Distinct endometrial protein profiles in spontaneous and stimulated cycles in women with poor ovarian response: A prospective case-crossover clinical trial.

The main role of the endometrium is to provide highly specific and precisely timed support for embryo implantation. For this purpose, the endometrium undergoes dynamic, hormone-regulated changes throughout the menstrual cycle. During most of the menstrual cycle, the endometrium remains unsuitable for embryo implantation. Dramatic physiological changes are required for the endometrium to become receptive and thus capable of accepting an embryo during a short period of the secretory phase of the menstrual cycle, referred to as the “window of implantation” (WOI) [ 1 ]. The duration of WOI is primarily determined by sex steroids, which regulate the expression of locally acting growth factors, transcription factors, cytokines, and chemokines. The transition of the endometrium to a receptive state is a highly coordinated process involving epithelial and stromal cell adhesion, trophoblast invasion, and immune modulation. Successful implantation requires accurate synchronization between the developing embryo and the receptive endometrium [ 2 ]. The field of assisted reproductive technologies has evolved considerably, with notable advancements in ovarian stimulation protocols and laboratory practices since its introduction. However, embryo implantation failure remains a major limiting factor in the success of in vitro fertilization (IVF) procedures. Although the quality of the embryo is considered to be an important determinant of successful implantation, impaired endometrial receptivity and dysregulated embryo-endometrial cross-talk are also responsible for implantation failure [ 3 ]. For this reason, numerous studies have focused on identifying markers of endometrial receptivity [ 4 ]. However, progress in developing reliable prognostic tests and treatments to enhance endometrial receptivity has been limited. One contributing factor is the lack of a clear definition of what constitutes a receptive human endometrium. Normal endometrium function relies on specific patterns of gene expression and downstream protein abundance. »Omics« techniques, which have the ability to determine changes in different molecular compartments (e.g., genomics, proteomics, metabolomics), have enabled a better understanding of endometrial physiology as well as disease [ 5 ]. The precise molecular profiles in the human endometrium under different hormonal conditions have not been fully characterized, despite extensive research. A better understanding of these profiles holds considerable promise for uncovering markers of endometrial receptivity [ 5 ]. Ovarian stimulation with gonadotropins leads to changes in hormone levels, which are believed to affect endometrial receptivity and reduce conception rates [ 6 , 7 ]. Studies have shown that the endometrium is more receptive in frozen-thawed embryo transfer cycles than in cycles with ovarian stimulation [ 8 ]. High levels of estrogen and progesterone during ovarian stimulation can affect the expression of endometrial genes and proteins that are involved in the embryo implantation process [ 9 ]. It is possible that a premature rise in progesterone during the ovarian stimulation causes a premature endometrial maturation and disrupts the WOI [ 10 ]. Comparisons of endometrial gene expression in spontaneous and stimulated cycles in the same patients have shown that gonadotropin stimulation induces the activation of genes which are not normally involved in the process of endometrial receptivity. On the other hand, expression of some genes that are known to be crucial for endometrial receptivity is reduced in stimulated cycles [ 11 – 13 ]. Due to numerous post-transcriptional modifications, gene expression data are not a reliable source for evaluating actual protein level dynamics. While many studies have used gene expression profiles to determine endometrial receptivity, investigations evaluating protein-level changes are limited. Therefore, this study aimed to determine differences in the endometrial protein profile during the WOI between spontaneous and stimulated cycles within the same patients with the goal of identifying clinically useful markers of endometrial receptivity that could enhance the embryo implantation process.

