The study was approved by the NHS National Research Ethics – Hammersmith and Queen Charlotte’s and Chelsea Research Ethics Committee (1997/5065). Endometrial biopsies from 12 patients were obtained in the Implantation Clinic, a specialized research clinic at University Hospitals Coventry and Warwickshire National Health Service Trust. Written informed consent was given by all participants in accordance with The Declaration of Helsinki 2000. Endometrial biopsies, timed 6 to 10 days following the preovulatory luteinizing hormone surge, were taken using a Wallach EndocellTM sampler (Wallach, Trumbull, United States) following a transvaginal ultrasound scan to exclude overt uterine pathology. Further, none of the participants had been prescribed hormonal treatment in at least 3 months prior to biopsy. Demographic details are summarized in Supplementary Table 1.

Single-cell suspensions of EnSC were isolated from 12 midluteal biopsies as described previously (Masuda et al., 2012). In short, samples were collected in DMEM/F-12 with 10% dextran coated charcoal activated fetal bovine serum (DCC-FBS), finely chopped, and enzymatically digested with deoxyribonuclease type I (0.1 mg/mL; Roche, Burgess Hill, United Kingdom) and collagenase (0.5 mg/mL; Sigma-Aldrich, Gillingham, United Kingdom) for 1 h at 37°C. The cells were passed through a 40 μm cell strainer (Fisher Scientific, Loughborough, United Kingdom), which retained epithelial cells more resistant to enzymatic digestion. Ficoll-Paque PLUS (GE Healthcare, Little Chalfont, United Kingdom) was utilized to remove erythrocytes from the stromal cell fraction. Subsequently, SUSD2+ and SUSD2− cells were isolated using magnetic bead sorting, as detailed previously (Masuda et al., 2012). Briefly, up to 1 × 10⁶ EnSC/100 μL of Magnetic Bead Buffer (0.5% BSA in PBS) were combined with 5 μL/1 × 10⁶ cells phycoerythrin (PE) conjugated anti-human SUSD2 antibody (BioLegend, London, United Kingdom). After 20 min on ice, approximately 1 × 10⁷ cells/80 μL of Magnetic Bead Buffer were incubated with anti-PE-magnetic-activated cell sorting MicroBeads (20 μL/1 × 10⁷ cells; Miltenyi Biotec, United Kingdom) for another 20 min. Cell suspensions (up to 1 × 10⁸ cells/500 μL of Magnetic Bead Buffer) were applied onto MS columns (Miltenyi Biotec) in a magnetic field, and washed multiple times with 500 μL of Magnetic Bead Buffer. SUSD2− cells passed freely through the column, whereas magnetically labeled SUSD2+ cells were retained. Following removal from the magnetic field, SUSD2+ cells were eluted from the columns with 1 mL of Magnetic Bead Buffer. Purified SUSD2− and SUSD2+ cells were expanded in DMEM/F12 with 10% DCC-FBS, 1% antibiotic-antimycotic solution (Invitrogen), 1% L-glutamine (Invitrogen), estradiol (1 nM; Sigma-Aldrich), insulin (2 μg/mL; Sigma-Aldrich), and basic fibroblast growth factor (10 ng/mL; Merck Millipore, Watford, United Kingdom). Decidualization experiments were carried out at passage 2. Briefly, paired SUSD2− and SUSD2 + cells were seeded 6-well plates at a density of 2 × 10⁵ cells/well and decidualized when confluent with 0.5 mM 8-bromoadenosine cAMP (8-bromo-cAMP; Sigma-Aldrich) and 1 μM medroxyprogesterone acetate (MPA; Sigma-Aldrich) in phenol red-free DMEM/F12 containing 2% DCC-FBS, 1% antibiotic-antimycotic solution. The spent medium was collected every 48 h for analysis, stored at −80°C and the differentiation medium refreshed. Decidual status was confirmed using transcript levels of the decidual marker gene PRL.

