Active chromatin marks and up-regulation of FOXC1 in uterine epithelial cells demarcate the onset of reproductive decline in aging females

Abstract

Advanced maternal age increases the risk of pregnancy complications due, in part, to changes in the uterine environment. Here, we show that uterine aging in mice is associated with a progressive increase in transcriptional variation, accompanied by a notable accumulation of activating histone marks at multiple genomic loci. Importantly, the transcriptional signatures of uterine aging differ substantially from senescence markers associated with organismal aging. We demonstrate that maternal age-induced effects largely originate in the epithelial compartment and entail a dramatic up-regulation of the pioneer transcription factor FOXC1, combined with a hyper-enrichment for H3K27ac and H3K4me3 across the locus. FOXC1 over-expression in human endometrial epithelial cells causes profound transcriptomic shifts and increased proliferation, recapitulating the aging phenotype. Using endometrial epithelial organoids of young and aged mice, we find that aging hallmarks including Foxc1 up-regulation and epithelial H3K27ac hyper-enrichment are conserved in vitro. Recapitulating the epithelial hyperplasia phenotype seen in vivo, endometrial epithelial organoids from aged mice are larger and mis-express key factors, such as SOX9, critical for uterine gland maturity and function. Collectively, our data highlight the susceptibility of uterine epithelial cells to early-onset aging, demarcated by an increase in activating epigenetic marks that converge on the mis-regulation of FOXC1.

Graphical Abstract

Graphical Abstract.

Introduction

Maternal age is the greatest risk factor for a healthy pregnancy outcome, a fact that is commonly attributed to the exponential increase in chromosomal abnormalities detected in oocytes of older females. However, a significant yet often overlooked contributor to this risk is the impact of advanced maternal age (AMA) on uterine, and in particular decidual function. In humans, this oocyte-independent component of AMA-associated risk manifests in increased rates of miscarriage, late fetal and perinatal death, stillbirth, preterm and extreme preterm birth, low birth weight, placenta previa, and pre-eclampsia, collectively resulting in significantly higher rates of neonatal intensive care unit admissions for women aged 35 years and over [1–11]. Moreover, AMA also increases the risk of various birth defects that cannot be explained by karyotypic abnormalities of the fetus, such as congenital heart defects, hypospadias and skull deformations [12–17]. Many of the pregnancy complications associated with AMA closely overlap with pathologies that have a proven placental origin, indicating that the placenta is likely the key target of uterine aging-induced defects that play out at the feto-maternal (i.e. decidual) interface and carry forward to affect the developing fetus [18]. This concept is strongly supported by mouse models in which a higher risk of congenital heart defects in the embryo has been tied to uterine aging as well as to uterine epithelial dysfunction [19, 20].

These findings have spurred our interest in elucidating the molecular impact of aging on uterine function. In previous work, we and others have demonstrated that AMA in the mouse model results in a greater frequency of developmental abnormalities such as growth retardation, pericardial edema and neural tube closure defects, as well as embryo resorptions [21, 22], that were closely correlated with the severity of placental defects. Importantly, embryo transfer experiments of fertilized preimplantation embryos from aged females into young recipient females resolved the entire spectrum of developmental abnormalities, thus demonstrating that the aging uterine environment, and not the oocyte, were causative of the phenotypes observed [21, 23, 24].

One of the major impacts of AMA in the context of reproductive health is a profoundly diminished capacity of uterine stromal cells to respond to the pregnancy hormones estrogen and particularly, progesterone [21, 25–27]. In mice, this blunted progesterone responsiveness causes a deficit in endometrial stromal cell decidualization and hence a reduced ability of the uterine environment to support a successful pregnancy. A dampened or refractory response of the endometrium to progesterone is also a common finding in women of AMA, which has often been interpreted as a “progesterone-resistant” phenotype and is associated with reduced fertility [28, 29]. Similarly, endometriosis is also associated with progesterone resistance and poorer pregnancy outcomes [30].

Since the reduced cellular responsiveness of “aged” uterine stromal cells to pregnancy hormones is observed in vitro under identical culture conditions, it must be caused by cell-intrinsic defects in transducing the hormonal signal, and not just altered hormone levels per se [21, 27]. The functional deficits in the uterine stromal cell compartment of aged females are accompanied by profound epigenetic differences, as determined by global and locus-specific changes in DNA methylation as well as in activating and repressive histone marks [27, 31–33]. Intriguingly, these data pointed to the possibility of insufficient priming of gene promoters with H3 lysine 4 trimethylation (H3K4me3), which correlates with impaired activation of the associated decidualization genes [27].

In addition to the hormonally induced changes that normally occur in the stromal cell compartment, uterine adaptations to pregnancy rely on the normal functioning of endometrial glands. This has been unambiguously demonstrated in mouse models that cause glandless uteri, which can be achieved by progesterone treatment of neonatal mice [34] or genetically by knockout (KO) of the genes encoding the transcription factor Foxa2 or the cytokine Wnt7a [35–38]. Females lacking uterine glands are infertile, primarily due to the inability to produce leukemia inhibitory factor (LIF) which is required for implantation in mice. However, while LIF supplementation can overcome this early bottleneck, glandless mice still exhibit decidualization defects leading to an abrupt loss of pregnancy at embryonic day (E)7.5. These experiments have provided explicit evidence for the vital importance of the crosstalk between uterine glands and the surrounding uterine stroma for the establishment and maintenance of pregnancy [39]. Yet how glandular function is affected by AMA has not been determined.

In the current study, we set out to explore when, and where, the first signs of uterine aging take effect. To this end, we assessed the impact of AMA in the mouse model in a refined manner over a time course of female age ranges. By pursuing combined transcriptomic and epigenomic profiling approaches, we find that uterine aging is accompanied by a progressive accumulation of epigenetic activating marks at multiple genomic loci that results in the mis-regulation of gene expression primarily in uterine epithelial cells. Endometrial epithelial organoids recapitulate this age-induced transcriptional mis-regulation. Our data suggest that this phenotype may be driven by FOXC1 hyperactivation that results in up-regulation of other critical determinants of uterine gland function, notably SOX9. Collectively, this over-activity generates a hyper-proliferative phenotype that results in glandular hyperplasia and demarcates one of the earliest signs of aging-induced uterine defects.

Materials and methods

Mice

All animal work was conducted in full compliance with regulatory requirements and approvals: Whole uterine tissues for RNA-seq and ChIP-seq were obtained from mice held at the Babraham Institute, Cambridge, UK, in full compliance with UK Home Office regulations and with approval of the local animal welfare committee (AWERB), and with the relevant personal and project licences in place (PPL 70/8276 and PPL PP0189059). All other experiments were conducted at the University of Calgary (UCalgary), Calgary, Canada, with approval by the University of Calgary’s animal care committee and with adequate animal use protocols in place (protocol AC22-0118). Mice were housed in IVC cages with 12-h light–dark cycles under ambient temperature (∼22°C) and humidity conditions [40].

All females assessed in this study were of the C57BL/6 strain, either C57BL/6Babr (Babraham Institute) or C57BL/6N (University of Calgary) obtained from Charles River (Saint Constant, QC, Canada), maintained as a breeding colony. Virgin females were caged in groups up to 5 until they had reached the desired age as indicated. For assessment of pregnancy outcomes, females of the indicated age ranges were mated to C57BL/6 males aged 8–30 weeks, counting the day of the vaginal plug as E0.5. Embryos were dissected at E11.5 and photographed using a Nikon dissecting microscope. No overt differences in the number of implantation sites were observed, as reported previously [21]. All mechanistic studies of uterine changes as a function of age (RNA-seq, ChIP-seq) were conducted on virgin C57BL/6 mice that had been mated to vasectomized CD1 males (8–36 weeks of age) to hormonally prime the females for receptivity. Uteri were dissected at E3.5 and each uterine horn cut into 2–4 pieces that were snap-frozen and stored at −80°C for RNA isolation and chromatin preparation. For histological analyses and endometrial epithelial organoid preparation, uteri were dissected from non-mated virgin C57BL/6 females and either fixed in 4% paraformaldehyde followed by embedding in Clear Frozen Section Compound (VWR, 95057-838), or processed for cell isolation as described below.

