Live birth in patients stimulated with r-hFSH or r-hFSH: r-hLH is strongly associated with cumulus cell derived gene expression models.

This observational cohort study was conducted at the Universitair Ziekenhuis Brussel (Brussels IVF, Brussels, Belgium) from January 2021 until June 2023. Ethical approval was obtained from the Ethical Committee of Vrije Universiteit Brussel – UZ Brussel (IEC: 2020.335) and the study was registered at ClinicalTrials.gov (ID NCT04710264 ). Patients aged 22–38 years, with a BMI between 17–33 kg/m2, were included in the study upon providing written informed consent. These patients underwent their first or second ICSI treatment cycle, and consented to individual oocyte denudation, RNA expression testing using their cumulus cells, and a fresh single blastocyst transfer. Eligibility required patients to be good ovarian responders defined as having an AMH level between 1 to 4.7 ng/ml and 7 to 18 follicles measuring 10–11 mm on the last ultrasound. Patients agreed to receive either r-hFSH (Gonal- f ) or r-hFSH: r-hLH (Pergoveris) using a GnRH antagonist protocol with hCG trigger. Women with a history of low oocyte maturation or known maturation defect, irregular menstrual cycles ( 35 days), BMI  33, smoking > 10 cigarettes per day, known low ovarian response based on the Bologna criteria [ 21 ], PCOS (polycystic ovary syndrome) defined by revised criteria ASRM ESHRE 2018 [ 22 ], severe endometriosis ≥ III (AFS classification), or combined use of urinary and recombinant gonadotropins in the current ovarian stimulation cycle were excluded from the study. Additionally, patients scheduled for PGT (preimplantation genetic testing), TESE (testicular sperm extraction) or couples with extreme oligo-astheno-teratozoospermia (OAT) with a sperm count below 100,000/ml were also excluded. Study participants were not enrolled in any other concurrent studies. The sample size calculation was based on three assumptions: (i) an ongoing pregnancy rate (OPR) of approximately 50% for fresh blastocyst transfers at UZ Brussels (ii) significant differences in gene expression between the two patient groups, and (iii) the need for at least 20 patients with positive OPR in each group to develop predictive gene expression models for OPR based on previous experiments [ 23 ]. Consequently, 40 patients for each group were initially planned, totalling 80 patients. An interim analysis was conducted as planned after 40 patients had been recruited. Due to an observed lower than expected OPR in both groups, recruitment continued until each group had at least 20 patients with positive OPR outcome. The primary objective was to identify potential differences in the expression of five key biomarkers genes in the CC of the two patient groups and to determine whether a unified prediction model could be applied or if separate models were necessary. The secondary objective was to evaluate the expression of 11 biomarkers for their potential to predict live birth outcomes utilizing cross-validation and linear regression modelling techniques. Ovarian stimulation was conducted using a GnRH antagonist protocol, with patients receiving either r-hFSH (Gonal- f , follitropin alfa) or r-hFSH: r-hLH (Pergoveris, combination of follitropin alfa and lutropin alfa in a 2:1 ratio) at a starting dose of r-hFSH ranging between 150–225 IU. Final oocyte maturation was induced by administering 250 μg of r-hCG (Ovitrelle) when at least 3 follicles ≥ 17 mm were observed by transvaginal ultrasound. Oocyte retrieval (OR) was scheduled 36 h post-trigger. Luteal phase support was provided using micronized vaginal progesterone, initiated according to the centres’ standard of care, and continued until ultrasound confirmation of a foetal sac with a heartbeat at 7 weeks of gestation. Individual denudation was performed on day 0 by enzymatic (Cumulase, Origio, CooperSurgical) and mechanical removal of cumulus cells [ 24 ] within two hours after oocyte retrieval. After individual denudation of all cumulus oocyte complexes (COC), ICSI was performed on all mature oocytes. Sperm motility was assessed according to the Tygerberg criteria [ 25 ]. Mature oocytes were in the second phase of meiosis (metaphase II or MII). Oocytes with two visible pronuclei (2PN) were considered normally fertilised. The collection of cumulus cells has been described before [ 16 ]. All individual cumulus cells were collected and snap-frozen in liquid nitrogen. Cumulus cell samples of all oocytes were transported at −196 °C to the molecular laboratory of Fertiga, Belgium and stored at −80 °C for further molecular analysis. Embryo culture procedures were previously described [ 26 ]. Briefly, ten individual 25μl media droplets were placed in a circular arrangement in an embryo culture dish (IVF round dish, Ø 60 mm, 353801, Falcon) with four additional 25μl media droplets positioned centrally. All droplets were then covered with 7 ml Ovoil (Vitrolife, Sweden). Following preparation, embryo culture dishes were equilibrated overnight in a tri-gas incubator containing 5% O2, 6% CO2, and 89% N2 at 37 °C. Single embryo culture was carried out from day 0 post-ICSI until day 3 in cleavage medium (OS Cleav, 83040010 A, Origio, Denmark). On day 3, embryos were transferred to blastocyst medium (OS Blast, 83060010 A, Origio, Denmark) and cultured until day 5 or 6. Embryo quality was scored using established grading systems. On day 3, cleavage-stage embryos were assessed and categorized as excellent, good, moderate or poor [ 27 ]. On day 5 blastocysts were categorized in the same four groups [ 28 ]. On day 5, all blastocysts were scored according to Gardner and Schoolcraft [ 29 ]. Good quality embryos in the current study are the embryos considered “good” and “fair” according to the ESHRE guidelines [ 30 ]. A single fresh blastocyst was selected for transfer based on these morphological criteria. Supernumerary blastocysts were vitrified individually using CBS high security straws in combination with VitNX (Fujifilm) for use in subsequent frozen embryo transfer cycles [ 31 ]. A βhCG positive pregnancy was defined as the detection of βhCG in the serum, indicating early pregnancy. Clinical pregnancy was defined as a pregnancy diagnosed by ultrasonographic or clinical documentation of at least one foetus with a discernible heartbeat in gestational week 6 to 8. An ongoing pregnancy was defined as a pregnancy defined by the presence of a gestational sac and a detectable foetal heart beat assessed by transvaginal ultrasound between 10 to 12 weeks of gestation. Live birth was defined as the birth of at least one newborn after 24 weeks of gestation who exhibits any sign of life. The CC RNA analysis was conducted at the Fertiga laboratory following the inclusion of all patients in the study. Total RNA was extracted using the RNeasy Micro kit (Qiagen, The Netherlands) with the Qiacube Connect (Qiagen). Subsequently, RNA was reverse transcribed into complementary DNA (cDNA) using the iScript cDNA synthesis kit (BioRad, Belgium) following the manufacturer’s protocol. The cDNA was stored at − 80 °C until further quantitative PCR analysis. Quantitative