Immunostaining was completed at passage 1. Cells were fixed with 4% paraformaldehyde (Sigma-Aldrich) for 10 min and blocked with 3% BSA or 5% normal goat serum in PBS for 30 min. Cells were then incubated with an anti-human VAP-1 antibody (HPA000980, 1:100; Sigma Aldrich) for 1 h at room temperature, followed by incubation with Alexa Fluor 488 anti-rabbit secondary antibody for 1 h at RT (1:1000; Thermo Fisher Scientific). Slides were mounted with VECTASHIELD anti-fade mounting media with DAPI (Vector Laboratories) to visualize the nuclei. Images were captured using a Zeiss Laser Scanning Microscope LSM 510 at 60× magnification.

Total RNA was extracted from EnSC and PVC cultures using RNA STAT-60 (AMS Biotechnology). Equal amounts of total RNA (1 μg) were treated with DNase and reverse transcribed using the QuantiTect Reverse Transcription Kit (Qiagen), and the resulting cDNA was used as template in RT-qPCR analysis. Detection of gene expression was performed with Power SYBR Green Master Mix (Thermo Fisher Scientific), and the 7500 Real-Time PCR System (Applied Biosystems). The expression levels of the samples were calculated using the ΔΔCt method, incorporating the efficiencies of each primer pair. The variances of input cDNA were normalized against the levels of the L19 housekeeping gene. All measurements were performed in triplicate. Melting curve analysis confirmed amplification specificity. Primer sequences used were as follows: AOC3, forward: TCC TGT GCC AGG ACT CTCTT and reverse CAA GGT TCA GTG TCC CCT GT; L19, forward: GCG GAA GGG TAC AGC CAA T and reverse: GCA GCC GGC GCA AA; PRL, sense 5′-AAG CTG TAG AGA TTG AGG AGC AAA C-3′, PRL antisense 5′-TCA GGA TGA ACC TGG CTG ACT A-3′.

For untargeted exometabolome analysis, samples were prepared as described previously (Zhu et al., 2011; Cui et al., 2013) with some modifications. Briefly 50 μL of EnSC and PVC conditioned media were thawed at 4°C, and quality control (QC) samples were prepared by mixing an equal amount of conditioned media from each of the samples. Both QC and the conditioned media from PVC and EnSC were processed in the same way. Proteins were precipitated with 200 μL ice-cold methanol containing 10 μg/mL 9-fluorenylmethoxycarbonyl-glycine as an internal standard (ISTD), and vortexed. Following centrifugation at 16,000 rpm for 10 min at 4°C, the supernatant was collected and evaporated until dry in a speed vacuum evaporator then resuspended in 200 μL of water/methanol (98:2; v/v) for liquid chromatography-mass spectrometry (LC-MS) analysis. All samples were kept at 4°C and analyzed within 48 h. The sample run order was randomized to remove batch effects and the QC samples were analyzed after every eight samples to monitor the stability of the system.

Untargeted exometabolomics was performed as previously described (Cui et al., 2017). The prepared conditioned media was analyzed using Agilent 1290 ultrahigh-pressure liquid chromatography system (Waldbronn, Germany) equipped with a 6520 QTOF mass detector managed by a MassHunter workstation. The oven temperature was set at 45°C.

The 5 μL of injected sample was separated using an Agilent rapid resolution HT Zorbax SB-C18 column (2.1 × 50 mm, 1.8 mm; Agilent Technologies, Santa Clara, CA, United States), 0.4 mL/min flow rate, and a gradient elution involving a mobile phase consisting of (A) 0.1% formic acid in water and (B) 0.1% formic acid in methanol. To start the mobile phase was set at 5% B, with a 7-min linear gradient to 70% B, then a 12 min gradient to 100% B. This was held for 3 min then returned to 5% B in 0.1 min.

Both positive and negative electrospray ionization were used to collect mass data between m/z 100 and 1000 at a rate of two scans per second. The ion spray was set at 4,000 V, and the heated capillary temperature was maintained at 350°C. The drying gas and nebulizer nitrogen gas flow rates were 12.0 L/min and 50 psi, respectively. Two reference masses (m/z 121.0509 (C5H4N4) and m/z 922.0098 (C18H18O6N3P3F24) were continuously infused to the system to allow constant mass correction during the run.

