s exact test for significantly enriched (red) and depleted (blue) associations. The recurrence rate (percentage) of a stalled decidual response is indicated in red letters. Asterisk (*) above a column in the contingency tables indicates nonuniform distribution of the relative frequency of samples in B biopsy quartile bins per A biopsy quartile bin at P < 0.05, chi-square test. (D) Scatterplots with curves of best fit (solid line) and 95% CI (colored areas).

Fig. 6.. Modeling decidual inflammation in assembloids.

(A) Schematic workflow for establishing assembloids from primary endometrial stromal and epithelial cells. BME, basement membrane extract. (B) Representative confocal image of E-cadherin and vimentin immunofluorescence in an assembloid grown for 8 days in a chemically defined ExM. Nuclei were stained with DAPI. Boxes highlight regions at higher magnification. (C) Schematic of experimental design and representative micrographs of assembloids maintained in an MDM. (D) FC in IL-6 secretion upon decidualization of endometrial assembloids. Colored dotted lines represent the FC in secreted IL-6 levels in five biological repeat experiments. The black solid line represents the median FC. (E) CFU activity in undifferentiated and decidualized endometrial assembloids. The data show median CFU activity (percentage) and individual data points of five biological repeat experiments. Different letters above the data points indicate significance at P < 0.05, Kruskal-Wallis test with Dunn’s multiple comparisons test. (F) Relative change in the SCARA5 and DIO2 mRNA ratio in five biological repeat experiments. Different letters above the data points indicate significance at P < 0.05, Kruskal-Wallis test with Dunn’s multiple comparisons test. (G) Time-dependent induction of MMP10 expression. The data show median MMP10 transcript levels, normalized to L19, and individual data points of five biological repeat experiments. (H) Schematic of experimental design and representative micrographs of assembloids first grown in ExM for 8 days and then alternatingly in MDM and ExM over 3 cycles. (I) Relative change in CFU activity in assembloids across 3 “cycles” of decidualization. The data show median FC in CFU activity and individual data points of five biological repeat experiments. Statistical analysis is based on Wilcoxon matched-pairs signed-rank test for comparison within each cycle.