PCR (qPCR) was performed in triplicate for the following 11 biomarkers: CAMK1D (Calcium/Calmodulin Dependent Protein Kinase ID), EFNB2 (Ephrin B2), SASH1 (SAM And SH3 Domain Containing 1), GOT1 (Glutamic-Oxaloacetic Transaminase 1), SLC6A9 (Solute Carrier Family 6 Member 9), STC2 (Stanniocalcin 2), HAS2 (Hyaluronan Synthase 2), VCAN (Versican), PTGS2 (Prostaglandin-Endoperoxide Synthase 2), HSPH1 (Heat Shock Protein Family H (Hsp110) Member 1), GSTA4 (Glutathione S-Transferase Alpha 4), and for two endogenous control genes: B2M (Beta-2-Microglobulin) and UBC (Ubiquitin C). These genes were selected from our portfolio of patented genes and were previously identified by our group using microarrays in r-hFSH and HP-hMG stimulated patients. Some of these genes were already validated with qPCR in previous published studies [ 15 , 23 , 32 , 33 ]. The qPCR analysis was done in two steps. First, CC from all 1,123 individually denudated oocytes were analysed for five marker genes (CAMK1D, EFNB2, SASH1, GOT1, SLC6A9) and two endogenous control genes (B2M, UBC). Subsequently, CC samples from 341 selected oocytes were analysed for the following six additional marker genes (STC2, HAS2, VCAN, PTGS2, HSPH1, GSTA4). Specificity of amplification during qPCR was assessed using melting curve analysis. Relative quantification of gene expression was performed using standard curves generated from serial dilutions of synthetic oligonucleotides corresponding to each amplicon. Gene expression data were normalized to the mean of the endogenous control genes and log 2 transformed prior to statistical analysis to approximate a Gaussian distribution, facilitating parametric statistical analysis. To detect potential differences between the two patient groups, patient characteristics, embryology, stimulation and cycle outcomes measures were compared with Mann–Whitney tests or two sided Fisher’s exact tests in GraphPad Prism V10. A significance threshold of p  < 0.05 was considered statistically significant. To correct for multiple testing in the Mann–Whitney analyses, a Bonferroni correction was applied adjusting the significance threshold to p  < 0.0025. For the analysis of cumulus cell (CC) gene expression in this study, two distinct cohorts were established based on treatment protocols and oocyte competence. A total of 1,123 CC samples were analysed for five key biomarkers: SASH1, EFNB2, CAMK1D, GOT1, SLC6A9. These genes were selected based on their roles in calcium homeostasis, cell signalling, and metabolic process, which are critical for oocyte maturation and embryo development. The analysis aimed to compare gene expression profiles between two treatment protocols: r-hFSH and a combination of r-hFSH with r-hLH. In the second phase of the study, gene expression in CCs was analysed from a subset of 146 oocytes that resulted in a transferred blastocyst. These CCs were analysed for six additional biomarkers: PTGS2, VCAN, HSPH1, HAS2, GSTA4, and STC2. These genes were chosen for their involvement in prostaglandin synthesis, extracellular matrix remodelling, stress response, and antioxidant defence, all of which are pertinent to oocyte quality and embryo development. By employing these two analytical strategies, the study aims to elucidate the impact of different gonadotropin treatments on CC gene expression and to identify biomarkers predictive of oocyte competence and IVF success. To investigate potential differences between the patient groups, correlation analyses were performed between clinical variables, stimulation, outcome measures and gene expression data (average expressions of all MII per patient). Separate analyses were conducted for patients treated with r-hFSH and those treated with r-hFSH: r-hLH as well as for the combined patient cohort. Prior to the correlation analyses, all parameters were assessed for a normal distribution. Log 2 transformations were applied to gene expression data and via Arc Sin transformations were applied to percentage variables to approximate normal distributions. Normal Q-Q (quantile–quantile) plots of residuals were generated to visually assess the distribution of the data. Despite these transformations not all variables achieved normality. Consequently, Spearman’s rank correlation coefficient r was calculated to evaluate monotonic relationship between variables. A significance threshold was established using the Bonferroni correction to account for multiple comparisons. Given 22 tests, the adjusted significance level was set at p   0.5 and p  < 0.0023 were considered statistically significant and were further analysed. For the primary objective, gene expression data of five key biomarkers (CAMK1D, EFNB2, SASH1, GOT1 and SLC6A9) were analysed across five distinct datasets: (i) CC of all oocytes, (ii) CC of all metaphase II (MII) oocytes, (iii) CC of all normally fertilized (2PN) oocytes, (iv) CC of all fertilized oocytes developing into Good-Quality Embryos (GQE) on day 3, and (v) CC of all fertilized oocytes developing into a GQE on day 5. One-way parametric analyses were performed in GraphPad Prism V10, calculating 95% confidence intervals (CIs) to assess differences between treatment protocols. These analyses aimed to exclude potential biases related to oocyte competence. Additionally, a multivariate analysis of variance (MANOVA) was conducted in R on the same biomarkers across two datasets: (i) CC of all oocytes (the largest dataset) and (ii) CC of all fertilized oocytes developing into a GQE on day 5 (the most specific dataset). For the secondary objective, gene expression data of 11 biomarkers (CAMK1D, EFNB2, SASH1, GOT1, SLC6A9, HAS2, PTGS2, HSPH1, VCAN, GSTA4, STC2) were analysed to identify predictive biomarkers for patients stimulated with r-hFSH as well as with a combination of r-hFSH and r-hLH. The analysis focused on CC from oocytes that either developed into transferable blastocysts leading to live birth or did not. This analysis was an inter-patient analysis. A “leave-one-out” cross-validation approach was employed to mitigate the risk of overfitting and ensure the robustness of the predictive model. Model construction utilized stepwise regression analysis based on a generalized linear mixed model (GLMM), incorporating patient as a random effect and biomarkers as fixed explanatory variables. Both forward and backward stepwise regression analysis was applied to identify the most predictive biomarkers, resulting in multivariate models containing up to four gene expression variables. This approach has been successfully applied in previous studies to identify predictive RNA profiles in fertility research [ 15 , 16 ]. To facilitate visual interpretation of the relation between gene expression and live birth outcome, scatter plots were generated with gene expression on the X-axis and predicted live birth outcomes on the Y-axis. A Locally Weighted Scatterplot Smoothed (LOWESS) line was included to highlight trends in the data. This visualization approach abstracts from other variables to provide a clear representation of the overall positive or negative relationships between gene expression and live birth outcomes.