A principal component analysis (PCA) of the protein expression data was performed on the complete array dataset ( Fig 2 ) and on the dataset filtered for differentially abundant proteins ( Fig 3 ). In the PCA plots, the location of the samples is defined by their first two principal components, i.e., linear combinations of protein features with the largest variance across the samples. Samples with a similar profile are located in close proximity. A total of 114 antibodies demonstrated differential abundance between samples from spontaneous and stimulated cycles. Several proteins were significantly increased in stimulated cycles, including Interleukin-8 (IL-8; log2FC = 2.07, 95% CI 1.35–2.79; FDR-adjusted p = 3.6 × 10 ⁻ ⁶), Matrix metalloproteinase-9 (MMP-9; log2FC = 2.01, 95% CI 1.57–2.45; FDR-adjusted p = 4.4 × 10 ⁻ ¹²), Cathelicidin antimicrobial peptide (CAMP; log2FC = 2.05, 95% CI 1.60–2.50; FDR-adjusted p = 4.4 × 10 ⁻ ¹²), and proteins S100-A8 and S100-A9 (log2FC = 2.24, 95% CI 1.70–2.78; FDR-adjusted p = 1.3 × 10 ⁻ ¹⁰). The results of the statistical analysis are summarized in the volcano plot ( Fig 4 ), while detailed information on all differentially abundant proteins, including effect sizes and confidence intervals, is provided in S5 Table . The distribution of effect sizes across all antibodies meeting the FDR criterion is provided in S6 Table , to allow comprehensive assessment of smaller but statistically reproducible changes. Relative protein expression levels of the five most significantly upregulated proteins in endometrial samples from stimulated cycles compared to spontaneous cycles are presented in Fig 5 . The functional associations of five upregulated proteins with significant differential abundance observed in endometrial samples from stimulated cycles were explored using the STRING v12.0 [ 24 ] and are presented in Fig 6 . When investigating interactions among the most abundant proteins in endometrial samples from stimulated cycles, functional associations were found among all five proteins. Fig 7 shows the protein-protein interactions among the proteins with the highest abundance in endometrial samples from stimulated cycles using the STRING v12 [ 24 ]. Exploratory pathway and network analyses were performed using the STRING database to provide a descriptive overview of biological processes associated with differentially abundant antibodies. These analyses were based on curated database annotations and were not designed as independent inferential enrichment tests, nor were pathway-level multiplicity corrections applied. KEGG pathway mapping within the STRING framework was used to contextualize the differentially abundant antibodies in relation to known biological pathways [ 25 ]. The results were presented as an overview of potentially involved biological processes, with particular emphasis on immune response and signaling pathways ( Fig 8 ). Selected KEGG pathway maps are provided in the Supporting Information ( S1 Fig ). Relative protein levels of all differentially abundant proteins are summarized in Fig 9 . Exploratory hierarchical clustering indicated differences in protein abundance patterns between endometrial samples from spontaneous and stimulated cycles. The heatmap demonstrates a clear separation of samples according to treatment exposure, indicating substantial proteomic shifts associated with ovarian stimulation. The clustering of differentially abundant proteins identified two clusters: Cluster 1 occupies most of the left side of the heatmap and contains a mixture of endometrial samples from spontaneous and stimulated cycles, whereas Cluster 2 is located on the far right and consists exclusively of samples from stimulated cycles. Following gonadotropin stimulation, all endometrial samples exhibited changes in protein abundance. However, in the group of samples classified as Cluster 2, stimulation induced more pronounced changes in protein abundance. To explore potential factors influencing increased protein abundance in Cluster 2, we analyzed several variables, including age, body mass index (BMI), duration of gonadotropin therapy, cumulative gonadotropin dose, pregnancy outcome, thyroid-stimulating hormone (TSH) levels, anti-Müllerian hormone (AMH) levels, smoking status, number of oocytes retrieved, and number of embryos obtained. No statistically significant associations were observed between cluster membership and the evaluated clinical variables at the nominal level. After adjustment for multiple testing using the Benjamini–Hochberg FDR procedure, no associations remained statistically significant. Spearman’s rank correlation coefficient and partial correlation analysis highlighted the strong positive association of AMH with oocyte count, as well as moderate positive associations between AMH and embryo count – findings that have been well-documented in the literature. These correlations were not specific to cluster membership. To provide clinical context, we summarized subsequent reproductive outcomes descriptively. Pregnancies were observed only among women whose stimulated-cycle samples clustered into Cluster 2 (3/8; including one miscarriage), whereas no pregnancies occurred among women in Cluster 1 (0/7). The pregnancies observed occurred in women from different POSEIDON subgroups, suggesting that outcomes were not confined to a single POSEIDON category; however, subgroup sizes were too small for formal inference. Because clinical outcomes were not a prespecified endpoint of this exploratory study, these observations are descriptive and hypothesis-generating.

Ovarian stimulation protocols involving high doses of exogenous gonadotropins have the potential to influence the endometrial environment. Such alterations may have significant implications for implantation success, as endometrial receptivity is crucial for embryo attachment and subsequent development. The present study demonstrates distinct alterations in the endometrium-specific proteome following ovarian stimulation. The findings indicate that altered protein abundance, particularly of proteins associated with the immune response and extracellular matrix remodeling, may reflect biological processes relevant to implantation biology. This study may serve as a foundation for investigating key biomarkers that could play an important role in regulating endometrial receptivity in both spontaneous and stimulated cycles.