Mouse endometrial epithelial organoid culture

To establish murine endometrial epithelial organoid (mEEO) cell cultures, uteri of C57BL/6 females falling into the young (Y) or into the aged 2/3 (A2/3) category were used. Vaginal smears were taken from all females to establish estrous cycle staging by crystal violet staining. A total of six females were in estrus, and the other six were not, distributed between the young and aged samples. Estrous cycle timing did not impact mEEO growth or expression signatures after the indicated culture periods. Uteri were dissected, longitudinally cut open and digested in 0.25% Trypsin, 2.5% Pancreatin solution for 1 h (Sigma, T4799-Trypsin and P3292-Pancreatin). The epithelial cell lining was gently scraped to release epithelial sheets, which were subsequently broken into epithelial cell fragments by trituration. Cell fragments were pelleted and then resuspended in ice-cold, growth factor-reduced and phenol red-free Matrigel (Corning, 356231). Cells were plated as Matrigel droplets in the center of each well of a 48-well plate. The Matrigel was left to polymerize for 30 min at 37°C, then overlaid with phenol red-free DMEM/F12 medium containing L-glutamine (Wisent Inc., 319-080 CL) supplemented with a cocktail of signaling and growth factors comprised of B27 (1×; Life Technologies, 12587010), N2 (1%; Life Technologies, 17502048), insulin-transferrin-selenium (ITS) (1%; Gibco, 51500-056), WNT3A (200 ng/ml; Abcam, ab81484), EGF (50 ng/ml; Gibco, PHG0313), NOGGIN (100 ng/ml; Peprotech, 120-10C), R-Spondin 1 (200 ng/ml; BioTechne, 4645-RS-100), FGF10 (100 ng/ml; Peprotech, 100-26), A83-01 (500 nM; Cayman CC, 9001799), nicotinamide (10 μM; Sigma, N0636), Y-27632 (10 μM; Cayman CC, 10005583), and anti-mycotic/antibiotic (1×; ThermoFisher, 15240-062). Organoids were used at first or second passage for all experiments. Hormonal treatment of organoids was performed by exposing them to 10 nM estradiol (Sigma, E2758) for 48 h, followed by exposure to 10 nM estradiol, 1 μM progesterone (Sigma, P8783), and 100 ng/μl cAMP (Sigma, B5386) for 72 h. 0.02% DMSO (ThermoFisher, J66650-AD) served as vehicle-control.

Ishikawa cells

Ishikawa cells, a human endometrial adenocarcinoma cell line (Sigma–Aldrich, 99040201), were cultured in RPMI 1640 medium with L-glutamine (ThermoFisher Scientific, 21875-034), supplemented with 10% fetal bovine serum (Wisent Inc., 098150), 2 mM sodium pyruvate (ThermoFisher Scientific, 11360-039), 1× anti-mycotic/antibiotic (ThermoFisher Scientific, 15240-062), and 50 μM 2-mercaptoethanol (Gibco, 31350) at 37°C in a 5% CO₂ incubator. To achieve overexpression of FOXC1 and SIX1, the full-length open reading frames of the murine orthologs were cloned into a pCAG-PiggyBac vector and sequence-verified. The amino acid sequences of mouse and human FOXC1 are 92% identical and 93% similar, and those of SIX1 are 99% identical. The expression constructs, or the corresponding empty vector, were co-transfected together with a PiggyBac transposase plasmid using Lipofectamine 2000 (Life Technologies, 11668019), following the manufacturer’s protocol. Stable cell lines were selected using 10 μg/ml Blasticidin S (Wisent Inc., 450-190-WL).

Cell proliferation rates were determined by plating 25000 cells as starting material in quadruplicate, followed by cell counting on an automated cell counter (Luna II, Logosbio) on each subsequent day. For PHH3 staining, cells were plated on coverslips in six-well plates at equal cell densities, fixed in 4% paraformaldehyde, permeabilized for 30 min in PBS, 0.1% Triton X-100 followed by blocking in phosphate-buffered saline (PBS) containing 0.5% bovine serum albumin (BSA; Sigma, A7906) and 0.1% Tween-20 (Sigma, P1379) for 30–60 min at room temperature. Primary antibody against PHH3 (Merck Millipore, 06-570) was used at 1:100 diluted in PBS, 0.5% BSA, 0.1% Tween-20, and incubated for 60 min at room temperature. Anti-rabbit AlexaFluor488-conjugated secondary antibody (ThermoFisher, A-21206) was used at 1:500 for detection. Nuclear counterstaining was performed with 4′,6-diamidino-2-phenylindole (DAPI; Sigma, D9542), after which coverslips were mounted onto slides with Fluoromount-G (SouthernBiotech, 0100-01) mounting medium. Images were taken on a Zeiss DMRE epifluorescence microscope, and mitotic rate analysis performed by cell counting in ImageJ.

RNA isolation and RNA-sequencing

Snap-frozen intact uterine fragments were immersed in 1 ml TRI reagent (Sigma, T9424) and immediately macerated with a tissue homogenizer. RNA was isolated according to the manufacturer’s protocol. RNA quality was assessed by gel electrophoresis and on a Tapestation. For organoids and Ishikawa cells, total RNA was extracted and DNase-digested using the Qiagen RNeasy Mini kit (Qiagen, 74104) or the Invitrogen PureLink^™ RNA Mini Kit (Invitrogen, 12183020) following manufacturer’s instructions.

RNA-sequencing libraries of uterine tissue were prepared by the Babraham Institute’s Sequencing Facility using the Illumina TruSeq^® Stranded mRNA Library Preparation kit (Illumina, 20020595) and sequenced on a Hiseq2500 sequencer using a Rapid run mode (100 bp single-end). RNA-sequencing of organoids and Ishikawa cells was performed by UCalgary’s Centre for Health Genomics and Informatics, using either the NEB Ultra II Directional RNA Library Prep kit (New England Biolabs, E7760L) or the Illumina Stranded mRNA prep, ligation kit (Illumina, 20040534), and sequenced on a NextSeq2000 instrument using the P4 100 cycle XLEAP protocol (50 bp paired-end).

Reads were aligned with hisat2 for whole-tissue uteri and with STAR (2.6.1a_08-27) for organoids and Ishikawa cells. Alignment of murine samples was to the mouse reference genome (GRCm38/mm10), and of human samples (Ishikawa cells) to the human reference genome (GRCh38). Count tables were assembled with htseq-count (bioconda 2018.11) using the reverse strand and non-unanimous settings, resulting in library sizes between 13M and 44M reads per sample.

Differential gene expression analysis was performed in SeqMonk program v1.48.1 (https://www.bioinformatics.babraham.ac.uk/projects/seqmonk/) with integrated R codes for DEseq2 analysis (https://www.R-project.org/). Differential expression was calculated using DESeq2 and adjusted for multiple testing correction using the Benjamini–Hochberg method (padj< 0.05). The SeqMonk program was also used for generation of heatmaps and correlation matrices. Principal component analysis (PCA) plots were generated using either R or SeqMonk. Read counts per million (RPM) were calculated normalized to library size and with merged transcript isoform settings. To avoid negative values after logarithmic transformation, a numerical value of 1 was added to each gene count prior to log2-transformation to generate log2(RPM + 1) values. Gene Ontology (GO) analyses were performed with DAVID [41] and motif analyses with Homer [42]. FOXC1 and SIX1 ChIP-seq data were integrated from GSM3227808 (kidney) and GSE108130 (E10.5 embryo), respectively.

Chromatin immunoprecipitation-sequencing (ChIP-seq)

Ultra-low input native ChIP-seq was performed on pieces of snap frozen endometrial tissue as previously described [43]. Briefly, tissue was thawed on ice in 100 μl nuclear isolation buffer (Sigma, NUC-101) and diced using a blade. An aliquot of 20000 nuclei in 10 μl NIB was then taken forward for MNase (NEB, M0247S) digestion at 21°C for 7.5 min. Digested chromatin was then suspended in Complete IP buffer [20 mM Tris–HCl (pH 8), 2 mM EDTA, 150 mM NaCl, 0.1% Triton-X100, 1× Protease inhibitor cocktail, and 1 mM phenylmethanesulfonyl fluoride (PMSF)]. Following pre-clearing the chromatin on a mix of protein A/protein G beads (ThermoFisher Scientific), and after 10% was set aside for input, the chromatin was divided for immunoprecipitation (IP). IP was performed overnight rotating at 30 rpm, at 4°C, using 250 ng antibodies against H3K4me3 (Diagenode, C15410003) and H3K27ac (Abcam, ab4729). DNA was washed in low salt and then high salt washes, before eluting in ChIP elution buffer (100 mM NaHCO₃ and 1% SDS). DNA was then SPRI bead-purified before continuing to library preparation using the MicroPlex Library Preparation kit (Diagenode, C05010001). Libraries were quantitated using the Agilent Bioanalyzer High Sensitivity DNA assay before pooling for sequencing. Libraries were sequenced on the Illumina NextSeq500, with 75 bp paired-end runs.