Adenosine, inosine, hypoxanthine, and xanthine were measured and analyzed by targeted LC-MS/MS analysis. The analysis was conducted as described previously with some modifications (Garcia-Manteiga et al., 2011). Briefly, LC-MS analysis was performed with Agilent 1290 ultrahigh-pressure liquid chromatography system (Waldbronn, Germany) coupled to an electrospray ionization with iFunnel Technology on an Agilent 6490 triple quadrupole mass spectrometer.

The auto-sampler was cooled at 4°C and 2 μL of injected sample was chromatographically separated using an Atlantis HILIC column (2.1 × 100 mm; Waters, Eschborn, Germany). The mobile phases were (A) 10 mM ammonium formate and 0.1% formic acid in water and (B) 0.1% formic acid in acetonitrile. Initially, 100% was utilized for 2 min, then reduced to 80% in a linear gradient for 11 min, and to 40% B over 1 min. This was held for 5 min and then the mobile phase was returned to starting conditions over 6 min. The column was kept at 45°C and the flow rate was 0.4 mL/min. Direct infusion of individual standard solutions allowed optimization of both the mass transition and collision energy for each compound by direct infusion. Both positive and negative electrospray ionization modes were performed with the following source parameters: drying gas temperature at 250°C with a flow of 14 L/min, sheath gas temperature at 400°C with a flow of 11 L/min, nebulizer gas pressure at 40 psi, capillary voltage 4,000 V and 3,500 V for positive and negative mode, respectively, and nozzle voltage 500 V for both positive and negative modes.

Raw spectrometric data in untargeted metabolomics were analyzed by MassHunter Qualitative Analysis software (Agilent Technologies, United States). The Molecular Feature Extractor algorithm was used to obtain the molecular features, employing the retention time (RT), chromatographic peak intensity and accurate mass as input. Next, the features were analyzed by MassHunter Mass Profiler Professional software (Agilent Technologies, United States). Here, only features detected in at least 80% of the samples at the same sampling time point signal with an intensity ≥20,000 counts, three-times the limit of detection of the LC-MS instrument, were kept for further processing. Pre-statistical filtering of samples was performed, including the removal of samples with more than 60% missing data. One sample had 62% missing data and was subsequently removed from the untargeted analysis. The tolerance window for alignment of RT and m/z values was set at 0.15 min and 2 mDa, respectively, and the data normalized to the 9-fluorenylmethoxycarbonyl-glycine ISTD spike.

Raw spectrometric data in targeted metabolomics were processed using MassHunter Workstation Quantitative Analysis software (Agilent Technologies, United States). Missing values were imputed using half the lowest value. For both untargeted and targeted analysis, metabolites that were differentially expressed were inputted into the Metaboanalyst Pathway Analysis Tool to identify pathways that were altered during decidualization or between PVC and EnSC. This incorporates both pathway topology and enrichment analysis to output pathways that have changed significantly, and the pathway impact (Chong et al., 2019). The hypergeometric test was used and relative-betweenness centrality for over-representation analysis and pathway topology analysis, respectively. The pathway impact is calculated by the number of compounds that are significantly altered in relation to the total number of compounds in the pathway (Chong et al., 2019).

For statistical analysis, non-parametric Test (Wilcoxon, Mann–Whitney test) with Holm-Sidak Multiple Testing Correction was employed and statistical significance was set at p < 0.05. For multivariate data analysis using hierarchical clustering or partial least squares regression (PLSR) analysis, data were normalized by applying log2, median-centering and dividing by standard deviation. Unsupervised Euclidean distance hierarchical clustering was performed using HemI (Deng et al., 2014). In addition, fold change (FC) analysis was performed to further filter the features and only those features with FC > 1.5 were selected as potential significantly altered metabolites across decidualization.