This prospective case-crossover clinical trial was conducted at the Department of Human Reproduction, Division of Obstetrics and Gynecology, University Medical Centre Ljubljana, Slovenia from September 2023 to June 2024. Protein profiling of endometrial samples was performed at Sciomics GmbH, Heidelberg, Germany. A total of 15 consecutive infertile patients aged <43 years, with primary infertility and poor ovarian response, were invited to participate in the study. Fig 1 shows the flowchart of patient inclusion. Patients were stratified into four groups according to the latest POSEIDON classification [ 14 ]. Representation across POSEIDON groups was limited, and subgroup sizes were small (POSEIDON 1: n = 5; POSEIDON 2: n = 3; POSEIDON 3: n = 4; POSEIDON 4: n = 3). Accordingly, the study was not powered for POSEIDON-stratified proteomic inference. The exclusion criteria were as follows: 1. Normal or high ovarian response to gonadotropin stimulation 2. Menstrual cycle disorders 3. Severe male factor infertility 4. Preimplantation genetic testing 5. Pathological uterine findings (fibroids, endometrial polyps, adenomyosis) and tubal pathology (distal occlusion of one or both fallopian tubes with hydrosalpinx). To assess the protein expression profile in spontaneous menstrual cycles, the first endometrial biopsy was performed in all subjects during the WOI. Ovulation was determined using urinary LH surge test strips (visual interpretation). In all participants, the endometrial biopsy in the spontaneous cycle was performed 7 days after a clearly positive urinary LH surge test (LH + 7), thereby standardizing sampling to the mid-luteal WOI. This timing typically corresponded to cycle days 20–24, depending on individual cycle length. The samples were collected using a plastic pipelle (Rampipella Ri.Mos.S.R.L. Mirandola, Italy), snap-frozen and stored at −80 °C until the final analysis. Following the endometrial biopsy, 4 mg of estradiol was administered to synchronize and coordinate follicular growth for a maximum of 10 days [ 15 ]. A uniform ovarian stimulation protocol was implemented in all patients during the follicular and luteal phases of the menstrual cycle. A double ovarian stimulation protocol was used, which has shown potential in patients with poor ovarian response [ 16 – 18 ] and requires freezing of all good-quality embryos, thereby allowing invasive endometrial biopsy during the WOI of stimulated cycles. On day 2 of the menstrual cycle, ovarian stimulation with high-dose recombinant FSH (Gonal-f 300 IU/day subcutaneously) was started according to the short antagonist stimulation protocol. On day 7 of the cycle, GnRH antagonist (Cetrorelix, Cetrotide 0.25 mg/day subcutaneously) was started. A GnRH agonist trigger (Gonapeptyl 0.1 mg subcutaneously) was administered for final oocyte maturation when at least three follicles reached 17–18 mm in diameter. Oocyte retrieval was performed 36 hours after GnRH agonist administration. No luteal phase hormonal supplementation was administered after oocyte retrieval, as all embryos were vitrified according to the freeze-all protocol required by the double stimulation approach . In the stimulated cycle, the endometrial biopsy was performed 7 days after the ovulation trigger, which was considered the functional equivalent of LH + 7 in the spontaneous cycle. On the same day, luteal phase ovarian stimulation was initiated using the same protocol as in the follicular phase. Fertilization, blastocyst culture, and embryo vitrification were carried out according to established laboratory methods described previously [ 19 ] and all good-quality embryos were vitrified. The warmed blastocysts were transferred in the following spontaneous or hormone-induced menstrual cycles. The study protocol is provided as Supporting information (S2 and S3 Trial study protocols). The laboratory outcomes related to IVF were not the focus of this manuscript and were not included in the results of this study. The primary objective of this exploratory proteomic study was to identify proteins with differential abundance in endometrial samples obtained during the WOI in spontaneous and stimulated cycles within the same patients. The estimand of interest was the within-patient log2-fold change (log2FC) in antibody signal intensity between stimulated and spontaneous cycles. Protein profiling of 30 endometrial samples (15 paired samples) was performed using antibody-based microarrays, enabling high throughput targeting studies of a large number of proteins in different biological samples. Fifteen samples represented endometrial tissue from spontaneous menstrual cycles, while the remaining fifteen were obtained following ovarian stimulation. Proteins were extracted with scioExtract buffer (Sciomics) according to the manufacturer’s standard operating procedures, and total protein concentration was determined by bicinchoninic acid (BCA) assay. A reference sample was established by pooling equal volumes of all individual samples. The samples were labeled at an adjusted protein concentration for 2 hours with scioDye 2 (Sciomics). The reference sample was labeled with scioDye 1 (Sciomics). After two hours, the reaction was stopped and the buffer was exchanged