Investigating cycle-to-cycle variation and the impact of endometrial injury through differential gene expression analysis.

The average woman will experience up to 450 menstrual cycles over their lifetime ( Chavez-MacGregor et al. , 2008 ). During each menstrual cycle, the endometrium grows, vascularizes, and becomes secretory in preparation for embryo implantation. If a pregnancy is not achieved, the endometrium breaks down, bleeds, and repairs to regenerate a new endometrial lining ( Critchley et al. , 2020 ). The endometrium is a complex and dynamic tissue, consisting of glandular epithelium, vascularized stroma, and immune cells. The presence of a ‘receptive’ endometrium when a viable embryo arrives at the uterine cavity is a prerequisite for successful implantation and ongoing pregnancy. In a normal ovulatory cycle, the receptive endometrium develops following sequential exposure to sex steroids–estrogen and progesterone. The endometrium undergoes a series of morphological and functional cellular changes as it moves towards receptive status, including stromal oedema, decidualization, and appearance of luminal epithelial pinopodes ( Teh et al. , 2016 ). Previous studies demonstrated significant changes in gene expression on a daily basis as the endometrium transitioned from the proliferative to secretory phase, including embryo implantation receptive phase ( Wang et al. , 2020 ; Teh et al. , 2023 ). Endometrial receptivity is essential for successful embryo implantation yet remains a major limiting factor in ART. Despite the availability of high-quality embryos, implantation failure continues to pose a significant clinical challenge. One intervention that has been proposed to enhance implantation is endometrial injury, a procedure involving intentional disruption to the endometrial lining during the luteal phase or prior to embryo transfer. The rationale stems from animal studies and early human observations suggesting that local endometrial injury may induce a regenerative or proinflammatory response that enhances receptivity during the subsequent cycle ( Barash et al. , 2003 ). Early clinical trials reported increased pregnancy rates following endometrial injury in women with previous IVF failure ( Barash et al. , 2003 ; Narvekar et al. , 2010 ; El-Toukhy et al. , 2012 ; Nastri et al. , 2012 ; Potdar et al. , 2012 ). However, subsequent large randomized controlled trials and meta-analyses have yielded conflicting results—some demonstrating no benefit in unselected IVF populations ( Lensen et al. , 2019 ; Vitagliano et al. , 2019 ), while others suggest a potential benefit ( van Hoogenhuijze et al. , 2023 ). These discrepancies have led to debate regarding the clinical utility of the intervention and underscored the need for an improved understanding of how endometrial injury might influence implantation potential. In addition to the conflicting evidence on whether endometrial injury improves implantation rates, there is no clear consensus on its underlying mechanism(s) of action. Notably, clinical protocols consistently perform the injury in the cycle preceding embryo transfer, as performing it in the same cycle is considered detrimental. This raises a key biological question: how could an intervention in one cycle influence implantation in the next? Any beneficial effect would need to persist beyond menstrual shedding and be retained in the regenerated functionalis layer of the subsequent cycle. It has been hypothesized that mechanical disruption of the endometrium may enhance uterine receptivity by inducing a tumour necrosis factor alpha (TNFα)-mediated inflammatory cascade ( Gnainsky et al. , 2010 ). This proinflammatory response is proposed to upregulate cytokines and chemokines, leading to recruitment of monocytes, macrophages, and dendritic cells, which in turn may influence uterine natural killer cell differentiation and modulate adhesion molecule expression ( Granot et al. , 2012 ). Additionally, injury-induced inflammation may promote endometrial regeneration via stem cell activation ( Liang et al. , 2015 ). These processes are considered integral to the transition of the endometrium from a non-receptive to a receptive state. However, studies investigating the transcriptomic effects of endometrial injury have been limited by absence of paired controls, or inadequate adjustment for menstrual cycle timing. Moreover, it remains unclear whether the observed molecular changes actively contribute to implantation success or simply reflect a general wound-healing response. In this study, we leveraged RNA sequencing of paired endometrial samples collected across two consecutive menstrual cycles (before and after endometrial injury) from the same individuals, including women who successfully conceived subsequently and those who did not. By integrating high-resolution transcriptomic data with precise cycle timing ( Teh et al. , 2023 ), we aimed to: (i) examine inter- and intra-cycle variation in endometrial gene expression, (ii) identify genes and pathways altered in response to endometrial injury, and (iii) explore whether these molecular signatures were associated with pregnancy success. This paired, time-corrected design provides a robust framework for understanding cycle-to-cycle variability and mechanistic assessment of endometrial injury.