The identification of the differential metabolites structure was based on our published work (Cui et al., 2017). Briefly, MassHunter software (Agilent) was used to calculate the elemental compositions of the metabolites based on their exact mass, the isotope pattern, and the nitrogen rule. The elemental composition in combination with the exact mass were used in open-source database searches, including HMDB¹, LIPIDMAPS², MassBank³, and METLIN⁴. Next, MS/MS experiments were performed to obtain structural information from the fragmentation pattern of the metabolite and these MS/MS spectra were searched and compared to compounds in the databases. Finally, the metabolites were confirmed by comparison with the standards where commercially available. The metabolites meet the minimum reporting standards for chemical analysis in metabolomics recommended by Metabolomics Standard Initiative (MSI) (Sumner et al., 2007).

The endometrial stromal fraction of 12 midluteal biopsies were separated into SUSD2+ PVC and SUSD2− EnSC by magnetic-activated cell sorting (MACS) (Figure 1). These cells were tested for AOC3/VAP-1 expression, as we have previously reported that AOC3 is highly enriched in PVC cells (Murakami et al., 2014). AOC3 mRNA levels were significantly enriched in freshly isolated PVC cells when compared to their EnSC counterparts (p = 0.0052; Figure 1B). Immunofluorescence showed VAP-1 immunoreactivity on the cell surface of SUSD2 + cells whereas more discrete, punctate cytoplasmic staining was observed in EnSC (Figure 1C and Supplementary Figures 1, 2). Both subpopulations were cultured to confluency and then decidualized over 8 days. Conditioned culture media of paired decidualizing PVC and EnSC cultures were collected every 48 h and subjected LC-MS based metabolomics.

A total of 145 annotated secreted metabolites and 5 unidentified metabolites were detected in the conditioned media of both PVC and EnSC (Supplementary Table 2). There were no metabolites unique to either PVC or EnSC. Hence, the metabolite data of PVC and EnSC were first combined to construct a temporal map of metabolic footprints associated with decidualization.

The QC samples clustered together in principal component analysis (PCA) scores plots (Supplementary Figure 3), indicating good stability and reproducibility of the chromatographic separation throughout the entire sequence. PLSR analysis separated the decidual time points into distinct groups (Figure 2A). Regression coefficients were used to identify metabolites that shaped the progression of decidualizing cells throughout the time-course. Metabolites with the five highest β-coefficient (β) at each timepoint are depicted in Figure 2B. Phenylalanyl-tyrosine (β: 0.02), 2,3-dinor-8-iso prostaglandin F2α (2,3-dinor-8-isoPGF2α) (β: 0.01), and hypoxanthine (β: 0.02) were identified, respectively, as influencers on day 0 (undifferentiated cells), day 2 and day 4 of the decidual time-course. Adenosine thiamine triphosphate (AThTP) (β: 0.01) and hyaluronic acid (β: 0.01) were the highest influencers of day 8 of decidualization.

The exometabolome analysis of conditioned media revealed a marked temporal change in metabolic footprints upon decidualization. Out of the 150 detected compounds, the secreted levels of 79 metabolites changed by 1.5-fold (p < 0.05) over the 8-day decidual time-course (Figure 3A and Supplementary Table 3). Unsupervised hierarchical clustering showed a major bifurcation in exometabolome profiles between undifferentiated and decidualizing cells. Further, each decidual time-point was characterized by a unique footprint, which supports the notion that decidualization is a multistep differentiation process (Figure 3A).