to phosphate-buffered saline (PBS). All labeled protein samples were stored at −20 °C until use. The samples were analyzed in a dual-color approach using a reference-based design on scioDiscover antibody microarrays (Sciomics) targeting 1,466 different proteins with 1,925 antibodies. Statistical inference was performed at the antibody (probe) level. When multiple antibodies targeted the same protein, each antibody was treated as an independent measurement, as different antibodies may recognize distinct epitopes or isoforms and yield divergent signal patterns. No probe aggregation to unique protein targets was performed. Each antibody was represented by four replicate spots on the array. The arrays were blocked with scioBlock (Sciomics) on a Hybstation 4800 (Tecan, Austria) and subsequently incubated competitively with the reference sample using a dual-color approach. Following incubation for 3 hours, the slides were thoroughly washed with 1 × PBSTT, rinsed with 0.1 × PBS and water, and subsequently dried with nitrogen [ 20 ]. Slide scanning was conducted using a Powerscanner (Tecan, Austria) with constant instrument laser power and photomultiplier tube settings. Spot segmentation and local background estimation were performed with GenePix Pro 6.0 (Molecular Devices, Union City, CA, USA). No additional background correction (e.g., normexp) was applied prior to normalization. For the scioDiscover antibody microarray platform, background levels are typically minimal due to optimized surface chemistry and blocking procedures; further background correction may increase variance when background signal is low. Acquired raw data were analyzed using the linear models for microarray data (LIMMA) package of R-Bioconductor after uploading the median signal intensities [ 21 ]. During spot segmentation, the intensity of each spot was calculated as the median of all pixel intensities within that spot. No summarization across the four replicate spots per antibody was performed prior to statistical analysis; instead, all replicate spots for each antibody and sample were retained and incorporated into the linear modeling framework. Data normalization was performed using an invariant Lowess method specifically developed for the scioDiscover antibody microarray platform [ 22 ]. This within-array normalization approach corrects for systematic intensity-dependent effects and potential dye bias inherent to dual-color microarrays. A pooled reference sample, consistently labeled with scioDye 1 across all arrays, provided a common baseline for normalization. No dye-swap experiments were performed. For differential protein abundance analysis, a multi-factorial linear model was fitted using the LIMMA package. In addition to the treatment factor (spontaneous vs. stimulated cycle), patient-specific effects were included as an additional factor in the linear model to explicitly account for the paired case-crossover design. Given the paired case-crossover design and the limited sample size, no additional covariate adjustment was applied in the proteomic models. Stable subject-level characteristics were inherently controlled by the within-subject comparison, while stimulation-related variables were considered biologically integral to the exposure of interest. Additional covariate adjustment was therefore avoided to reduce the risk of model overfitting and over-adjustment. Differential abundance between spontaneous and stimulated cycles was assessed using moderated t-statistics. All p-values were adjusted for multiple testing using the Benjamini–Hochberg false discovery rate (FDR) procedure [ 23 ]. Protein abundance differences were reported as log2FC with corresponding 95% confidence intervals derived from the paired multi-factorial linear model. Differential protein abundance was defined a priori as an absolute log2FC greater than 0.5 with a FDR-adjusted p value < 0.05. A log2FC threshold of 0.5 (corresponding to an approximately 1.4-fold change) was selected to balance statistical significance with biological relevance and to reduce the likelihood of emphasizing very small effect sizes that, although statistically significant, may not reflect meaningful biological differences in the endometrial microenvironment. Antibodies meeting both criteria were classified as differentially abundant and were presented in S5 Table . Antibodies meeting the FDR threshold but not exceeding the predefined log2FC cutoff were retained to allow transparent assessment of smaller yet statistically robust effects, and were presented separately in S6 Table . Pathway enrichment analyses and descriptive assessment of selected proteins with established relevance to endometrial receptivity were conducted as secondary exploratory analyses. To explore whether stimulated-cycle samples exhibited distinct proteomic patterns beyond the overall stimulated versus spontaneous comparison, we performed an exploratory unsupervised hierarchical clustering analysis of antibody-level expression profiles and visualized the output as a heatmap ( Fig 9 ). Protein signals were centered and scaled by antibody prior to clustering. Based on the hierarchical clustering dendrogram (using a two-cluster cut), samples were assigned to two