Reads were aligned to the mouse genome assembly GRCm38 using bowtie2. We assessed the data quality using ChIPQC (v1.34.0). Peaks were identified using MACS2 (v2.2.9.1) with q < 0.01. The DiffBind R package (v3.8.0) was used to find consensus peaks in different sample groups and perform differential binding analysis. Consensus peaks were defined as those present in >66% of samples within each age group. A binding matrix was generated using read counts from consensus peaks overlapping within each group (summit width = 200 bp), followed by normalization. This matrix was then used to identify differentially enriched regions between age groups using DESeq2 (as implemented by default in DiffBind), applying a false discovery rate (FDR) threshold of ≤ 0.05. Peaks were assigned to genes or nearby genes using ChIPpeakAnno (v3.32.0). In addition to gene assignment, peaks were annotated for genomic feature overlap—whether located in TSS, exon, 5′ UTR, 3′ UTR, intronic, or intergenic regions using ChIPseeker (v1.34.1). deepTools (v3.5.1) was used to visualize the average profiles of histone modifications across gene regions ±5 kb in different groups.

To identify super-enhancers, H3K27ac peaks were called with MACS2 (with parameter — broad) and used as input for the ROSE (Rank Ordering of Super-Enhancers) software, which stitches together enhancer peaks located within 12.5 kb of each other [44]. HOMER was used to identify motifs in the predicted enhancer regions [42].

Cell type-specific deconvolution analysis

The AUCell R package (v1.16.0) [45] was used to evaluate the enrichment of DE genes in scRNA-seq data. AUCell builds a gene expression ranking matrix for each cell and computes the enrichment of the DE gene set among the top-expressed genes in each cell. To compute AUCell scores, we re-analyzed and annotated publicly available wild-type uterine scRNA-seq data (GEO accession: GSE118180) [46] using canonical marker expression. Seurat (v5.0.3) [47] modules were used for processing, analyzing, and visualizing the scRNA-seq data. DE gene sets were input into AUCell for area under the curve (AUC) value calculation. Cells expressing more genes from the DE gene set exhibit higher AUC values than those expressing fewer. After manual inspection of AUC thresholds, we selected cells with AUCell scores > 0.04. Based on AUCell mapping, cells in the UMAP embedding were color-coded. Analysis code is available at: doi.org/10.5281/zenodo.16782162.

Immunofluorescence staining

Frozen 7 μm sections were warmed up at 37°C for 30 min, then washed in PBS, permeabilized in PBS, 0.1% Triton X-100, and blocked in PBS, 0.5% BSA, 0.1% Tween-20 for at least 1 h at room temperature. Primary antibody incubations were performed overnight at 4°C. Antibodies used were against FOXC1 (1:100; Abcam, ab227977); Ki67 (1:250; ThermoFisher, 14-5698-82); CD31 (1:50; BD Biosciences, 550274); NG2 (1:100; ThermoFisher, MA5-24247); SOX9 (1:200; Abcam, ab185966); CDH1 (1:300; BD Transduction Laboratories 610 182), H3K27ac (1:500, Abcam ab4729), and PHH3 (1:100; Merck Millipore, 06–570), all diluted in PBS, 0.5% BSA, 0.1% Tween-20. Primary antibodies were detected with appropriate AlexaFluor488- or AlexaFluor568-conjugated secondary antibodies (1:500, ThermoFisher). Counterstaining was performed with DAPI (Sigma, D9542). Slides were mounted with Fluoromount-G (SouthernBiotech, 0100-01) mounting medium and imaged at a Leica DMRE epifluorescence microscope, using identical exposure and gain settings for all immunofluorescence stainings that were subsequently quantitatively assessed.

Quantification of immunofluorescence stainings in glandular epithelium (SOX9) was performed in ImageJ (v1.53) by using the DAPI channel to generate a mask identifying epithelial nuclei, and then measuring the relative intensity of the channel-of-interest in these marked areas. For quantification of FOXC1 and H3K27ac intensities on uteri, the density of luminal epithelial cells interfered with automated thresholding on the DAPI channel. Therefore, we defined small squares (for glandular epithelial and stromal cell nuclei) or rectangles (for luminal epithelial cell nuclei) that just captured individual nuclei, and manually measured luminal, glandular and stromal cell staining intensities. Identical-sized areas were used for each individual compartment. Glandular and luminal epithelial cells were identified based on morphological appearance and by CDH1 staining. Equally-sized areas were used to determine background intensities in the uterine and glandular lumen, that was subtracted. Measurements (in arbitrary units) were statistically analyzed using GraphPad Prism v10.4.1.

Western blotting

Approximately 1 cm long pieces of uterine tissue from young (11.4 weeks) and aged 2 (52 weeks) females were snap-frozen immediately on dissection and macerated into tissue powder under liquid nitrogen. Protein lysates were prepared in RIPA buffer (Sigma R0278), followed by sonication in a Bioruptor (30 s on/off cycles, two repeats). Fifty micrograms of protein lysate per sample was loaded onto 4%–20% gradient mini-PROTEAN TGX pre-cast SDS gels (BioRad, 4568094), transferred onto polyvinylidene difluoride (PVDF) membrane (Novex Life Technologies, LC2002), and blocked with 1% milk powder in Tris-buffered saline containing 0.1% Tween-20 (TBST). Primary antibodies were against H3K27ac (Abcam, ab4729; used at 1:1000) followed by anti-rabbit HRP detection (Cell Signaling Technology, 7074S) using Pierce^™ ECL Western Blotting Substrate (ThermoFisher, 32209) reagent. After stripping, the blot was probed for pan-H3 (Invitrogen, PA5-16183; used at 1:1000). Quantification of band intensities was performed in ImageJ v1.54p and statistical analysis in GraphPad Prism v10.4.1.

Results

Female aging results in a progressive decline of reproductive performance

Previous studies have demonstrated that in mice, the predominant impact of AMA in the context of reproductive performance is mediated by declining uterine function and not reduced oocyte fitness [21, 22]. These studies have assessed endpoint effects of AMA in females toward the end of their reproductive lifespan, which in C57BL/6 mice is around 52 weeks of age [21, 22, 27, 31]. Yet it has remained unknown whether the observed reproductive decline, defined here as the impact of maternal age on developmental progression in utero, occurs progressively as females grow older, or whether a critical “cliff-edge” exists after which pregnancy success rates are precipitously reduced. To shed light onto this question, we set up timed matings between C57BL/6 stud males and virgin females of four defined age ranges: young (Y) females were 8–20 weeks (wks) of age, aged 1 (A1) 24–36 wks, aged 2 (A2) 37–52 wks, and aged 3 (A3) 53–65 wks (Fig. 1A). Following observation of a vaginal plug indicating a successful mating, we dissected the females at E11.5 to visually assess the developmental progression of each embryo in the litter. Of note, our previous studies had demonstrated that paternal age had little influence on gross developmental outcome [21], and hence we used stud males between 8 and 30 wks of age throughout.

Figure 1.

The impact of advancing maternal age on reproductive outcome. (A) Representative images displaying gross morphology of E11.5 embryos developed in primiparous C57BL/6 females of increasing age ranges grouped as follows: young (8–20 weeks), aged 1 (24–36 weeks), aged 2 (37–52 weeks), and aged 3 (>52 weeks). Each row represents embryos from a single litter. Total numbers of litters analyzed is provided in panel (B); scale bars: 1 mm. (B) Scoring of the severity of developmental defects by visual assessment of embryo appearance. Severity scores were qualitatively assigned with 0 = normal, 1 = mild growth retardation, 2 = severe growth retardation and/or obvious developmental pathology (heart, brain defects), and 3 = dead or resorbing embryos. Violin plots of the frequency of these scores highlight the age-associated increase in the severity of developmental defects. Statistical analysis was performed using two-way ANOVA followed by Holm-Šídák’s multiple comparisons test; *P< 0.05, **P< 0.01, and ****P< 0.0001. The number of litters and embryos assessed for each age group is indicated. Specifically, the “young” group assessed consisted of 16- to 20-week-old females to establish the upper limit of maternal age for which no prevalent decline in embryo development was observed. (C) Schematic of experimental strategy for subsequent experiments. Virgin C57BL/6 females of increasing age ranges (Y, A1, A2, and A3) were primed for pregnancy by mating to vasectomized males, and E3.5 pseudopregnant uteri collected for downstream transcriptomic and epigenomic analyses. Histological analysis was performed on an independent set of uterine samples.