A total of 23 women were recruited for the study, and 45 endometrial biopsies were collected. Three participants were excluded: one due to biopsy collection at the incorrect time point (identified by molecular dating), one due to incomplete data collection, and one who withdrew prior to the second biopsy. Thus, 20 matched pairs of endometrial biopsies were included in the final analysis. The mean age of participants included in the final analysis was 36.9 ± SD 4.6 (range 26–45) years, and the average menstrual cycle length was 28.6 ± SD 2.3 (range 25–35) days. Of the 20 women, 11 had a history of prior pregnancy. Relevant clinical characteristics included a history of endometriosis (n = 7), polycystic ovary syndrome (PCOS; n = 2), pelvic inflammatory disease (n = 1), hydrosalpinx (n = 2), uterine fibroids (n = 5), bicornuate uterus (n = 1), and autoimmune thyroiditis (n = 1). Following participation in the study, 10 women conceived through IVF and were classified as the fertile group for sub-analysis. The remaining 10 women did not achieve pregnancy by the end of the 2-year follow-up period and were categorized as the infertile group. The mean age was 34.8 years (range 26–42) and the average cycle length was 27 days in the fertile group, whereas the mean age was 39 years (range 32–45) and the average cycle length was 28.75 days in the infertile group. More participants in the infertile group had a history of previous pregnancy (7 vs 4), but there were no apparent differences in relevant gynaecological conditions between the groups. The average number of unsuccessful embryo transfer cycles in the infertile group was 4.3. Within the infertile group, nine women met the criteria for recurrent IVF failure (RIF), defined as a > 60% cumulative chance of clinical pregnancy following multiple unsuccessful embryo transfers ( Cimadomo et al. , 2023 ) ( Tables 1 and 2 ). Baseline characteristics of participants. Baseline characteristics of fertile and infertile participants. The majority of endometrial samples (n = 35) were collected at clinical LH + 7–8. Two samples from the same woman were collected at LH + 4, and three samples from different women were collected at LH + 10. One serum progesterone level (sample N2) was missing; all other serum progesterone levels were within the secretory range ( Table 3 ). Biopsy day based on clinical dating and progesterone levels on the day of biopsy (with one missing progesterone level for woman N). A total of 17 753 genes were included in the analysis from 20 paired samples. To explore variation in gene expression across samples, we performed PCA on log-transformed CPM values of paired biopsies ( Fig. 1a and b ). The first PCA plot demonstrated a lot of variation in grouping positions of biopsies, with only one sample (M) clustering close together ( Fig. 1a circled). This is however likely due to sequencing batch effect. Endometrial molecular dating model enables precise dating of the endometrial biopsy and identification of differentially expressed genes between endometrial samples ( Teh et al. , 2023 ). From the second PCA plot, it could be seen that the samples with similar menstrual cycle timing clustered together ( Fig. 1b ). These data indicated the while two samples (M and N) grouped together at similar time points (indicating they were collected at the same endometrial cycle stage on consecutive cycles), a majority of the samples were collected across several time points of the secretory phase from post-ovulation days (POD) 3–12, based on molecular dating. These findings indicate the variation in gene expression observed is likely due to the variable collection times of the samples. The findings associated with the molecular dating of samples also highlights the potential inaccuracy of using the LH + 7 time point as a true indicator of endometrial dating. Pairwise trajectories between pairs were also visualized using line segments connecting corresponding points in PCA space before correcting for cycle time ( Fig. 1c ) and after correction for cycle time ( Fig. 1d ). Principal component analysis (PCA) was performed to explore variation in gene expression across samples. Samples were projected into PCA space, coloured for molecular cycle time and labelled in pairs. ( a ) The PCA plot demonstrated a lot of variation in clustering positions of biopsies with only one sample (M) clustering close together. ( b ) Samples with similar menstrual cycle timing also clustered together. ( c ) Pairwise trajectories between pairs were visualized using line segments connecting corresponding points in PCA space before