Figure 3B highlights several regulated secreted metabolites, some of which are already implicated in decidualization. For example, the prostaglandin metabolite, 2, 3-dinor-8-isoPGF2α, which can be derived from PGE2, was undetectable in undifferentiated cultures but consistently upregulated in differentiating cells throughout the decidual time-course. Secreted levels of glucosylsphingosine, a sphingolipid metabolite, followed the same pattern, in keeping with the dependency of the decidual process on de novo sphingolipid synthesis (Mizugishi et al., 2007; Ding et al., 2018). Several metabolites were selectively secreted by the decidualizing cells on day 8, including 5-amino-3-oxohexanoate and hyaluronic acid (HA). HA prevents apoptosis of decidual cells through binding to its receptor CD44 (Takizawa et al., 2014). 20-trihydroxy-leukotriene-B4 and PE (18:3/14:1) are examples of metabolites whose extracellular concentrations declined upon decidualization, whereas phosphatidylcholine (PC) (16:0/16:0) and lactosamine exemplify metabolites exhibiting a biphasic secreted pattern, peaking on day 2 and day 4 of decidualization, respectively. A difference in molecular mass distinguishes endogenous cyclic AMP levels which increased upon decidualization from the 8-bromo-cAMP utilized to stimulate decidualization.

Metaboanalyst was used to identify metabolic pathways regulated upon decidualization (Chong et al., 2019). Twenty-one metabolic pathways were identified across the decidual time-course (Supplementary Figure 4). As shown in Figure 3C, prominent pathways enriched across the decidual time-course included valine, leucine, and isoleucine biosynthesis (VLIB), glycerophospholipid metabolism (GM), and phenylalanine, tyrosine, and tryptophan metabolism (PTTM). Pyrimidine metabolism (PM) was enriched on day 8 of decidualization, suggesting the requirement for pyrimidines as nitrogenous bases in DNA and RNA.

The temporal changes in the metabolic footprints of decidualizing PVC and EnSC were comparable but not identical, as demonstrated PLSR analysis (Supplementary Figure 5). Notably, apart from a single metabolite (N2, N2-Dimethylguanosine), no significant difference was observed in the exometabolomes of undifferentiated PVC and EnSC. By day 2 of decidualization, however, the secreted levels of 12 annotated metabolites were significantly higher in PVC compared to EnSC and 4 were lower (1.5 fold-change, p < 0.05; Supplementary Table 4). By day 4 and day 8 of decidualization, secreted levels of 6 and 9 metabolites, respectively, were significantly different between PVC and EnSC (Supplementary Table 4). Out of a total of 32 differentially secreted, annotated metabolites, 20 were more abundantly expressed by PVC. Figure 4A shows examples of metabolites differentially secreted between PVC and EnSC at each timepoint of the decidual pathway.

To further elucidate the metabolic differences between PVC and EnSC upon decidualization, pathway enrichment analysis was performed (Figure 4B). Interestingly, distinct metabolomic profiles emerged. The tricarboxylic acid cycle (TCA), and glycine, serine and threonine metabolism (GSTM) were enriched in PVC cells on day 2 of decidualization. Enrichment of sphingolipid metabolism (SM) arose at in PVC at day 4, whereas inositol phosphate metabolism (IPM) and the pentose phosphate metabolism (PPM) were more prominent in EnSC. Finally, on day 8, sphingolipid (SM) and purine metabolism (PuM) were enriched in PVC. Purine metabolism had the highest pathway impact and significance (Figure 4B and Supplementary Figure 6), suggesting a greater autocrine or paracrine requirement of purine signaling upon decidualization in PVC when compared to EnSC.

The enrichment in purine metabolism (Figure 5A) prompted us to investigate its wider metabolomic network in greater depth. We developed an analytically robust targeted mass spectrometry analysis (average coefficient of variation = 23.9%) to assess adenosine, inosine, hypoxanthine, and xanthine levels. Adenosine (7.8 fold-change, p = 0.0001), inosine (1.53 fold-change, p < 0.02) and xanthine (3.1 fold-change, p = 0.0001) were significantly higher in PVC on day 8 of the decidual time-course in comparison to undifferentiated cells whereas uridine (0.21 fold-change, p < 0.0001) and cytidine (0.42 fold-change, p = 0.001) were significantly lower. Furthermore, xanthine levels were significantly lower in PVC compared to EnSC conditioned media in both undifferentiated and decidualized cultures (Figure 5B and Supplementary Table 5).