Developmental progression of individual embryos in litters to Y females (each row of embryos represents one litter) was highly uniform by visual appearance, as expected (Fig. 1A and B, and Supplementary Fig. S1). Specifically, the litters examined were to females aged 16–20 wks, yet they were indistinguishable from the developmental outcomes we had reported for 8- to 12-wk-old females [21], thus determining the upper limit of maternal age for which no prevalent decline in embryo development was observed. By contrast, in age groups A1–A3, the spectrum of developmental progression became more and more variable. Generally, with increasing age of the female, fewer embryos per litter developed normally while an increasing number of embryos exhibited developmental defects. These defects ranged from an increasing severity of growth retardation combined with obvious developmental brain and heart defects to implantation sites with embryos destined to fail or already resorbed (Fig. 1A and Supplementary Fig. S1). Implementing a severity scoring system from “normal” to “dead/resorbed” provided a visual readout of the progressive rise in developmental defects as a function of maternal age (Fig. 1B and Supplementary Fig. S1). These findings demonstrated that the AMA-induced decline in reproductive competence in female C57BL/6 mice occurs from about 6 months of age onward, with increasing numbers of embryos failing to develop normally with advancing maternal age.

Increasing variability of uterine transcriptomes underpins the AMA-associated reproductive decline

To gain molecular insights into the gene regulatory changes that are associated with this progressive reproductive decline, we set up females of the same four age groups with vasectomized males to trigger hormonal priming of the uterus, which induces a state of pseudopregnancy. Uteri were dissected at E3.5, i.e. the developmental stage equivalent to the peri-implantation window when the blastocyst would start to attach and then penetrate into the uterine wall, and cut into multiple pieces for processing toward transcriptomic and epigenomic analyses (Fig. 1C).

Bulk RNA sequencing (RNA-seq) analysis of these uteri revealed increasingly divergent gene expression profiles with advancing maternal age. Thus, PCA performed on the entire transcriptomes showed a relatively tight cluster of uterine samples from Y females whereas those from age groups A1, A2, and A3 spread further and further along PC1, with a wide spectrum of divergence between biological replicates (Fig. 2A). This increase in sample dissimilarity was further visualized in a heatmap of Euclidean distances which highlighted the tight grouping of Y samples but an increasing transcriptional heterogeneity between individual uteri of age groups A1–A3 (Fig. 2B).

Figure 2.

Uterine aging increases transcriptional variation but remains largely distinct from organismal aging. (A) PCA on RNA-seq data from bulk uterine tissue derived from females of the age ranges defined in Fig. 1. PCA was performed on the top 10% most variable genes, with the clouds indicating the 95% confidence interval. Curves on top of the graph depict the smoothened density plots for each principal component, reflecting the continuous probability distribution of data, which showcases the increase in transcriptional variability in aging females. All RNA-seq data are from Y, n = 8; A1, n = 4; A2, n = 6; A3, n = 9 independent biological replicates (i.e. uteri from different females). (B) Correlation matrix based on Euclidean distances of uterine RNA-seq data showing increased transcriptomic divergence in samples of AMA, indicating a progressive increase in inter-sample variability with age. (C) Bar graph depicting the number of significantly up- and down-regulated genes identified by DESeq2 analysis, highlighting an age-dependent increase in DE genes. Exponential and linear trend lines were added in Excel to the up- and down-regulated genes, respectively. (D) Venn diagram illustrating the overlap of DE genes identified in pair-wise comparisons of young (Y) versus age groups aged 1 (A1), aged 2 (A2), and aged 3 (A3). (E) GO analysis on the 45 DE genes that are shared between all aging groups. (F) Venn diagram illustrating the limited overlap between common senescence markers (“organismal aging genes”), as defined by the Tabula Muris consortium [48], and the DE genes identified in aging uteri. Overlaps are shown for genes that are DE in A1 or in any of the aging uteri groups (“A any”). (G) Expression trend plots of genes that are most robustly up-regulated in the Tabula muris dataset. No consistent trends are observed in aging uteri, and inter-sample variability is high even for genes that show a propensity toward up-regulation.

In line with the rise in abnormally developed embryos, we also observed a dramatic increase in the number of differentially expressed (DE) genes between Y uteri and those of the aging females. Performing pair-wise differential gene expression analyses using DESeq2 between Y and A1, A2, and A3, we identified 87, 528, and 1280 DE genes, respectively. Notably, the majority of these DE genes were up-regulated in aging uteri (Fig. 2C). Thus, the number of up-regulated DE genes increased almost exponentially between A1, A2, and A3, whereas the down-regulated genes followed a more linear trend (Fig. 2C). These data indicated that uterine aging is associated with a loss of transcriptional fidelity and specifically with an overall tendency toward gene activation that included about 30% of up-regulated genes that are normally barely expressed in uteri of Y females [303/1047 up-regulated DE genes have a log₂ (RPM + 1) ≤ 1 in young uteri] (Supplementary Fig. S2A).

Uterine aging is largely distinct from organismal aging

Intersecting the DE gene lists obtained from the pair-wise comparisons between Y versus A1, A2, and A3 revealed a rather small overlap of 45 genes that were dysregulated in all advanced age groups (Fig. 2D), 33 of which were up-regulated and 12 down-regulated. This finding underpinned the general observation of an increased transcriptional heterogeneity in uteri of aged females, as opposed to a growing set of shared DE genes. GO analysis on these 45 common DE genes highlighted terms related to cellular senescence and aging, as well as positive regulation of transcription (Fig. 2E), compared to much broader and less specific terms on A3 DE genes (Supplementary Fig. S2B). However, when comparing the uterine aging DE genes with a reported consensus list of senescence markers that are dysregulated in various aging mouse tissues [48], a minimal overlap of only four genes was observed (Fig. 2F). These included Cdkn2a and Cdkn2b (encoding p16INK4a and p15INK4b, respectively) (Supplementary Fig. S2C), which are some of the most well-established aging genes [49–51]. However, these were the only robust aging signatures that were identified in the uteri. Even when relaxing the criteria to include DE genes identified in any of our Y versus A1/A2/A3 comparisons, only an additional 14 genes were identified but their expression dynamics showed little consistency between samples, even in age group A3 (Fig. 2F and Supplementary Fig. S2D). We then plotted the expression dynamics of the most significant Tabula Muris senescence-associated genes on our data; however, these factors also exhibited no consistent mis-regulation in the aging uteri (Fig. 2G and Supplementary Fig. S2E). These data suggested that the gene expression changes associated with uterine aging are largely distinct from common senescence-associated “organismal” aging genes.

Uteri of aging females display a striking peak enrichment for active histone modifications

Leading on from these global changes in gene expression in uteri of aging females, we aimed to investigate whether these transcriptional shifts were influenced—or potentially driven—by underlying epigenetic mechanisms. Because of the bias toward transcriptional activation, we focussed on the modifications histone 3 lysine 4 trimethylation (H3K4me3) and histone 3 lysine 27 acetylation (H3K27ac) that demarcate active promoters and enhancers. We performed chromatin immunoprecipitation followed by sequencing (ChIP-seq) using an ultra-low input protocol [43] on uterine tissue pieces from the same females for which we had generated the RNA-seq data (Figs. 1C and 3A). After MACS2 peak calling on individual samples, we identified a total of 25836 H3K4me3 and 54362 H3K27ac high-confidence consensus peaks using DiffBind (Fig. 3A and Supplementary Fig. S3A). Around 40% of H3K27ac peaks overlapped with H3K4me3 peaks, predominantly at gene promoters (Fig. 3B and Supplementary Fig. S3B). We then performed differential enrichment analysis between age groups using the R Bioconductor package DiffBind, which identified a sharp increase in the number of differentially enriched peaks for both histone marks with advancing female age (Fig. 3C–H). For both modifications, the mean enrichment across all differentially marked sites was elevated in aging uteri compared to Y, and rose progressively from A1 through to A3 (Fig. 3D and G). This trend was not evident for non-differentially enriched peaks (Supplementary Fig. S3C). Perhaps most strikingly, however, was the fact that the vast majority of differentially enriched sites gained in H3K4me3 or H3K27ac enrichment in age groups A1 and A2 (Fig. 3E and H, and Supplementary Figs. S3D, S4A and B). This was despite a global trend toward lower H3K27ac levels in uterine tissue with age (Supplementary Fig. S3E and F), pointing toward a genomic re-distribution or cell type-specific enrichment of H3K27ac. Integrating the epigenomic tracks with the transcriptomic data showed that 14 out of the 33 (42%) core up-regulated genes (Fig. 2D) were associated with an enriched H3K27ac peak (Fig. 3I). Conversely, an increasing and decreasing enrichment of H3K27ac was also broadly associated with transcriptionally up- and down-regulated DE genes, respectively (Fig. 3J). While differentially marked H3K4me3 and H3K27ac peaks in A3 were more evenly split between enrichment and depletion, the extent of these differences was less pronounced compared to A1 and A2 (Supplementary Figs. S3D and S4A and B). Collectively, these findings suggested that the broadly changed epigenetic landscape in A3 was likely the result of secondary tissue alterations due to advanced aging, as perhaps corroborated by the wide spectrum of GO terms enriched in A3 DE genes (Supplementary Fig. S2B). By contrast, the early epigenetic and transcriptional aberrations in age groups A1 and A2, i.e. from 24 wks of age onward, were likely the drivers of uterine aging linked to the initiation of uterine dysfunction, as defined by the onset of developmental abnormalities in embryos to females of these age groups (Fig. 1 and Supplementary Fig. S1).