correcting for cycle time and ( d ) after correction for cycle time. To assess intra-individual vs inter-individual variability in endometrial gene expression, we compared transcriptomic variance between paired samples from the same patient (1st and 2nd biopsy) against variances between randomly paired samples from different individuals. The average gene expression variance was significantly lower among same-patient pairs compared to random pairs (132.8 vs 156.7), respectively ( t = −3.8, P  = 0.001) ( Fig. 2a ). Comparison of transcriptomic variance between paired (within-patient) and randomly paired (between-patient) endometrial samples. ( a ) Paired analyses showed that Euclidean distances were significantly lower among same-patient pairs (mean = 132.8) compared to random pairs (mean = 156.7; t = −3.8, P  = 0.001), suggesting high intra-individual consistency and supporting the strength of the paired design. ( b ) After adjustment for batch effect and cycle stage, the distances between samples from the same individual remained lower than between randomly selected pairs (115.1 vs 123.6), but this difference did not reach statistical significance ( t = −1.8, P  = 0.09, 95% CI −18.0 to 1.5). To determine what role the menstrual cycle phase had on sample variability, we corrected the gene expression matrix for batch effects and estimated cycle timing. After adjustment, the mean transcriptomic variance between samples from the same individual remained lower than between randomly selected pairs (115.1 vs 123.6), but this difference did not reach statistical significance ( t = −1.8, P  = 0.09, 95% CI −18.0 to 1.5) ( Fig. 2b ). Taken together, these data suggest that while individual-level gene expression profiles remain more similar post-adjustment, much of the original variation observed may be attributable to differences in cycle phase. We modelled gene expression as a function of biopsy round (1st vs 2nd biopsies), using linear models with patient modelled as a random effect. Covariates included batch and estimated molecular POD based upon the molecular staging model of Teh et al. (2023) . No genes reached significance after multiple testing correction (FDR > 0.05) ( Table 4 ). The absence of significant changes in gene expression between 1st and 2nd paired biopsies after cycle stage correction suggests the programmed regeneration of endometrial tissue is not notably influenced by prior injury. Differential gene expression analysis following endometrial injury. No genes reached statistical significance after correction for multiple testing (FDR > 0.05). Genes with positive logFC values are considered upregulated, whereas negative logFC values indicate downregulation. Paired analysis of biopsies collected before and after endometrial injury revealed no significant changes in cell lineage composition after adjusting for cycle timing and batch, although total immune proportion showed a small, non-significant upward trend of 0.021 (se = 0.011, P  = 0.068) ( Table 5 ). Investigation of cellular proportion changes as a result of endometrial injury demonstrated no significant changes in cell lineage composition after adjusting for cycle timing and batch. We examined whether the transcriptional response to endometrial injury differed according to age. Analyses were performed using age as both a continuous and a dichotomous variable. When age was modelled as a continuous covariate interacting with endometrial injury, genes showing the strongest age-modulated changes (e.g. LTF, SAA1, MMP9, IL18) were enriched for innate immune and inflammatory pathways; however, none reached genome-wide significance (FDR > 0.1). When age was modelled dichotomously (<35 vs ≥35 years), neither the main age effect nor the endometrial injury interaction effect yielded significant differential expression. These findings suggest that the endometrial transcriptomic response to an endometrial injury is not affected by age. To examine the impact of prior endometrial injury on fertility outcomes, differential gene expression analysis was performed on the transcriptional data of the second biopsy sample to determine transcriptional differences between women who did and did not achieve pregnancy (within a 2-year follow-up period). No significant genes were identified in this comparison ( Table 6 ). Endometrial DGE comparing pregnancy success following endometrial injury: No genes reached statistical significance after correction for multiple testing (FDR > 0.05). Genes with positive logFC values are considered upregulated, whereas negative logFC values indicate downregulation.