In this single centre study, 137 patients were initially recruited and signed the informed consent. Of these, 24 patients (10 in the r-hFSH group and 14 in the r-hFSH: r-hLH group) were considered as drop-outs for the study (Fig.  1 ) due to the following reasons: (i) Covid infection ( n  = 2), (ii) freeze-all decision at the time of oocyte retrieval ( n  = 13), (iii) low sperm count ( n  = 1), (iv) BMI > 33 kg/m2 ( n  = 1), (v) withdrawn by the investigator ( n  = 6), and (vi) technical reasons (due to an incorrect sample transport) ( n  = 1). Fig. 1 Flowchart patient inclusion and allocation to the two study groups. For all 113 patients, cumulus cells from all oocytes were individually collected on the day of oocyte retrieval, ICSI was performed and SET was scheduled on day 5 Flowchart patient inclusion and allocation to the two study groups. For all 113 patients, cumulus cells from all oocytes were individually collected on the day of oocyte retrieval, ICSI was performed and SET was scheduled on day 5 The remaining 113 patients were assigned into one of the two study groups: 47 received stimulation with r-hFSH and 66 patients were treated with r-hFSH: r-hLH. For all 113 patients, cumulus cells from all oocytes were denuded/collected individually on the day of oocyte retrieval followed by ICSI. This study aimed to identify potential biomarkers predictive for oocyte competence in two patient groups undergoing ART. To assess the applicability of a common set of biomarkers across both groups, the expression of five key biomarkers was initially compared. To ensure comparability, patient characteristics and stimulation protocols were compared (Table  1 ). Table 1 Patient characteristics & stimulation in the two patient groups Both patient groups were compared with Mann-Whitney tests Bonferroni corrected p values < 0.0025 were significant Patient characteristics & stimulation in the two patient groups Both patient groups were compared with Mann-Whitney tests Bonferroni corrected p values < 0.0025 were significant The ages of the patients were 32 in both the r-hFSH group and in the r-hFSH: r-hLH group (Table  1 ). All participants underwent their first or second ART cycle. For at least 75% of the patients it was their first cycle (Q3 = 0 in Table  1 ) in both patient groups. SET was scheduled on day 5 post-ICSI for all patients. In both treatment groups, similar r-hFSH doses were administered, resulting in comparable serum FSH levels (Table  1 ). However, in the r-hFSH: r-hLH patient group, significantly higher serum LH levels ( p  < 0.0001) and oestradiol (E₂) concentrations ( p  = 0.0002) were observed two days prior to oocyte retrieval. Additionally, the E2/P4 ratio differed significantly between the two groups ( p  < 0.0001) (Mann–Whitney test, Table  1 ). Embryology outcomes are summarized in Table  2 . Clinical outcomes are detailed in Table  3 . No significant differences were observed between the two patient groups regarding the βhCG rate, clinical pregnancy rate, ongoing clinical pregnancy rate, and live birth rate. These comparative analyses of clinical parameters, embryology and clinical outcomes were conducted as post hoc analyses, as detailed in Tables 1 ,  2 and 3 . Table 2 Embryology results in the two patient groups Mean maturation (MII), fertilization and day 5 transferable embryo rates were calculated per patient Embryology results in the two patient groups Mean maturation (MII), fertilization and day 5 transferable embryo rates were calculated per patient Table 3 Outcome of the fresh single embryo transfer in the two patient groups Differences between the two patient groups were not significant Outcome of the fresh single embryo transfer in the two patient groups Differences between the two patient groups were not significant To minimize potential bias between the two patient groups, correlations between clinical patient characteristics and their gene expression data were analysed. In the r-hFSH stimulated group, Spearman correlation analysis revealed a significant positive correlation between serum LH and serum E2 (Spearman r  = 0.54 and p  < 0.0001) (Supplementary Table  1 a). Conversely, in the r-hFSH: r-hLH group, a significant negative correlation was observed between serum AMH and total FSH dose ( r  = −0.55; p  < 0.0001 (Supplementary Table  1 b). These differing correlations may suggest variations in the two ovarian stimulation protocols r-hFSH and r-hFSH: r-hLH and the resulting endocrine environment between the two groups. In the initial phase of the gene expression analysis, 12 out of 1135 CC samples (1%) were excluded due to too low gene expression in at least two of the studied genes. Consequently, 1123 CC samples (532 from the r-hFSH group and 591 from the r-hFSH: r-hLH group) were retained for further analysis, as depicted in the sample inclusion flowchart (Fig.  2 ). Fig. 2 Sample inclusion and analysis flowchart for both patient groups for the first study objective to compare CC gene expression between the two patient groups. Each subgroup was compared with the other, in total 5 comparisons Sample inclusion and analysis flowchart for both patient groups for the first study objective to compare CC gene expression between the two patient groups. Each subgroup was compared with the other, in total 5 comparisons The expression of five key biomarker—EFNB2, CAMK1D, SASH1, GOT1 and SLC6A9—was assessed across five datasets: (i) CC of all oocytes, (ii) CC of all MII oocytes, (iii) CC of all 2PN oocytes, (iv) CC of all oocytes forming a GQE day 3, and (v) CC of all oocytes forming a GQE day 5 (Fig.  2 ). Detailed one-parametric analysis results of these five biomarker genes in the five sample sets are shown in Table  4 . No statistically significant differences were observed in gene expression for CAMK1D, SASH1 and SLC6A9 between the two stimulation protocols across all sample groups. However, EFNB2 exhibited significantly higher expression in the r-hFSH: r-hLH group compared to the r-hFSH group, while GOT1 showed lower expression in the r-hFSH: r-hLH group (Table  4 , Fig.  3 ). Table 4 One-parametric analysis of the gene expression of five biomarkers in CC of r-hFSH and r-hFSH: r-hLH stimulated patients Significant results are shown in bold *All expression values were normalized for the expression of the endogenous controls and log2 transformed One-parametric analysis of the gene expression of five biomarkers in CC of r-hFSH and r-hFSH: r-hLH stimulated patients Significant results are shown in bold *All expression values were normalized for the expression of the endogenous controls and log2 transformed Fig. 3 Estimation plot depicting the expression difference observed when comparing the CC of all oocytes available in both stimulation groups. Violins represent all samples, thin dotted lines represent 25%, 50% (median), 75% quartiles and thick dotted lines represent means. Error bars depict 95%CI Estimation plot depicting the expression difference observed when comparing the CC of all oocytes available in both stimulation groups. Violins represent all samples, thin dotted lines represent 25%, 50% (median), 75% quartiles and thick dotted lines represent means. Error bars depict 95%CI To further investigate the differential expression of the key five biomarkers—EFNB2, CAMK1D, SASH1, GOT1, and SLC6A9 – between the two treatment groups, a multiparametric analysis using MANOVA in R. This approach was applied to two a multiparametric analysis was performed datasets: (i) CC of all oocytes (largest sample set) and (ii) CC of all oocytes forming a GQE on day 5 (most selective sample set). The results of this analysis corroborated the findings from the univariate (one-parametric) analyses, confirming