Figure 3.

Active histone marks H3K4me3 and H3K27ac are dramatically enriched in aging uteri. (A) Schematic diagram of the experimental strategy and the number of consensus ChIP-seq peaks identified for H3K4me3 and H3K27ac. All ChIP-seq data are from Y, n = 7; A1, n = 4; A2, n = 6; A3, n = 10 independent biological replicates (i.e. uteri from different females). (B) Overlap between H3K27ac and H3K4me3 peaks. (C) Bar graph of the number of differentially enriched H3K4me3 peaks identified in pair-wise comparisons between the young group and aged 1, aged 2, or aged 3 samples. Note that data are plotted on a log₁₀ scale. (D) Box whisker plots displaying the mean enrichment levels of the n = 138 H3K4me3 peaks that are differentially enriched in A1 and/or A2 groups. Overall H3K4me3 enrichment increases across these sites. The box represents the 25th and 75th percentiles of the data sets around the median, which is shown by a horizontal line. The whiskers show the median plus/minus the interquartile (25%–75%) range multiplied by 2. (E) Heatmap of H3K4me3 occupancy levels at peaks differentially enriched in A1 or A2. The vast majority of sites gains in H3K4me3 enrichment. (F) Bar graph of the number of differentially enriched H3K27ac peaks identified in pair-wise comparisons between the young group and aged 1, aged 2, or aged 3 samples. Note that data are plotted on a log₁₀ scale. (G) Box whisker plots displaying the mean enrichment levels of the n = 62 H3K27ac peaks that are differentially enriched in A1 and/or A2 groups, depicted as described in (D). (H) Heatmap of H3K27ac occupancy levels at peaks differentially enriched in A1 or A2. All sites except for one single peak gain in H3K27ac enrichment with age. (I) Heatmap of gene expression levels of DE genes that are up-regulated with age and overlap with differentially enriched H3K27ac peaks. (J) H3K27ac enrichment at genes DE in any of the age groups, separated into up- and down-regulated genes. H3K27ac enrichment aligns with global trends of gene expression. The box represents the 25th and 75th percentiles of the data sets around the median, shown by a horizontal line. The whiskers extend as calculated by the Tukey method.

Foxc1 hyperactivation is a hallmark of uterine aging

To explain the global trend of transcriptional activation coupled to the enrichment in active histone modifications in uteri of aging females, we first speculated that a strong transcriptional repressor or repressive chromatin modifier might be down-regulated as a function of age. We assembled a large list of such candidate factors based on GO terms including histone deacetylase activity (GO:0004407), PRC2 and PRC1 complex components (GO:0035098, GO:0035102), histone H3K9 methyltransferase activity (GO:0140949, GO:0140948, and GO:0140947), transcription corepressor activity (GO:0003714), components of the NuRD, CoRest, and BCOR complexes, TRIM28, KRAB-ZFPs, and DNA methyltransferases. Unexpectedly, none of these genes exhibited any substantial changes in expression across the uterine aging time course (Supplementary Fig. S4C). Hence, we turned to the alternative possibility that the induction of a strong transcription factor or transcriptional activator may account for our findings. Amongst the up-regulated DE genes were only three genes matching these criteria, the transcription factors Foxc1, Six1, and Zfhx4 (Fig. 4A and Supplementary Fig. S5A).

Figure 4.

Foxc1 is an early target of uterine aging and drives many age-associated transcriptional changes. (A) Expression dynamics of Foxc1 across the uterine aging time course as determined by RNA-seq. Data are displayed as mean ± S.E.M. Statistical analysis was performed by one-way ANOVA with Dunnett’s multiple comparisons test. **P< 0.01, ****P< 0.0001. Y, n = 8; A1, n = 4; A2, n = 6; A3, n = 9 independent biological samples. (B) Wiggle plots of expression and histone modification enrichment across the Foxc1 locus. FOXC1 ChIP-seq binding sites are indicated by the yellow bar. Dashed rectangles highlight regions differentially enriched for H3K4me3 and H3K27ac. Foxc1 was the locus with the most significantly enriched H3K27ac peak in Y versus A1, both in terms of magnitude and significance of enrichment (P = 3.89E-07). It remained the most significantly enriched H3K27ac peak also in the Y vs A2 comparison (P = 2.15E-10). (C) Fraction of genes differentially expressed between uteri from young females and age groups A1 or A2 that are in proximity to a FOXC1 genomic binding site (5 kb distance cut-off), and hence are likely regulated by this transcription factor. FOXC1 ChIP-seq data were from [52]. (D) Transcription factor motif discovery analysis on H3K27ac-enriched super-enhancers using Homer [42] identifies the FOX motif. (E) Tree diagram based on Euclidean distances of RNA-seq data generated from human endometrial Ishikawa cells stably carrying empty vector (EV) as control, or FOXC1-, SIX1-, or FOXC1- and SIX1-encoding expression plasmids, in the presence or absence of the pregnancy hormones progesterone and estrogen combined with cAMP (PEc). Biological replicates were performed in quadruplicate. The main branch point in sample separation is determined by the presence or absence of FOXC1. (F) PCA performed on whole transcriptomes of Ishikawa cells carrying stable expression constructs as described in panel (E). PC1 is driven by FOXC1, whereas PC2 is determined by hormonal stimulation. SIX1 has a comparatively minor effect on global expression profiles. (G) Fractions of the DE gene sets in aging mouse uteri that are determined by FOXC1. The overlap was calculated by intersecting the DE gene lists from aging uteri with those of FOXC1- (or FOXC1 + SIX1)-over-expressing Ishikawa cells. Statistical analysis was performed with Fisher’s Exact test; *P< 0.05.

Transcripts for these three transcription factors were consistently more abundant across all aging categories. However, compared to the Y control samples, Foxc1 exhibited the earliest and most consistent transcriptional induction in A1 and A2, i.e. the age groups when the first (epi)genomic changes start to manifest. Importantly, the Foxc1 locus was also the first to display epigenetic changes, harboring the most significantly enriched H3K27ac peak in A1 (P = 3.89E-07) and joint H3K4me3 and H3K27ac gains in A2 and A3 (Fig. 4B and Supplementary Fig. S5B, Supplementary Tables S1 and S2). With Foxc1 being a comparatively small gene encompassing only ∼1.6 kb, both histone marks were distributed broadly over the entire intron-less Foxc1 gene locus. Zfhx4 was differentially enriched for H3K27ac but only in A3, whereas the abundance of both active histone marks at the Six1 promoter regions were visually present but evaded statistical significance in our strict peak calling method (Supplementary Fig. S5C).

Very little is known about ZFHX4, but FOXC1 and SIX1 have been previously characterized in various other tissue contexts. We therefore downloaded existing ChIP-seq data [52, 53] and assessed the relationship of DE genes to FOXC1 and SIX1 binding sites, using a conservative cut-off of 5 kb. This analysis showed that around 18% of all genes mis-expressed at either A1 or A2 are close to a FOXC1 binding site, and this number was even higher (20%) when considering the up-regulated genes only (Fig. 4C). When including SIX1 binding sites, this number rose to over 40% of genes that were specifically up-regulated in A1, and to 26% of genes up-regulated in A2, that were in proximity to either FOXC1- or SIX1-binding sites (Supplementary Fig. S5D). We also used our H3K27ac ChIP-seq data to call super-enhancers using ROSE [44, 54], which stitches together H3K27ac peaks located within 12.5 kb of each other. Thereby, we identified 121 putative super-enhancers that exhibited differential enrichment in aged uterine samples, again with the vast majority being more abundantly marked by H3K27ac. Motif analysis revealed that these differentially bound super-enhancers were significantly enriched for the forkhead homeobox (FOX) consensus sequence (Fig. 4D), which is similar between various FOX domain-containing transcription factors. We also noted that Foxc1, Six1, and Zfhx4 all contained FOXC1- and SIX1-binding sites themselves, indicating that these genes likely establish a self-reinforcing network.