The study was conducted at the Reproductive Services Unit at Royal Women’s Hospital (RWH) in Melbourne between 2014 and 2017. Ethical approval was obtained from the RWH Research and Human Research Ethics Committee (Project no: 13/38). All patients provided informed written consent for participation in the study. Twenty women aged 26–45 years with regular monthly menstrual cycles were recruited to undergo endometrial biopsies during the mid-secretory phase of two consecutive menstrual cycles. Exclusion criteria included current use of hormonal treatments or supplements, and a history of endometrial hyperplasia or malignancy. For study sub-analysis, participants were subdivided into fertile and infertile groups based on whether they successfully conceived within the 2-year follow-up period after participating in the study. Participants were instructed to use urinary LH detection kits (Seratec, GmbH, Germany) to identify their LH surge. In cases where urinary LH results were inconclusive (n = 8), transvaginal ultrasound for follicle tracking and/or blood tests (measuring serum LH and progesterone levels) were performed. An LH level >20 IU/l was considered indicative of a positive LH surge, and a progesterone level >5.0 nmol/l confirmed ovulation. Participants attended the clinic for outpatient endometrial biopsy and venesection on LH + 6–8 days. A Pipelle catheter (Pipelle de Cornier, Laboratoire, France) was inserted through the cervix into the uterine cavity, and tissue was obtained by moving the Pipelle back and forth while maintaining suction via the withdrawn piston. If a participant had been recommended for hysteroscopy, the endometrial biopsy was performed during that procedure instead. Peripheral blood was collected following the biopsy. The same procedure was repeated in the subsequent menstrual cycle. Demographic, menstrual, fertility, and medical history—including medication use—were obtained via participant questionnaires and review of medical records. Endometrial tissue samples were stored in RNAlater (Life Technologies, Grand Island, NY, USA) at 4 °C before long-term storage at −80 °C for total RNA extraction. Serum was extracted from centrifuged blood samples and stored in clean polypropylene tubes at −80 °C for subsequent hormone assays. Total RNA was extracted from homogenized endometrial tissue using RLT buffer and the RNeasy Plus Mini Kit, following the manufacturer’s instructions (QIAGEN, Valencia, CA, USA) followed by RNA quantification using the NanoDrop ND-6000 spectrophotometer (Thermo Fisher Scientific, Scoresby, VIC, Australia). RNA integrity was assessed using the Agilent Bioanalyzer 2100 (Agilent Technologies, Santa Clara, CA, USA). All RNA samples exhibited high integrity, with RNA Integrity Numbers greater than 8. Samples were treated with the Turbo DNA-free kit (Thermo Fisher Scientific, USA) prior to RNA-seq library preparation, as previously described ( Mortlock et al. , 2020 ). Stranded RNA-seq libraries were generated using the Illumina TruSeq Stranded Total RNA Gold protocol, which includes ribosomal RNA depletion and sequenced using Illumina NovaSeq6000 (Illumina, USA). We received assistance from Australian Genome Research Facility for next-generation sequencing (majority of the samples were single-ended reads, up to 30M reads per sample). Read quality was assessed using FastQC v0.12.1 ( Simon Andrews et al. , 2010 ) and low-quality reads and adapter sequences were trimmed using Trimmomatic v0.36 with the following parameters: ‘LEADING:3 TRAILING:3 SLIDINGWINDOW:4:20 MINLEN:35’ ( Bolger et al. , 2014 ). For probes to be included in the analysis, probe quality was required to be listed as ‘Good’ or ‘Perfect’ and we required the probe to have a detection P -value < 0.05 in at least 90% of samples. Reads were mapped to the human reference genome ( Homo sapiens GRCh38) using HISAT2 v2.21 ( Kim et al. , 2019 ) and quantified using StringTie v2.2.1 ( Pertea et al. , 2015 ) with the Ensembl Homo sapiens GRCh38 release 91 reference annotation. Raw gene counts were normalized for library size and composition bias using the Trimmed Mean of M -values method implemented in edgeR v4.4.1 ( Robinson et al. , 2010 ). Genes with low expression—defined as counts-per-million (CPM) <0.5 in less than 50% of samples—were excluded from further analysis. Menstrual cycle time was estimated using the endest v.0.1.0 R package ( Teh et al. , 2023 ), where the post-ovulatory day (POD) for each sample was estimated using the ‘secretory’ model. For visualization, log-transformed CPM values were calculated, and the removeBatchEffect function in limma v3.62.2 ( Ritchie et al. , 2015 ) was used to remove sequencing batch and menstrual cycle time effects, modelling cycle time as a B-spline with 5 degrees of freedom. Principal component analysis (PCA) and pairwise distances of samples were used to examine sample-to-sample variation. A PCA was performed on the normalized expression data and cycle-corrected expression data to assess whether samples