the differential expression patterns observed for EFNB2 and GOT1 between the two treatment groups. For detailed statistical outcomes and further insights, please refer to Supplemental Tables 2a and b. In both the r-hFSH and the r-hFSH: r-hLH treatment groups, no strong correlations were observed between the clinical variables and the expression of the 11 biomarker genes analysed (Supplementary Table 3a and b). This suggests that clinical factors did not significantly influence gene expression levels, ensuring that observed gene expression patters are not confounded by clinical variables. Within each treatment group, several significant correlations were identified among the 11 biomarker genes. In the r-hFSH group, eight significant correlations were found between: SASH1 and CAMK1D ( r  = 0.54; p  < 0.0001), EFNB2 and HAS2 ( r  = 0.68; p  < 0.0001), EFNB2 and VCAN ( r  = 0.60; p  < 0.0001), GOT1 and SLC6A9 ( r  = 0.51; p  = 0.0002), GOT1 and STC2 ( r  = 0.69; p  < 0.0001), GOT1 and GSTA4 ( r  = 0.60; p  < 0.0001), SLC6A9 and STC2 ( r  = 0.50; p  = 0.0004) and HAS2 and VCAN ( r  = 0.56; p  < 0.0001). In the r-hFSH: r-hLH group, ten significant correlations were found between: CAMK1D and EFNB2 ( r  = 0.55; p  < 0.0001), CAMK1D and PTGS2 ( r  = 0.50; p  < 0.0001), EFNB2 and HAS2 ( r  = 0.74; p  < 0.0001), EFNB2 and PTGS2 ( r  = 0.51; p  < 0.0001), EFNB2 and VCAN ( r  = 0.71; p  < 0.0001), GOT1 and STC2 ( r  = 0.54; p  < 0.0001), GOT1 and HSPH1 ( r  = 0.59; p  < 0.0001), GOT1 and GSTA4 ( r  = 0.66; p  < 0.0001), HAS2 and VCAN ( r  = 0.60; p  < 0.0001) and PTGS2 and VCAN ( r  = 0.59; p  < 0.0001). For the secondary study objective, gene expression was analysed on CC from transferred blastocysts in both patient groups to identify the most effective combination of biomarkers, predictive for live birth. In total, 146 CC samples associated with transferred (fresh and frozen transfers) blastocysts were analysed for expression levels of 11 biomarker genes. Multiparametric model building was performed using gene expression data from the following 11 genes: CAMK1D, EFNB2, SASH1, GOT1, SLC6A9, HAS2, PTGS2, HSPH1, VCAN, GSTA4, STC2. Interpatient comparisons were conducted with the endpoint of live birth. The analysis aimed to distinguish gene expression patterns in CCs associated with transferred blastocysts that did or did not result in a live birth. In the group of patients stimulated with r-hFSH, gene expression of 33 CC samples linked to blastocysts that resulted in a live birth was compared with 28 CC samples linked to blastocysts that did not. In the group stimulated with r-hFSH: r-hLH, 36 CC samples associated with live birth were compared with 49 with non-live birth outcomes (Fig.  4 ). Fig. 4 Sample inclusion and analysis flowchart for both patient groups for the second study objective to use CC for the prediction of live birth in the two patient groups Sample inclusion and analysis flowchart for both patient groups for the second study objective to use CC for the prediction of live birth in the two patient groups To identify the most potent biomarkers, a “leave-one-out cross-validation” approach was applied—61 iterations in the r-hFSH group, and 85 iterations in the r-hFSH: r-hLH patient group. Table 5 presents the frequency with which each biomarker gene was selected as one of the top predictors across these iterations. Biomarkers ranked higher in frequency were considered more relevant within the respective patient group. In the r-hFSH group, GOT1, HAS2, SASH1 and PTGS2 emerged as the most predictive genes for live birth. In the r-hFSH: r-hLH group, GOT1, HAS2, EFNB2, STC2 and GSTA4 were identified as key predictors. Notably, GOT1 ranked at the top predictor in both stimulation protocols, followed closely by HAS2 (Table  5 ). Table 5 “Leave-one-out” cross-validation for the r-hFSH patient group ( n  = 61 CC from transferred oocytes) and the r-hFSH: r-hLH patient group ( n  = 85 CC from transferred oocytes) to identify the strongest biomarkers r-hFSH patient group ( n  = 61) r-hFSH: r-hLH patient group ( n  = 85) Gene Frequency of occurrence Gene Frequency of occurrence GOT1 38 GOT1 47 HAS2 38 HAS2 43 SASH1 38 EFNB2 40 PTGS2 37 STC2 26 HSPH1 1 GSTA4 23 VCAN 7 SASH1 2 “Leave-one-out” cross-validation for the r-hFSH patient group ( n  = 61 CC from transferred oocytes) and the r-hFSH: r-hLH patient group ( n  = 85 CC from transferred oocytes) to identify the strongest biomarkers Following this, stepwise linear regression was employed to develop predictive models evaluating multiple combinations of up to four biomarkers in interpatient comparisons. The best performing models predictive of live birth are summarized in Table  6 (for r-hFSH group) and Table  7 (for r-hFSH: r-hLH group). Table 6 Overview of the stepwise linear regression model building for the r-hFSH patient group showing the number of samples, the genes and the predictive value of the live birth models with ROC analysis, showing the area under the curve (AUC) and accuracy of the obtained live birth predictive models Overview of the stepwise linear regression model building for the r-hFSH patient group showing the number of samples, the genes and the predictive value of the live birth models with ROC analysis, showing the area under the curve (AUC) and accuracy of the obtained live birth predictive models Table 7 Overview of the stepwise linear regression model building for the r-hFSH: r-hLH patient group showing the number of samples, the genes and the predictive value of the live birth models with ROC analysis, showing the area under the curve (AUC) and accuracy of the obtained live birth predictive models Overview of the stepwise linear regression model building for the r-hFSH: r-hLH patient group showing the number of samples, the genes and the predictive value of the live birth models with ROC analysis, showing the area under the curve (AUC) and accuracy of the obtained live birth predictive models In the r-hFSH group, the optimal model included GOT1, HAS2, SASH1 and PTGS2 achieving an AUC of 0.7284 and an accuracy of 70% (Fig.  5 a and b). Fig. 5 a ROC (Receiver-operating characteristic) curve for the live birth predictive model with GOT1, HAS2, SASH1 and PTGS2 in the r-hFSH patient group with AUC (area under the curve) of 0.7284 and an accuracy of 70%. b Relations between the gene expression of GOT1, HAS2, SASH1 and PTGS2 and the predicted outcome in r-hFSH patients. These graphs are a two-dimensional representation of the relation of one gene with the probability of live birth, while abstracting the other influencing parameters a ROC (Receiver-operating characteristic) curve for the live birth predictive model with GOT1, HAS2, SASH1 and PTGS2 in the r-hFSH patient group with AUC (area under the curve) of 0.7284 and an accuracy of 70%. b Relations between the gene expression of GOT1, HAS2, SASH1 and PTGS2 and the predicted outcome in r-hFSH patients. These graphs are a two-dimensional representation of the relation of one gene with the probability of live birth, while abstracting the other influencing parameters In the r-hFSH: r-hLH group, the most predictive model was a two-gene model including GOT1 and HAS2, with an AUC of 0.9529 and an accuracy of 88% (Fig.  6 a and b). Fig. 6 a ROC (Receiver-operating characteristic) curve for a live birth predictive model with GOT1 and HAS2 in r-hFSH: r-hLH patients with an AUC (area under the curve) of 0.9529 and an accuracy of 88%. b Relations between the gene expression of GOT1 and HAS2 and the predicted outcome in r-hFSH: r-hLH patients. These graphs are a two-dimensional representation of the relation of one gene with the probability of live birth, while abstracting the other influencing parameters a ROC (Receiver-operating characteristic) curve for a live birth predictive model with GOT1 and HAS2 in r-hFSH: r-hLH patients with an AUC (area under the curve) of 0.9529 and an accuracy of 88%. b Relations between the gene expression of GOT1 and HAS2 and the predicted outcome in r-hFSH: r-hLH patients. These graphs are a two-dimensional representation of the relation of one gene with the probability of live birth, while abstracting the other influencing parameters To assess the potential clinical impact of the newly developed CC-based predictive gene expression models in a standard-of-care context, the models were retrospectively applied to patients in the current study who had multiple good or top-quality embryos available for transfer. In this retrospective application, the CC-based models would have suggested a different top/good-quality blastocyst for the first transfer in 32% of r-hFSH patients (12 out of 37) and 63% of r-hFSH: r-hLH patients (27 out of 43).