To elucidate the impact of FOXC1 and SIX1 on the endometrial transcriptome, we generated FLAG-tagged expression constructs and established stable over-expressing cell lines (Supplementary Fig. S5E), using the human endometrial Ishikawa cell line as a model of the endometrial epithelium [55, 56]. Ishikawa cells express low-to-moderate levels of FOXC1 or SIX1 in ground state conditions. Transcriptomic analysis of empty vector control, FOXC1-, SIX1-, and FOXC1+ SIX1-over-expressing cells revealed wide-ranging changes that were predominantly driven by the presence or absence of exogenous FOXC1 (Fig. 4E and F). Thus, a tree diagram based on Euclidean distances highlighted the major separation of samples depending on FOXC1 over-expression (Fig. 4E) which drove sample separation along PC1 on a PCA graph (Fig. 4F). Exposure to the steroid hormones progesterone and estrogen, together with cAMP (“PEc”) separated samples along PC2, whereas the presence or absence of exogenous SIX1 had only minor effects on global gene expression profiles (Fig. 4F). Intersecting the FOXC1- and SIX1-driven transcriptomic changes in Ishikawa cells with the DE gene lists of aging uteri highlighted that the uterine aging signatures were primarily driven by FOXC1: About one-third of A1 DE genes and half of all A2 and A3 DE genes overlapped with genes dysregulated in FOXC1-overexpressing cells (Fig. 4G). This result did not shift significantly by the concomitant presence of SIX1 or the sex steroid hormones prominent at the time of implantation and during pregnancy (Fig. 4G and Supplementary Fig. S5F). These results demonstrated that Foxc1 hyperactivation represents one of the earliest changes in aging mouse uteri and that FOXC1 has a major impact on global gene expression profiles in endometrial epithelial cells.

FOXC1 up-regulation demarcates the onset of aging in the uterine epithelial compartment

Leading on from these results, we aimed at determining the cell type-of-origin of FOXC1-driven genome regulatory changes. For this purpose, we first overlaid the genes that were up-regulated in age groups A1 or A2 onto two previously published single-cell RNA-sequencing (scRNA-seq) data sets from mouse uteri [46, 57]. After careful re-annotation of cell cluster identity, we found that most of our de-regulated genes fell within the epithelial cell cluster, with fewer genes also present in the endothelial cluster (Fig. 5A, and Supplementary Fig. S6A and B). This distribution was in stark contrast to the evenly distributed cell cluster overlap observed with random gene lists (Fig. 5A and Supplementary Fig. S6B). Despite the fact that GO term analyses did not highlight a specific epithelial-related functional enrichment, the predominant overlap of DE genes with the epithelial cell cluster was observed even with the genes de-regulated in A3 (Supplementary Fig. S6B and C). We also performed an estimation of cell type proportions on our bulk RNA-seq data based on single cell sequencing data using the bisque script (Supplementary Fig. S6D and E) [20, 57, 58]. While this analysis showed an increase in the proportion of epithelial cells in A3, in line with the uterine epithelial hyperplasia phenotype characteristic of females of this age group, the fraction of epithelial cells in A1 and A2 was statistically unchanged from Y (Supplementary Fig. S6D and E), thus ruling out that the epithelial cell cluster enrichment of DE genes we noted in A1 and A2 was driven by a bias in cell type proportions.

Figure 5.

FOXC1 up-regulation demarcates the initiation of aging in the epithelial compartment. (A) UMAP plots of re-analyzed single-cell RNA-sequencing data of the mouse uterus [46] identifies four distinct clusters composed of myocytes, endometrial stromal, epithelial, and perivascular endothelial cells. Super-imposition of genes DE in A1 and/or A2 uterine age groups highlights a strong enrichment in the epithelial cell cluster, and a minor enrichment in the perivascular endothelial cell cluster. Random gene lists serve as a control. (B) Immunofluorescence staining for FOXC1 (green) and epithelial cell marker CDH1 (red) on uterine cross-sections over an aging time course. FOXC1 is gradually and progressively up-regulated in glandular (highlighted by filled arrowhead in A1) and luminal (highlighted by open arrowhead in A1) epithelial cells with age. The position of the uterine lumen is indicated (lu), all other CDH1-enclosed spaces are uterine glands. Images are representative of n ≥ 3 samples per age group. (C) Quantification of FOXC1 immunofluorescence staining intensities (in arbitrary units, AU) in glandular (GE) and luminal (LE) epithelial cells across a time course of female aging, illustrating the progressive up-regulation of FOXC1 protein abundance. For simplicity, statistical significance is only shown for the pair-wise comparisons of GE and LE in the aged 1 (25 and 32 weeks of age) and Y→A1 intermediate sample aged 22 weeks (wks). All pair-wise comparisons of 8, 12, 18, and 22 wks LE are significantly different from 22 wk GE and all older samples with at least P< 0.05. Individual datapoints of nuclear intensities are shown as colored dots, with the horizontal black bar depicting the mean ± S.E.M. Statistical significance was calculated by one-way ANOVA with Tukey’s multiple comparisons test; *P< 0.05, ****P< 0.0001. (D) Left: Immunofluorescence staining of young and aged (aged 2) uterine cross-sections highlighting the appearance of single FOXC1-positive cells in the stromal cell compartment of aged samples, in addition to the FOXC1 up-regulation in glandular and luminal epithelial cells demarcated by CDH1. Right: Double immunofluorescence staining with FOXC1, endothelial cell marker CD31, and perivascular mural marker NG2 (a.k.a. CSPG4, commonly considered a pericyte marker) identifies the individual FOXC1-bright cells in the stromal cell compartment as perivascular mural cells. Images are representative of n = 3 samples per age group. (E) Confocal images of H3K27ac staining on uteri of young and aged (aged 2) females shows higher staining intensity in glandular epithelial (GE) cells (arrowheads), compared with luminal epithelium (LE, arrows) and uterine stromal (STR) cells. Images are representative of n = 3 samples per age group. (F) Quantification of H3K27ac immunostaining signals. Data points of individual area measurements are shown, with the horizontal bar depicting the mean and the error bars S.E.M. Statistical analysis was performed with one-way ANOVA; ****P< 0.0001, ns = not significant. LE, luminal epithelium; GE, glandular epithelium; STR, stromal cells.

We then investigated the cell type-specific expression pattern of FOXC1 in the aging mouse uterus by immunostaining samples capturing the onset of the reproductive decline (Fig. 1A and B), i.e. spanning an age range from 8 wks (Y) to 41 wks (A2). Virtually no FOXC1 staining was detectable in uteri of 8-wk-old females (Fig. 5B and C). However, we observed a gradual up-regulation of FOXC1 across the time course (Fig. 5B and C), with few FOXC1-bright cells becoming obvious within the stromal compartment of the endometrium at 18 wks of age. Glandular and luminal epithelial cells also exhibited a faint signal at 18 wks, but these staining intensity changes remained statistically insignificant (Fig. 5B and C). Strikingly, the epithelial FOXC1 signal intensified dramatically as aging progressed, with the first significant enrichment observed in the glandular epithelium at 22 wks of age. All samples A1 and older exhibited significantly higher FOXC1 abundance compared to Y in both epithelial compartments, with a more pronounced enrichment in glandular epithelial cells, in particular in A1 (Fig. 5B and C). Using a series of double-immunostainings, we determined that the FOXC1-positive single cells within the endometrial stroma were perivascular mural cells, as they surrounded CD31-positive vascular endothelial cells and stained positive for CSPG4 (also known as NG2)—a characteristic marker of pericytes (Fig. 5D). This cell type association was also corroborated in the single-cell data overlay (Supplementary Fig. S6B). To test whether this up-regulation of FOXC1 in perivascular cells translated into an overall increase in the abundance of pericytes, we assessed a larger set of perivascular niche markers [59] in our bulk RNA-seq data but found no evidence for an overall increase in the expression of genes specifically associated with pericytes or endothelial cells (Supplementary Fig. S6F). Thus, either existing pericytes start to mis-express FOXC1 or the perivascular niche becomes biased in favor of a FOXC1-positive subset of cells. Overall, the distribution of FOXC1 immunostaining signals in the aging uteri was fully in line with the mapping of DE genes onto the epithelial and endothelial/perivascular cell clusters (Fig. 5A, and Supplementary Fig. S6B and C). We also stained young and aged uteri for H3K27ac and observed a specific accumulation of this mark in uterine epithelial cells, in particular of uterine glands (Fig. 5E and F). By contrast, the stromal cell compartment tended toward lower H3K27ac levels, in line with the Western blot data of bulk tissue which is proportionately dominated by stromal cells (Supplementary Fig. S3E and F) Taken together, these findings indicated that aging affected the epithelial portion of the uterus, and specifically the glandular epithelium, the most, with an increasingly pronounced up-regulation of FOXC1 coupled to increased H3K27ac levels as aging progressed.