clusters (Cluster 1 and Cluster 2), and cluster membership was treated as a binary exploratory grouping variable (Cluster 1 vs. Cluster 2) for subsequent post hoc comparisons with clinical and stimulation-related characteristics. Scatter plot displaying the first two principal components of the samples’ protein signal data using complete array data. The percentages given in the axis labels indicate the ratio of total variance explained by the respective principal component. Scatter plot displaying the first two principal components of the samples’ protein signal data based on differentially abundant proteins. The percentages given in the axis labels indicate the ratio of total variance explained by the respective principal component. Protein abundance differences are expressed as log2FC, and statistical significance is shown as FDR-adjusted p-values. The horizontal red line indicates the significance threshold (FDR-adjusted p = 0.05), and the vertical lines indicate the predefined log2FC cut-offs (±0.5). Positive log2FC values indicate higher protein abundance in endometrial samples from stimulated cycles, whereas negative log2FC values indicate higher protein abundance in endometrial samples from spontaneous cycles. Proteins meeting the predefined criteria for differential abundance (|log2FC| > 0.5 and FDR-adjusted p value < 0.05) are labeled in blue. Proteins meeting the FDR-adjusted p value criterion but not exceeding the log2FC threshold are shown in green. Each sample was measured by four replicate spots per array. Diamonds indicate sample group means. Whiskers indicate one standard deviation. A. Interleukin-8 (IL-8); B. Matrix metalloproteinase-9 (MMP-9); C. Cathelicidin antimicrobial peptide (CAMP); D. Proteins S100-A8 and S100-A9. Colored nodes indicate direct interactions with other proteins. The threshold for establishing interaction links has been set to the highest level of confidence (0.900). The node representing the studied protein is shown in red. A. Interleukin-8 (IL-8); B. Matrix metalloproteinase-9 (MMP-9); C. Cathelicidin antimicrobial peptide (CAMP); D. Proteins S100-A8 and S100-A9. The threshold for establishing interaction links has been set to the high level of confidence (0.700). The green line represents the simultaneous mention of related proteins in the published literature. The pink line represents the discovery of putative homologs engaging in interactions in other organisms. The black line represents the co-expression of proteins in humans or the co-expression of their putative homologs in other organisms, suggesting functional association among the proteins. Interleukin-8 (IL-8), Matrix metalloproteinase-9 (MMP-9), Cathelicidin antimicrobial peptide (CAMP), proteins S100-A8 and S100-A9. Values were centered and scaled by protein. Cluster 1: left cluster on the heatmap; Cluster 2: right cluster on the heatmap. Post hoc exploratory analyses were conducted to evaluate whether cluster membership (Cluster 1 vs. Cluster 2) was associated with selected clinical and stimulation-related variables (age, BMI, stimulation duration, cumulative gonadotropin dose, smoking status, AMH, TSH, number of oocytes retrieved, number of embryos obtained, and pregnancy outcome). Continuous variables were compared using the Mann–Whitney U test and categorical variables using Fisher’s exact test. To account for multiplicity across these exploratory cluster–clinical association tests, p-values were adjusted using the Benjamini–Hochberg FDR procedure. Associations among clinical and laboratory variables were assessed using Spearman’s rank correlation and partial correlation analyses. All comparisons of clinical and laboratory characteristics were performed at the subject level, with each woman contributing a single observation. Clinical variables were not analyzed at the sample level and were therefore not duplicated across paired endometrial samples obtained from spontaneous and stimulated cycles. These analyses were considered exploratory and hypothesis-generating, given the limited sample size. This study was reported in accordance with the CONSORT 2025 guidelines; the completed checklist is provided in the Supporting Information ( S4 Checklist ). The study involving human participants was reviewed and approved by the Medical Ethics Committee of the Republic of Slovenia (0120–319/2021/3). All participants provided written informed consent to participate in this study. The trial was registered retrospectively at ClinicalTrials.gov ( NCT06804174 ) after the initiation of participant recruitment. This delay occurred due to a procedural oversight in recognizing registration requirements at the time the study was initiated, despite prior ethics committee approval and written informed consent for all participants. The study protocol, eligibility criteria, sampling schedule, and planned analyses were defined before data analysis, and we report all outcomes described in this manuscript transparently. Since then, we have implemented an internal pre-enrollment checklist requiring trial registration before recruitment for all prospective interventional studies, to ensure compliance with international standards in future work.

(TIF) (DOCX) (DOCX) (DOCX) (XLSX) (XLSX)