For the past four decades, assisted reproductive technologies (ART), including ovarian stimulation combined with in-vitro fertilization (IVF) or intracytoplasmic sperm injection (ICSI), have been utilized to address fertility challenges. These procedures involve single embryo transfer (SET)—either fresh or frozen—or double embryo transfer (DET) on day 3, or SET on day 5, to enhance conception rates in infertile couples. The decision to transfer a specific embryo typically relies on morphological assessment of the embryos. Until today, traditional morphological embryo scoring is still considered the gold standard for ranking embryos for transfer. Invasive embryo biopsy techniques, such as Preimplantation Genetic Testing for aneuploidy (PGT-A), are employed to detect chromosomal abnormalities typically on a subset of blastocyst cells and to inform decisions regarding embryo transfer. However, the utility of PGT-A as a prognostic screening tool remains under debate [ 1 ]. In a randomized clinical trial, the benefit of PGT-A on blastocysts was only demonstrated for patients over 35 years of age [ 2 ], a finding further supported by a meta-analysis [ 3 ]. Concerns regarding the reliability of invasive embryo selection methods, such as trophectoderm biopsy in PGT-A, have also been raised with evidence suggesting that viable embryos may be unnecessarily excluded [ 4 ]. By showing that euploid embryos have a high likelihood of implantation across multiple attempts, the study highlighted the limitations of relying solely on invasive selection techniques and emphasized the need for careful interpretation of PGT-A results. These observations underscore the importance of continued research into alternative, less invasive embryo selection strategies. One such approach, non-invasive PGT-A on spent blastocyst media, has been proposed; however, its reported sensitivity and specificity remain limited [ 5 ], reviewed by [ 6 ]. Time-lapse imaging and artificial intelligence (AI) have been introduced more recently to the embryology lab to optimize the embryo ranking and selection for transfer (review by [ 7 ]. The goal is to reduce the time-to-pregnancy leading to a healthy live birth. However, a Cochrane review on time-lapse systems showed low-quality evidence that time-lapsed imaging increased live birth and clinical pregnancy in the first transfer [ 8 ]. Furthermore, unlike time-lapse imaging, which has been studied both retrospectively and prospectively (including RCTs) over the past decade, AI-based embryo selection is still in the exploratory stage (review by [ 9 ]). Nonetheless, there is still a large need to improve clinical outcomes and shorten the time-to-pregnancy in ICSI cycles, ideally through the application of other non-invasive methods. mRNA expression profiles in cumulus cells have been identified as indicators of oocyte and embryo developmental potential in ART treatments [ 10 , 11 ]. Oocyte developmental competence is defined as the capacity of the oocyte to be fertilized, sustain development to the blastocyst stage and subsequently lead to implantation, clinical pregnancy and/or live birth [ 12 ]. There are multiple advantages to cumulus cell (CC) based oocyte/embryo grading: 1) CC are somatic cells that surround and support the oocyte during its growth and maturation and are as such an excellent surrogate cell source for correlation analysis [ 10 , 13 ]. 2) During an ICSI treatment, CC are removed from the oocyte before injection as part of standard care and are considered ‘waste material’ from the procedure, as such avoiding certain legal and ethical concerns that may be relevant for PGT-A blastocyst biopsies. 3) The test can be performed within one day, and allows transfer in the fresh pick up cycle, while PGT-A requires vitrification of the blastocysts and will delay the embryo transfer. 4) This test may also add critical information on oocyte competence for women considering oocyte freezing for fertility preservation or oocyte donation cycles. 5) Alternatively, the test can be used to identify the most competent oocytes/embryos and limit the number of oocytes to be fertilised/grown and thus limit the number of supernumerary embryos generated (e.g. embryo protection law Germany [ 14 ]. It also could limit the number of embryos needed to be biopsied for PGT-A. 6) Most importantly, the test reduces the stress and burden of failed transfers for the patient by significantly improving the live birth rate per transfer, and shortens time-to-pregnancy. The main disadvantage of the test is that it requires extra hands-on time for the individual denudation of the oocytes, but this is limited and all alternative technologies also require extra expertise and hands-on time for biopsies, extra washing of COC/blastocysts, or extended picture/pattern observations. A minor remark is that the test is currently only applied on ICSI patients. This is to ascertain sample homogeneity and avoid sample drop out due to too few cells, so only for technical reasons. Previously, a non-invasive cumulus cell test predictive of pregnancy was developed and clinically validated by our group for HP-hMG stimulated ICSI patients [ 15 , 16 ]. However, it has been established by our group [ 17 ] and confirmed by others [ 18 , 19 ] that the hormone class used for ovarian stimulation influences the cumulus cell gene expression. Other groups have retrospectively studied numerous cumulus cell predictive biomarkers (reviewed by [ 20 ]). Until now, a similar predictive test for r-FSH stimulated patients has been lacking. For the discovery of r-FSH specific tests, cumulus cells from patients stimulated with either r-hFSH (Gonal- f , follitropin alfa) or r-hFSH: r-hLH (Pergoveris, combination of follitropin alfa and lutropin alfa in a 2:1 ratio) were analysed using gene expression profiles of selected biomarker genes by RT-qPCR. These biomarkers were selected from our patent portfolio of 45 differentially expressed genes from prior discovery work. This prior work involved high throughput screening of independent patients using Affymetrix microarrays, with validation conducted through qPCR (not published). The aim of the current study was to identify CC gene expression profiles predictive for live birth in ICSI patients stimulated with either r-hFSH or r-hFSH: r-hLH. This was achieved by comparing the expression of key biomarkers between both patient groups to determine whether a single predictive model could be applied to both or if separate models would be necessary. Subsequently, cross-validation was employed to identify the most important biomarkers from a set of 11 biomarker genes, and stepwise linear regression was used to construct models for live birth outcomes.