Endometrial epithelial organoids from aged females recapitulate Foxc1 up-regulation and exhibit overgrowth in vitro

The ability to derive and grow mEEOs provided a valuable opportunity to assess the impact of maternal age on the uterine epithelial compartment in greater detail. First, we investigated the efficiency of mEEO formation from young and aged uteri and confirmed that organoids can be successfully derived and grown from uteri across the age spectrum at similar rates (Fig. 6A).

Figure 6.

Uterine epithelial organoids display age-associated hyperproliferation and elevated progenitor marker expression. (A) Representative images of organoids established from endometrial epithelial fragments isolated from young and aged (aged 3) mice. Epithelial fragments self-organize into organoids within 24 h, then continue to expand with further culture. (B) Experimental design for RNA-sequencing of mEEOs derived from young (n = 6) and aged (n = 6, of which A2: n = 5 and A3: n = 1) mice cultured for 2, 10, and 14 days.(C) Foxc1 expression in young and aged mEEOs at the three sampled time points. Elevated Foxc1 expression levels with uterine age are recapitulated in the cultured mEEOs. Data points represent biological replicates (i.e. organoids derived from different females) and are shown as mean ± S.E.M. Statistical significance was calculated using two-way ANOVA with Šídák’s multiple comparisons test; ****Padj < 0.0001, ***Padj = 0.0004, *Padj = 0.0079. (D) Immunofluorescence staining of young and aged mEEOs for FOXC1 (green) and proliferation marker Ki67 (red). Images are representative of n = 3 biological replicates. (E) Immunofluorescence staining for H3K27ac on mEEOs from young and aged (A2) females. Images are representative of n = 3 biological replicates. (F) Images of mEEOs from young and aged (A2) females after 14 days in culture, quantified in (G). Images are representative of n = 6 biological replicates. (G) Quantification of organoid sizes. Each data point represents a single organoid measurement from young (n = 39 organoids derived from three females) or aged (n = 64 organoids derived from 3 A2 females) mEEOs grown for 14 days. Statistical significance was calculated using Mann–Whitney test; **P = 0.0055. (H) Proliferation curve of Ishikawa cells stably transfected with empty vector or with a FOXC1 over-expression plasmid over a 4-day time course. Statistical analysis was performed with unpaired two-tailed t-test of the areas under curve; **P = 0.009. (I) Quantification of the percentage of proliferation marker PHH3-positive Ishikawa cells stably transfected with empty vector (EV) or with a FOXC1 over-expressing (FOXC1 o/e) plasmid. Statistical analysis was performed with unpaired two-tailed t-test; ***P = 0.0007. (J) Sox9 expression in mEEOs from young and aged females as described in (B), shown as log₂(RPM + 1) values derived from RNA-sequencing data. Data are shown as mean ± S.E.M. Statistical significance was calculated using unpaired two-tailed t-test; **P = 0.0035. (K) Immunofluorescence staining of young and aged (A2) mEEOs for uterine glandular progenitor marker SOX9, illustrating the upregulation of SOX9 in aged organoids. Images are representative of n = 3 samples per age group. (L) Diagram of the changes in pioneer transcription factor FOXC1 and SOX9 expression between young and aged mEEOs, and the functional consequences that likely ensue as a result.

To gain more detailed insights into the global transcriptional changes that occur in epithelial organoids derived from aged uteri, we performed RNA-sequencing on young (n = 6) and aged [n = 6, of which A2: n = 5 (44–51 wks) and A3: n = 1 (53 wks)] mEEOs that were cultured in vitro for 2, 10, and 14 days post-isolation (Fig. 6B). We identified 908 DE genes in the 2-day cultured mEEOs, of which 41 overlapped with uterine DE genes. Setting the uterine A1&A2 DE genes as denominator, this translated into 11% shared up-regulated genes, but only 1.5% shared down-regulated genes, emphasizing the greater consistency of genes that become aberrantly induced in uterine epithelial cells with age (Supplementary Fig. S7A). However, we noticed that the transcriptional divergence of young and aged mEEOs diminished over the 14 day period as Y and A mEEOs became more transcriptionally similar to one another and acquired a “culture-specific” phenotype (Supplementary Fig. S7B and C). This notwithstanding, FOXC1 mRNA and protein was consistently and significantly up-regulated in mEEOs from aged females, thus recapitulating our observations in uterine tissue (Fig. 6C and D). Treatment of 10-day cultured mEEOs with the hormonal PEc cocktail caused Foxc1 up-regulation, but this was even more pronounced in mEEOs from aged females (Supplementary Fig. S7D). We also found that mEEOs from aged females exhibited consistently strong H3K27ac staining, whereas those from young females generally stained weakly for H3K27ac with only a few organoids exhibiting pronounced H3K27ac intensity (Fig. 6E). Since most organoids are derived from the glandular epithelium and fewer from the luminal epithelium, and isolated epithelial fragments preferentially form organoids by closing in on themselves, this pattern matches well the H3K27ac immunostaining in uteri that shows an increase of H3K27ac specifically in the glandular epithelial cells (Fig. 5E and F).

When observing mEEOs from Y and A females over the culture period, those from aged females appeared consistently larger (Fig. 6A and F). This was confirmed by quantifying mEEO sizes at day 14 (Fig. 6G). We then sought to determine whether FOXC1 up-regulation caused increased cell proliferation rates, as is the case in the context of various cancers [60–63]. We stained mEEOs for the proliferation marker Ki67 and found that those from aged females exhibited more Ki67-positive cells, but the comparatively small number of cells in organoids did not allow for a robust quantitation of this effect (Fig. 6D). Thus, to gain robust insights into the impact of FOXC1 on cell proliferation, we performed proliferation time course and phospho-histone H3 (PHH3) staining experiments on FOXC1 over-expressing compared to empty vector (EV)-control Ishikawa cells. These experiments established that elevated FOXC1 levels indeed caused endometrial epithelial cell hyperproliferation (Fig. 6H and I). This finding provided a molecular explanation for the epithelial hyperplasia phenotype with frequently cystic appearance that is characteristic of aging mouse uteri (Supplementary Fig. S8A).

Focusing on the RNA-seq data from day 2 mEEOs as the most robust representation of the epithelial aging signatures, we observed that amongst the 15 genes with ≥10-fold higher expression was Sox9, a well-known uterine epithelial cell progenitor marker associated with uterine gland hyperproliferation (Fig. 6J) [64]. Staining of mEEOs confirmed the dramatic increase in SOX9-positive cells in aged mEEOs (Fig. 6K). We also observed increased SOX9 staining intensities in luminal and glandular epithelial cells of aged female uteri (Supplementary Fig. S8B and C). Other glandular epithelial hallmark genes associated with a progenitor-like state were also up-regulated in aged mEEOs, notably Foxa2 and Prom1 (Supplementary Fig. S8D). Moreover, Six1, but not Zfhx4, was up-regulated in mEEOs as the transcription factor we had identified in the early time points of aging uterine tissue. Overall, these results indicated that endometrial aging is associated with increased expression of FOXC1 which causes a pro-proliferative phenotype that is exacerbated by profound up-regulation of SOX9, impacting the uterine epithelial compartment.

Discussion

A reduced capacity of the uterus to support a successful pregnancy constitutes the main reason for the reproductive decline observed in female mice of AMA [21, 23, 65]. Previous studies have shown that one-year-old females display a pronounced deficit in initiating the decidualization response triggered by the action of the steroid hormones, estrogen and progesterone [21, 26]. In particular, progesterone responsiveness is most affected, creating a prolonged pro-estrogenic bias that manifests, for example, in a delayed shift of cell proliferation from the uterine luminal epithelium into the stroma [21, 66, 67]. Decidual stromal cells of aged females also exhibit extensive epigenetic changes that may underpin these transcriptional and functional deficiencies [27, 31]. However, what has remained unknown is whether these age-related changes occur abruptly at a certain age or whether they accumulate over time, and whether the stromal cell compartment of the uterus is the main driver or merely the affected bystander of these age-associated changes.