that originated from the same patient clustered together. Similarly, distance matrices were generated using Euclidean distance, and hierarchical clustering was performed to visualize which samples were most similar to each other. Pairwise distances between intra-patient pairs and all permutations of inter-patient pairs were compared to each other using Welch’s t -tests. Differential gene expression analysis was conducted with the limma R package using a paired design. Gene expression, modelled as log-transformed CPM, was fit using a linear mixed-effects model with the patient modelled as random effect and the before/after injury status as a fixed effect. Sequencing batch and estimated POD were also included as covariates in the model. Empirical Bayes moderated t -statistics were used to assess if genes were significantly differentially expressed. The resulting P -values were corrected for multiple hypothesis testing using the Benjamini–Hochberg method to control the false discovery rate (FDR). To assess whether age influenced the transcriptional response to endometrial injury, additional models were fitted including age as either a continuous variable or a dichotomous variable (<35 vs ≥35 years), modelled as an interaction term. To assess differences between patients that did and did not achieve pregnancy following participation in the study, only the post-injury samples of each patient were used in the analysis. Cell-type proportions were estimated using reference-based bulk RNA-seq deconvolution using the Human Endometrial Cell Atlas ( Marečková et al. , 2024 ) and MuSiC v1.0.0 ( Wang et al. , 2019 ). Linear mixed-effects models were used to evaluate cell lineage proportion differences before and after injury, with sequencing batch and estimated POD included as covariates, and patient modelled as a random effect.

This prospective transcriptomic study of the human endometrium during the secretory phase did not identify any consistent differentially expressed genes following endometrial injury, a common intervention technique used in ART to enhance implantation success ( Barash et al. , 2003 ). After applying a rigorous normalization pipeline—including correction for variation in menstrual cycle stage—no statistically significant or reproducible gene expression patterns were observed between the 1st and 2nd consecutive endometrial biopsies. These findings challenge the biological rationale for endometrial injury and raise important questions about its therapeutic utility in clinical practice. A previous microarray study reported differential gene expression between women with prior endometrial injury (n = 4) and those without (n = 4), suggesting a potential effect of local injury on endometrial receptivity ( Kalma et al. , 2009 ). However, this small study by Kalma et al. warrants caution in interpretation. Their analysis was conducted retrospectively, including only patients who conceived after endometrial injury and comparing them with non-injured individuals who did not conceive. This approach introduces selection bias and limits generalizability. Furthermore, the data were not adequately adjusted for the specific day of the menstrual cycle. In contrast, the current study employed a prospective paired-sample design, comparing gene expression profiles before and after endometrial injury within the same individuals, thereby minimizing inter-patient variability and potential confounding. Gnainsky and colleagues proposed that endometrial injury-induced local inflammation, demonstrated by increased macrophage and dendritic cell infiltration, and elevated proinflammatory cytokine expression in samples obtained during the same cycle (n = 42) compared to controls (n = 22). They found a positive correlation between inflammatory markers and pregnancy outcomes ( Gnainsky et al. , 2010 ). Similarly, Liang et al. (2015) reported increased levels of cytokines, chemokines, and growth factors in endometrial secretions from women with unexplained infertility who had undergone endometrial injury in a preceding cycle (n = 38), compared to untreated controls (n = 28). However, gene expression data in these studies were not adjusted for menstrual cycle stage, a critical confounder in endometrial transcriptomic analyses. The absence of appropriately paired controls further introduces the risk of bias and limits the reliability of the reported findings. Importantly, in clinical practice, endometrial injury is typically performed in a cycle preceding embryo transfer, but can also be performed up to Day 3 of the same cycle. Given the regenerative nature of the endometrium and its shedding during menstruation, any immediate inflammatory response induced by endometrial injury may not persist into subsequent cycles. A recent study ( Tian et al. , 2023 ) involving paired endometrial biopsies collected from consecutive cycles at LH + 7 from eight women with recurrent implantation failure (not defined by authors) reported no significant differences in gene expression between