In this study, cumulus cell (CC) gene expression was systematically analysed for the first time in patients undergoing ICSI and day 5 embryo transfer following two types of ovarian stimulation protocols: one using recombinant follicle stimulating hormone alone (r-hFSH, Gonal- f : follitropin alfa), and the other combining recombinant follicle stimulation hormone with recombinant luteinizing hormone (r-hFSH: r-hLH, Pergoveris: combination of follitropin alfa and lutropin alfa). Among five preselected biomarker genes, two (GOT1 and EFNB2) were found to be significantly differently expressed between these two ovarian stimulation protocols. GOT1 expression was consistently higher in CC from r-hFSH-stimulated patients compared to those of the r-hFSH: r-hLH group. This pattern was observed when considering CC of oocytes with different developmental capacities, including all retrieved oocytes, mature MII oocytes, fertilized 2PN oocytes and embryos that developed to day 3 and day 5. In contrast, EFNB2 expression was lower in the r-hFSH group compared to the r-FSH: r-hLH group across the same sets of CC samples. These findings suggest that the presence of recombinant LH in the stimulation regimen has a measurable impact on the transcriptomic profile of cumulus cells. Given that two out of five studied genes show significant differential expression, it is likely that genome-wide transcriptomic analysis would identify a much broader set of differentially expressed genes between the two stimulation protocols. The observed differences are presumably driven by the biological actions of human luteinizing hormone (hLH), which is included in the second stimulation protocol but absent in the r-hFSH-alone regime. hLH plays essential roles in follicular recruitment and maturation, the resumption of meiosis, extrusion of the first polar body and decidualization of endometrial stromal cells, all of which are critical for embryo implantation [ 34 ]. Why the role of r-hLH in ovarian stimulation (OS) across a diverse infertility profile remains a topic of ongoing debate, there is growing evidence supporting its utility in specific patient populations [ 35 , 36 ]. Although the clinical characteristics and stimulation regimens were similar between the two ovarian stimulation groups, significant differences were observed in their hormonal profiles. In the r-hFSH:r-hLH group, serum levels of LH ( p  < 0.0001) and oestradiol (E2; p  = 0.0002) were significantly higher two days prior to oocyte retrieval compared to the r-hFSH-only group. The E2/P4 ratio was also significantly different ( p  < 0.0001). These findings are consistent with earlier studies comparing r-hFSH and HP-hMG stimulation protocols [ 37 ]. Whether the additional r-hLH in the r-hFSH:r-hLH group has the same clinical and molecular effects as the urinary hCG present in HP-hMG (e.g., Menopur) remains to be determined. Furthermore, in the r-hFSH:r-hLH group, a significant negative correlation was found between serum AMH levels and the total FSH dose administered ( r  = −0.55; p  < 0.0001), which is expected as lower AMH levels reflect reduced follicular reserve and increased gonadotropin requirements. No significant correlation was observed between serum LH and E2 concentrations ( r  = 0.23; p  = 0.06), likely due to the presence of exogenous LH in the stimulation regimen. The fact that human luteinizing hormone is changing the expression signature in cumulus cells of patients receiving OS with r-hFSH is in analogy with previous results that gene expression in cumulus differs in women stimulated with highly purified urinated human menopausal gonadotropin (HP-hMG, Menopur) or with r-hFSH [ 17 – 19 ]. Here, one major difference between HP-hMG and r-hFSH is the presence of human chorionic gonadotropin in the first. The differential expression of GOT1 and EFNB2 in response to stimulation protocols differing only in the presence of r-hLH is particular compelling, as both genes have already been linked to oocyte competence in independent studies [ 15 , 23 , 32 , 33 ] as well as our own unpublished micro array and qPCR data. Notably, both genes were also retained as predictive biomarkers in our live birth prediction models. EFNB2, a transmembrane ligand in the ephrin family, is predominantly expressed during luteinization in granulosa cells [ 38 ]. Its receptor, EPHB2, is regulated by FSH and the oestrogen receptor beta [ 39 ]. Increased EFNB2 expression may promote improved follicular vascularization [ 40 ] thereby supporting oocyte competence. Moreover, expression of the EFNB2 receptor in the CCs has also been associated with oocyte/embryo competence and euploidy [ 41 ]. GOT1 (glutamic-oxaloacetic transaminase 1) plays a central role in the malate-aspartate shuttle, converting glutamate to aspartate using oxaloacetate. This metabolic process supports redox homeostasis and NAD + regeneration during oocyte maturation [ 42 ]. Aspartate, a non-essential amino acid, is among the most consumed amino acid in mouse, cow, and pig blastocysts [ 43 – 47 ]. Increased GOT1 expression may therefore enhance cellular redox status and NAD + availability, contributing to oocyte development potential [ 45 , 48 , 49 ]. While the exact mechanism underlying the differential CC expression of GOT1 and EFNB2 in response r-hLH exposure is currently unclear, our findings add to the growing body of evidence supporting the functional importance of gonadotropin-induced transcriptional changes in CCs and their potential utility in improving embryo selection strategies. As a next step, predictive gene expression models for live birth were systematically constructed using a panel of 11 biomarker genes and evaluated for their performance using linear regression methods. Model performance was assessed based on area under the curve (AUC) and overall accuracy. Given the previously observed difference in gene expression profiles between the two ovarian stimulation protocols, a stimulation group-specific approach was adopted for model development. The 11 genes evaluated were CAMK1D, EFNB2, SASH1, GOT1, SLC6A9, HAS2, PTGS2, HSPH1, VCAN, GSTA4, and STC2, all previously identified as expression biomarkers informative of oocyte competence. For model construction, gene expression data from cumulus cells of oocytes that developed into blastocysts and resulted live births were compared to those from blastocysts that failed to yield live births after transfer. ‘Leave-one-out’ cross validation identified the most frequently occurring genes in predictive models. In the r-hFSH-alone group, the most consistently selected genes were