In the current study, we find that maternal aging is associated with a progressive increase in transcriptional variation that is accompanied by an accumulation of activating histone marks at a multitude of genomic loci. These changes begin to manifest between 20 and 24 wks (5–6 months) of age and become progressively more pronounced in older females. Importantly, the transcriptional signatures associated with uterine aging differ substantially from common senescence markers associated with organismal aging, highlighting the unique character of the decidualizing endometrium and the early onset of reproductive aging compared with organismal aging [68]. Thus, except for the age-associated hallmark genes Cdkn2a/b, uterine aging is globally characterized by a loss of transcriptional fidelity that manifests as a profound increase in transcriptional variability, with only a small subset of genes overlapping between all age groups analyzed. These transcriptomic changes are accompanied by a drastic increase in the active histone modifications H3K4me3 and H3K27ac at dozens of genomic sites in the early age groups A1 and A2 (i.e. between 24 and 52 wks of age). Intriguingly, a recent study reported a global loss of H3K27ac associated with reduced progesterone receptor (PGR) expression in the aging endometrium of humans and mice [69]. While we, too, have reported reduced PGR expression in aging uteri and stromal cells [21, 27] and observe a trend toward a global reduction of H3K27ac levels that appears to stem from the stromal compartment (Fig. 5F and Supplementary Fig. S3D), our genome-wide RNA-seq and ChIP-seq data collectively point toward a loss of transcriptional fidelity, a propensity toward gene activation, and an accumulation of H3K27ac and H3K4me3 marks at many genomic sites. A trend toward gene activation was also observed in our previous study on uterine stromal cells from aged uteri [27]. Our data corroborate the notion that the fine-tuning of H3K27ac distribution may be of particular importance in the peri-implantation window, when many decidualization genes need to become activated [69]. Of note, ChIP-seq enrichment at individual sites may diverge from global histone modifications levels, a notion that closely resonates with the well-known opposing behavior of global versus locus-specific DNA methylation levels.

Our data show an overall correlation of H3K27ac levels at gene promoters with gene expression (Fig. 3J). Moreover, deconvolving our bulk RNA-seq data from uterine tissues by projection onto single cell sequencing data [46] highlights that maternal age-related effects appear to emanate from the epithelial compartment during early time points of aging. This finding is corroborated by recent single-cell sequencing analysis of aging female reproductive tissues in which an over-representation of uterine glandular epithelial cells was also observed [57]. Specifically, we find that these changes entail a dramatic up-regulation of the pioneer transcription factor FOXC1, which is underpinned by a profound hyper-enrichment for H3K27ac and H3K4me3 across this gene locus. Using endometrial epithelial organoids, we show that FOXC1 activation is recapitulated in vitro and goes hand-in-hand with the up-regulation of the endometrial gland progenitor marker SOX9, suggesting a reversal of glandular epithelial cell maturity to a more immature, proliferative state (Fig. 6L). This is a remarkable and unanticipated observation, as aging would not otherwise be associated with reprogramming-like events. Collectively, our data show that up-regulation of FOXC1 is one of the earliest hallmarks of uterine epithelial cell aging. Aberrant induction of FOXC1 initiates a cohort of transcriptional changes that are associated with the age-associated functional decline.

The increase in the active histone modifications H3K27ac and H3K4me3 from about 6 months of age onward is an intriguing finding of our study. This elevated abundance was more pronounced for H3K27ac compared to H3K4me3, and identified Foxc1 as the most affected locus where the gain in H3K27ac was already highly significant in A1 and clearly preceded that of H3K4me3 (significant from A2 on). This finding aligns with studies showing that the ectopic introduction of H3K27ac in promoter regions leads to H3K4me3 enrichment around transcription start sites and triggers transcriptional activation [70, 71]. Thus, the hyper-enrichment of H3K27ac may constitute a key driver of Foxc1 activation, underpinning the gene regulatory changes associated with uterine epithelial cell aging.

Despite intensive interrogation of our data against a large cohort of epigenetic repressors including histone deacetylases, H3K27 demethylases (Supplementary Fig. S4C) and epigenetic modifiers associated with H3K27ac deposition or accessibility (Supplementary Fig. S8E), we could not find evidence for the de-regulation of any of these factors, at least on the transcriptional level. More subtle changes in expression may be obscured by the opposing global enrichment trends of H3K27ac in epithelial glands and stromal cells. Nevertheless, this prompts the question of how the increase of H3K27ac at Foxc1 comes to pass. Here, it is intriguing to note that several highly unusual characteristics of Foxc1 might make this locus particularly prone to dysregulation. Foxc1 is an intron-less gene characterized by an unusually high GC content of ∼70%, and is devoid of DNA methylation in uteri of both, young and aged females [31]. In experiments where sequences of different GC content were placed into an ectopic gene locus, extremely GC-rich sequences were shown to gain H3K27ac [72]. Moreover, we find in our data that the FOX-binding motif is specifically enriched in super-enhancers that gain H3K27ac. Previous studies have found that super-enhancers are significantly more GC-rich than typical enhancer elements, indicating that GC richness may have important roles in the formation of super-enhancers [73]. Finally, Foxc1 is associated with the most significantly enriched binding site for estrogen receptor (ESR1) and is the most up-regulated gene in response to estradiol exposure in mouse and human osteoblasts [74, 75]. This finding is highly relevant to uterine aging which is associated with a strong pro-estrogenic bias due to the loss of progesterone responsiveness. Thereby, the “unopposed” estrogen action in the aging uterus may lead to a specific propensity for Foxc1 activation. Thus, the conjunction of hormonal signaling imbalance, the highly unusual DNA sequence composition of the Foxc1 locus and the pioneer transcription factor role of FOXC1 at super-enhancers appear to make Foxc1 a key sensor of early aging. Foxc1 up-regulation establishes a self-reinforcing positive feedback loop together with SOX9, another pioneer transcription factor that has been shown to physically interact with FOXC1, leading to global genome activation [76].

Using the human endometrial Ishikawa cell line, we demonstrate that FOXC1 overexpression causes profound transcriptional shifts with some 10000 de-regulated genes. This effect far outweighs that of other genes also mis-expressed at the onset of uterine aging, such as Six1. Nevertheless, de-regulation of Six1 corroborates the predominant impact of early aging on the epithelial compartment, as SIX1 is an indicator of human endometrial health which regulates aberrant endometrial epithelial cell differentiation and cancer latency [77]. Indeed, FOXC1 has emerged as one of the key targets in numerous cancers including those of the human endometrium, where elevated FOXC1 levels are generally associated with higher proliferation rates and poor prognosis [78, 79]. This general correlation directly resonates with our data from the mouse uterus, as we find that endometrial epithelial organoids from aged females grow larger in size, and that the luminal epithelium of the uterus of older female mice exhibits far more involutions than that of young females [21, 24]. Notably, FOXC1 has also emerged as an important cell-intrinsic regulator of adult stem and progenitor cell function across multiple tissues, with higher FOXC1 levels inducing de-differentiation and the reversion to a progenitor state of previously differentiated cells [79, 80]. This appears to occur very similarly in the uterus, demonstrated by the marked up-regulation of genes indicative of a more stem- or progenitor-like state of uterine glandular epithelial cells, including SOX9 mRNA and protein, Foxa2 and Prom1 [20, 35–37, 64, 81–83]. Up-regulation of these factors is also associated with uterine epithelial hyperplasia [64, 82]. Perhaps most compellingly, the pathological transformation of the epithelial compartment is highlighted by the striking resemblance of aged uteri to those over-expressing Sox9 experimentally, which show the same epithelial hyperplasia phenotype even at a young age [64].

Collectively, we demonstrate that the uterine epithelial compartment is a key sensor of maternal aging, and that FOXC1 up-regulation in these cells is amongst the first and most robust changes in the epigenetic and transcriptional landscape, triggering a cascade of downstream alterations to uterine epithelial cell function, including increased cell proliferation. Due to the reliance of uterine stromal cell decidualization on the continuous crosstalk with uterine glands, these early epithelial defects propagate to the uterine stroma and ultimately result in the progressive decline in reproductive performance in female mice with age.

Supplementary data

Supplementary data is available at NAR Molecular Medicine online.

Conflict of interest

The authors declare that there is no competing conflict of interest that could be perceived as prejudicing the impartiality of the study reported.

Funding

This work was supported by Canadian Institutes of Health Research (CIHR) project grant PJT-174982 to M.H. and W.D., by BBSRC-NC3R ageing project grant BB/S002995/1 to H.C. and M.H., by a Tier I Canada Research Chair in Developmental Genetics and Epigenetics (CRC-2018-00 240) to M.H., by the UK Biotechnology and Biological Sciences Research Council grant BBS/E/B/000C0423 to G.D.K., by a Natural Sciences and Engineering Research Council of Canada (NSERC) Discovery Grant RGPIN-2021-02 417 to M.H., and by funds of the Alberta Children’s Hospital Research Institute grant ACHF22-0911 to M.H. and W.D.

Data availability

All primary sequencing data have been deposited in the GEO repository under accession numbers GSE296745 (uteri and organoid RNA-seq), GSE296787 (Ishikawa cell RNA-seq) and GSE298816 (uteri ChIP-seq). Single cell deconvolution code is available at doi.org/10.5281/zenodo.16782162.