the two biopsies, including genes associated with endometrial receptivity—findings consistent with our results. Our study included a larger cohort, applied more stringent cycle-stage correction using a molecular staging model, and incorporated both women with proven fertility and those with recurrent IVF failure, thereby enhancing the robustness and generalizability of the findings. Several tests have been developed to assess endometrial receptivity with the goal of optimizing embryo transfer timing in assisted reproduction. The most established is the Endometrial Receptivity Analysis (ERA), which uses transcriptomic profiling of 248 genes to define a ‘window of implantation’ and classify the endometrium as receptive or non-receptive ( Díaz-Gimeno et al. , 2011 ). Other commercial assays include BeReady, which similarly uses RNA-seq-based profiling but incorporates machine learning for interpretation ( Teder et al. , 2018 ), and ERPeak, which utilizes RT-qPCR for a smaller panel of genes ( Enciso et al. , 2018 ; Ohara et al. , 2022 ). The clinical utility of receptivity testing remains debated, with some studies suggesting improved outcomes in patients with recurrent implantation failure, while others report no benefit in unselected populations ( Simón et al. , 2020 ; Glujovsky et al. , 2023 ). All of these endometrial tests involve biopsy of endometrium the cycle before embryo transfer cycle, a design we mirrored in our research. Although our data show that endometrial samples from the same woman had a marginally more similar molecular profile than samples from different women, these similarities were weak. Importantly, we observed considerable variability between samples even from the same individual across cycles. This has important implications for the clinical utility of endometrial receptivity testing. Tests such as the ERA are designed to pinpoint the optimal implantation window by staging the endometrium in a given cycle. However, because the timing of the receptive phase can vary from one cycle to the next—even in the same woman—and because LH surge detection is an imprecise marker of ovulation ( Lenton et al. , 1984 ), identifying a receptive day in one cycle does not guarantee accurate prediction in the next. Our data show marked variability in endometrial gene expression across cycles, suggesting that even if a woman is found to have a receptive endometrium in one cycle, her endometrial state at the time of embryo transfer in a subsequent cycle may be meaningfully different. Together, these findings raise concerns about the reliability of single-cycle biopsies to guide personalized embryo transfer timing. A particular strength of our study is the ability to rigorously account for gene expression changes related to menstrual cycle stage as a covariate. Endometrium is a dynamic tissue that undergoes dramatic cyclical changes in gene expression in response to changing levels of circulating estrogen and progesterone during the menstrual cycle ( Ponnampalam et al. , 2004 ). This rapidly changing gene expression landscape makes accurate menstrual cycle staging critical for measuring differential endometrial gene expression between samples. To address the challenges posed by variability in gene expression and cycle length, we applied a bioinformatics approach to accurately determine endometrial cycle stage based on global gene expression profiles ( Teh et al. , 2023 ). This ensures that any differential gene expression identified between samples is not confounded by differences in cycle stage. The endometrium is divided morphologically into an upper two-thirds functionalis layer and a lower one-third basalis layer. Given that most of the functionalis is lost during menstruation, any injury-induced inflammatory response would need to be long-lived and/or affect the basalis to have an ongoing impact on endometrial receptivity in subsequent cycles. Like most studies on human endometrium, endometrial samples in our study were collected through Pipelle biopsy from the functionalis. Sampling of the basalis may be able to provide more information with regards to the impact of endometrial injury on the endometrium. In summary, our differential gene expression analysis of paired endometrial samples including women suffering from RIF did not reveal any statistically significant changes in gene expression after endometrial injury. These findings do not support the use of endometrial injury to improve endometrial receptivity for IVF patients. Our findings also further highlight a worrying discordance in predicting ovulation and/or endometrial receptivity based on clinical dating relative to more sophisticated molecular endometrial dating. While there was a tendency for samples from the same individual to exhibit more similar molecular profiles compared to those from different individuals, these similarities were modest. This intra-individual variability may help explain the inconsistent clinical outcomes observed with endometrial receptivity testing.