GOT1, HAS2, SASH1, and PTGS2 (Table  5 ). In the r-hFSH: r-hLH group, the top genes were GOT1, HAS2, EFNB2, STC2, GSTA4 and VCAN (Table  5 ). The best performing model for r-hFSH patient group consisted of four genes (GOT1, HAS2, SASH1, PTGS2), yielding an AUC of 0.7284 and an overall accuracy of 70%. In the r-hFSH: r-hLH group, the optimal model was a two-gene combination (GOT1 and HAS2), achieving an AUC of 0.9529 and an overall accuracy of 88%. The superior performance in the r-hFSH: r-hLH group likely reflects the availability of approximately 30% more informative samples (CCs of transferred embryos with known outcomes) in that group. Correlation analysis of gene expression data further informed model refinement by identifying genes that likely function within shared regulatory pathway. Genes showing high co-correlation are less suitable for inclusion together in a predictive model due to potential redundancy. For instance, while GOT1 correlated with three other genes, it was not correlated with HAS2 and EFNB2, supporting their combined inclusion in predictive models. Additionally, three gene pairings (GOT1 vs HPSH1; SASH1 vs CAMK1D; SLC6A9 vs STC2) exhibited differing correlation patterns between the two patient groups, potentially reflecting stimulation protocol-specific effects (e.g. elevated GOT1 expression in the r-hFSH group). GOT1 and HAS2 were among the strongest predictors of live birth across both stimulation protocols. However, the correlation of HAS2 with live birth varied by group, indicating protocol-dependent dynamics. A similar phenomenon has been described previously for PTGS2, where studies reporting opposing associations with oocyte competence depending on whether patients were stimulated with rFSH [ 50 ] or HP-hMG which includes both FSH and LH components [ 10 ]. This apparent contradiction was further validated in a comparative study involving both patient populations [ 17 ], underscoring the need for stimulation-specific predictive models in CC-based analysis. HAS2 encodes hyaluronan synthase 2, an enzyme critical for hyaluronan acid production and cumulus cell expansion, processes that are crucial during oocyte maturation. HAS2 is upregulated in response to GDF-9, a key oocyte-secreted factor, and supports the structural remodelling required for optimal oocyte function [ 51 , 52 ]. Previous work by our group linked HAS2 expression in CCs to increased rates of 2PN fertilization, suggesting it may serve as a marker of oocyte maturity or over maturity [ 17 ]. SASH1 (SAM and SH3 domain-containing protein 1) belongs to the SLY1-family of signal adapter proteins. Unlike most family members which are haematopoietic, SASH1 is ubiquitously expressed including in the ovary, uterus, placenta, prostate and testis. This gene is also included in the Aurora Test, a pregnancy prediction tool prospectively validated for HP-hMG-stimulated patients [ 15 , 16 ]. SASH1 has been implicated in the Toll-like receptor 4 pathway [ 53 ] and is known to be LH-responsive in bovine granulosa cells [ 54 ]. PTGS2 (prostaglandin-endoperoxide synthase 2) has long been associated with oocyte maturation and cumulus expansion and has been identified as predictive CC biomarker by multiple research groups [ 10 , 13 , 55 – 57 ]. Its inclusion in the predictive model for the r-hFSH group reinforces its value in the context of FSH-driven folliculogenesis. Taken together, these stimulation-specific predictive biomarkers reflect the diverse molecular pathways involved in cumulus-oocyte signalling and highlight distinct mechanisms through which a competent oocyte and, ultimately, a viable embryo can be generated to result in clinical pregnancy and live birth. The CC based live birth prediction models developed here would have suggested a different top/good-quality blastocyst for the first transfer in 32% of r-hFSH patients of the current study. This finding aligns with previous observations from an unbiased, interventional prospective study using the Aurora Test in HP-hMG stimulated patients, where only 27% concordance was found between embryo ranking based on morphology and ranking based on CC gene expression analysis. In that study prioritizing the transfer of day 3 embryos based on the CC analysis resulted in a doubling of the pregnancy rate compared to the controls with transfer based on morphology only [ 15 , 16 ]. This confirms the impact a CC based oocyte/embryo grading can have on routine ART. This clinical study was an observational cohort study and not a randomized clinical trial and was conducted between 2021 and 2023 at a single tertiary university IVF centre. Patients were enrolled until at least 20 patients with a confirmed ongoing pregnancy were included in both study groups, which was necessary to support the development of predictive models for the secondary study objective. One limitation is that equal recruitment across study arms was not achieved. Ultimately, 19 more patients were included in the r-hFSH: r-hLH arm. It remains unclear whether the observed difference between the two patient groups in the number of COC and MII oocytes between the two groups reflect a clinically relevant effect, a difference in the baseline characteristics or simply a random variation due to the limited patient number. The CC gene expression-based prediction models developed here are intended to be used in conjunction with standard morphological blastocyst grading. Efforts to develop models that predict oocyte competence independently of morphologic grading are currently ongoing. This study should be regarded as a discovery study. While the gene expression models demonstrated high predictive accuracy, particularly in both ovarian stimulation protocols, the findings are based on a non-randomized, single-centre design and should be interpreted accordingly. Implementation of CC-based gene expression analysis in clinical practice would require individual denudation of oocytes, which is not current standard practice. While oocyte denudation is typically performed in groups for efficiency, individual denudation adds approximately 30 min per patient, representing a manageable but relevant increase in embryology lab workload. Finally, while several of the biomarker genes used in the new predictive models such as GOT1, HAS2, SASH1 and PGTS2 have been previously associated with oocyte competence or cumulus function, their exact mechanistic roles remain to be fully elucidated. Although a complete understanding of their function is not necessary for the utilization of the predictive models, further functional studies would strengthen the biological foundation.

This study identified novel cumulus cell gene expression models that predict oocyte competence and are associated with live birth outcomes in patients undergoing ART. Biomarker expression profiles differed between patients stimulated with r-hFSH alone and those receiving r-hFSH in combination with r-hLH, underscoring the need for stimulation-specific predictive approaches. The most robust gene models achieved an AUC between 0.72 and 0.95, with an overall accuracy for predicting live birth ranging from 70 to 88%. Given their potential to improve embryo selection, increase live birth per transfer and shorten the time-to-pregnancy in ICSI cycles, these models merit validation in prospective, randomized clinical trials.

Supplementary Material 1. Supplementary Table 1a: Spearman correlation analysis for patient characteristics in the r-hFSH stimulated group. Correlations with Spearman r >0.5 or <-0.5 and p 0.5 or <-0.5 and p < 0.0023 were highlighted. Supplementary Table 2a and b: The analysis was repeated in a multiparametric approach (MANOVA, using R) for the largest ( n =1123, Table 2a) and the smallest sample set ( n =285, Table 2b). The results obtained for “CC of all oocytes ( n =1123)” and “CC of all oocytes forming a GQE on day 5 ( n =285)” were comparable. The expression in the two treatment groups was statistically different. This difference was detectable for EFNB2 and GOT1 but also for CAMK1D and SLC6A9. The largest differences observed were between 0.5 and 0.75. As this were log2 values actual differences were between 1.4-1.7 fold or 40% to 70% higher/lower expression between the two stimulation groups. Supplementary Table 2a: MANOVA results when comparing the CC gene expression in “All oocytes” sample set. P values are enumerated in the columns for 6 hypothesis (difference between the 2 groups ><0, <0.1, <0.5, <0.75, <1, <1.5log2), and the analysis was repeated for all combinations comprising 1, 2, 3, 4 or 5 gene expressions. All analyses were performed in R. The green line (indicating significance) is added for the ease of interpretation of the table. The first column detects difference between the 2 treatment groups. If this is 0=significant then the 5 columns to the right are considered. These allow an estimation on how large the difference is between the 2 treatments. Blue arrows indicate the largest differences observed when considering the expression of only 1 or 2 or 3 genes. Supplementary Table 2b: MANOVA results when comparing the CC gene expression in “All oocytes forming a GQE on day 5” sample set. P values are enumerated in the columns for 6 hypothesis (difference between the 2 groups ><0, <0.1, <0.5,<0.75, <1, <1.5 log2), and the analysis was repeated for all combinations comprising 1, 2, 3, 4 or 5 gene expressions. All analyses were performed in R. The green line (indicating significance) is added for the ease of interpretation of the table. The first column detects difference between the 2 treatment groups. If this is 0=significant then the 5 columns to the right are considered. These allow an estimation on how large the difference is between the 2 treatments. Blue arrows indicate the largest differences observed when considering the expression of only 1 or 2 or 3 genes. Supplementary Table 3a: Spearman correlation analysis for patient characteristics and gene expression data in the r-hFSH stimulated group. Correlations with Spearman r >0.5 or <-0.5 and p < 0.0023 were highlighted and discussed in detail. Supplementary Table 3b: Spearman correlation analysis for patient characteristics and gene expression data in the r-hFSH: r-hLH stimulated group. Correlations with Spearman r >0.5 or <-0.5 and p < 0.0023 were highlighted and discussed in detail. Supplementary Material 1. Supplementary Table 1a: Spearman correlation analysis for patient characteristics in the r-hFSH stimulated group. Correlations with Spearman r >0.5 or <-0.5 and p 0.5 or <-0.5 and p < 0.0023 were highlighted. Supplementary Table 2a and b: The analysis was repeated in a multiparametric approach (MANOVA, using R) for the largest ( n =1123, Table 2a) and the smallest sample set ( n =285, Table 2b). The results obtained for “CC of all oocytes ( n =1123)” and “CC of all oocytes forming a GQE on day 5 ( n =285)” were comparable. The expression in the two treatment groups was statistically different. This difference was detectable for EFNB2 and GOT1 but also for CAMK1D and SLC6A9. The largest differences observed were between 0.5 and 0.75. As this were log2 values actual differences were between 1.4-1.7 fold or 40% to 70% higher/lower expression between the two stimulation groups. Supplementary Table 2a: MANOVA results when comparing the CC gene expression in “All oocytes” sample set. P values are enumerated in the columns for 6 hypothesis (difference between the 2 groups ><0, <0.1, <0.5, <0.75, <1, <1.5log2), and the analysis was repeated for all combinations comprising 1, 2, 3, 4 or 5 gene expressions. All analyses were performed in R. The green line (indicating significance) is added for the ease of interpretation of the table. The first column detects difference between the 2 treatment groups. If this is 0=significant then the 5 columns to the right are considered. These allow an estimation on how large the difference is between the 2 treatments. Blue arrows indicate the largest differences observed when considering the expression of only 1 or 2 or 3 genes. Supplementary Table 2b: MANOVA results when comparing the CC gene expression in “All oocytes forming a GQE on day 5” sample set. P values are enumerated in the columns for 6 hypothesis (difference between the 2 groups ><0, <0.1, <0.5,<0.75, <1, <1.5 log2), and the analysis was repeated for all combinations comprising 1, 2, 3, 4 or 5 gene expressions. All analyses were performed in R. The green line (indicating significance) is added for the ease of interpretation of the table. The first column detects difference between the 2 treatment groups. If this is 0=significant then the 5 columns to the right are considered. These allow an estimation on how large the difference is between the 2 treatments. Blue arrows indicate the largest differences observed when considering the expression of only 1 or 2 or 3 genes. Supplementary Table 3a: Spearman correlation analysis for patient characteristics and gene expression data in the r-hFSH stimulated group. Correlations with Spearman r >0.5 or <-0.5 and p < 0.0023 were highlighted and discussed in detail. Supplementary Table 3b: Spearman correlation analysis for patient characteristics and gene expression data in the r-hFSH: r-hLH stimulated group. Correlations with Spearman r >0.5 or <-0.5 and p < 0.0023 were highlighted and discussed in detail.