Epigenetic age and fertility timeline: testing an epigenetic clock to forecast in vitro fertilization success rate

Of the 400 women recruited, 11 did not start the IVF cycle for personal reasons and 10 did not provide blood sample for the DNA methylation study, leaving 379 for data analysis. A total of 204 women (54% of our cohort) achieved a live birth. Table  1 lists the clinical characteristics of the entire cohort and a comparison of women who did and did not achieve a live birth. Age, body mass index (BMI), FSH, AMH, period of infertility duration, previous deliveries, smoking and indications for access to ART procedures were examined. The analysis of the two groups showed that the women who were successful were significantly younger and with better ovarian reserve variables (higher AFC and AMH) with lower BMI and duration of infertility than those without the live birth, as expected. Table 1 Baseline clinical characteristics of the entire cohort and comparison of women who achieved and did not achieve a live birth Characteristics Whole cohort Live birth No Live birth p n  = 379 n  = 204 n  = 175 Age (years) 36 ± 4 35 ± 4 37 ± 4 < 0.001 BMI (Kg/m 2 ) 22.9 ± 4.2 22.6 ± 3.9 23.4 ± 4.4 0.05 AMH (ng/mL) 2.4 ± 1.9 2.6 ± 1.9 2.1 ± 1.9 0.02 AFC 12 ± 8 13 ± 8 11 ± 9 0.002 Duration of infertility (months) 38 ± 26 35 ± 21 42 ± 31 0.03 Previous deliveries 37 (10%) 16 (8%) 21 (12%) 0.22 Smoke 65 (17%) 31 (15%) 34 (19%) 0.28 IVF indications 0.07  Unexplained 157 (41%) 93 (46%) 64 (37%)  Endometriosis 43 (11%) 17 (8%) 26 (15%)  Tubal factor 53 (14%) 32 (16%) 21 (12%)  Ovulatory disorder 16 (4%) 10 (5%) 6 (3%)  Uterine factor 4 (1%) 1 (0.5%) 3 (2%)  Male factor 78 (21%) 41 (20%) 37 (21%)  Mixed 28 (7%) 10 (5%) 18 (10%) Data are reported as mean ± SD or number (percentage); AFC: Antral Follicle Count; AMH: Anti-mullerian hormone; IVF: in vitro fertilization Baseline clinical characteristics of the entire cohort and comparison of women who achieved and did not achieve a live birth Data are reported as mean ± SD or number (percentage); AFC: Antral Follicle Count; AMH: Anti-mullerian hormone; IVF: in vitro fertilization In Table  2 , we reported the outcomes of the first IVF cycle in the two groups identified (those who achieved or not the live birth). The type of ovarian stimulation protocol (GnRH antagonist, Long and Flare-up), gonadotropin dose, duration of stimulation, number of oocytes retrieved, number of mature ones, fertilization rate, number of embryos at 72 h were examined. Women who delivered had a significantly higher number of retrieved oocytes, suitable oocytes, and embryos available at the cleavage stage. Table 2 IVF cycle outcome in the whole cohort and comparison between women who did and did not obtain a live birth Characteristics Whole cohort Live birth No Live birth p n  = 379 n  = 204 n  = 175 Protocol 0.53  GnRH antagonist 357 (94%) 194 (96%) 163 (93%)  Long protocol 9 (2%) 5 (2%) 4 (2%)  Flare up protocol 13 (3%) 5 (2%) 8 (5%) Total dose of gonadotropins (IU) 1670 ± 516 1622 ± 456 1724 ± 575 0.06 Duration of stimulation (days) 8 ± 2 9 ± 2 8 ± 2 0.14 Cancelled cycles 35 (9%) 11 (5%) 24 (14%) 0.007 N of oocytes retrieved 7 ± 6 8 ± 7 6 ± 6 < 0.001 N of suitable oocytes 6 ± 5 7 ± 6 5 ± 4 < 0.001 Fertilization rate (%) 0.70 ± 0.30 0.71 ± 0.26 0.67 ± 0.34 0.23 N cleavage stage embryos (72 h) 3 ± 2 4 ± 2 3 ± 2 0.001 Data are reported as mean ± SD or number (percentage); IU: International Units; N cleavage stage embryos (72 h) refers to the number of embryos in D3 which met adequate morphological assessment IVF cycle outcome in the whole cohort and comparison between women who did and did not obtain a live birth Data are reported as mean ± SD or number (percentage); IU: International Units; N cleavage stage embryos (72 h) refers to the number of embryos in D3 which met adequate morphological assessment Then, we analysed epigenetic age and EPA in women who did and did not obtain a live birth, and a difference emerged only for epigenetic age. Specifically, the mean ± SD epigenetic age was 36 ± 5 and 39 ± 5 years respectively ( p  < 0.001) for women who did and did not have a live birth. EPA was − 0.3 ± 4.0 and + 0.3 ± 3.8 years, respectively ( p  = 0.12). We proceeded to perform a multivariate analysis adjusting for the variables which were found to be different between the two groups (Table  3 ). In Model I we corrected for the baseline pre-treatment variables of ovarian reserve (AFC), BMI and duration of infertility and in Model II we also corrected for the number of oocytes collected at the oocyte retrieval. We did not also correct for AMH because of the high collinearity with AFC. Table 3 Univariate and multivariate analyses on the relation between epigenetic aging and live birth Variable Univariate analysis Model I a Model II b OR 95% CI p Adj OR 95% CI p Adj OR 95% CI p DNAm age 0.91 0.87–0.95 < 0.001 0.91 0.86–0.96 < 0.001 0.92 0.87–0.97 0.001 AgeAcc 0.96 0.91–1.01 0.12 0.96 0.910–1.01 0.14 0.97 0.91–1.03 0.32 OR: odds ratio; CI: confidence interval; DNAm age: epigenetic age; ageacc: age acceleration; Ia) data was adjusted for AFC , BMI , and duration of infertility using a multivariate logistic regression model; IIb) data was adjusted for AFC , BMI , duration of infertility , and oocyte yield using a multivariate logistic regression model Univariate and multivariate analyses on the relation between epigenetic aging and live birth OR: odds ratio; CI: confidence interval; DNAm age: epigenetic age; ageacc: age acceleration; Ia) data was adjusted for AFC , BMI , and duration of infertility using a multivariate logistic regression model; IIb) data was adjusted for AFC , BMI , duration of infertility , and oocyte yield using a multivariate logistic regression model Adjusting for the ovarian reserve variables, epigenetic age significantly differed between the two groups, with an adjusted odds ratio (adj. OR) of live birth per year of 0.91 (95%CI: 0.87–0.96, p  < 0.001). Its meaning was that for every additional year of DNAm Age, the likelihood of achieving a live birth decreased by 9%. This effect was independent of traditional markers such as AFC. This association remained significant even after adjusting for both baseline ovarian reserve and IVF response (oocyte yield), with adjusted OR of live birth per year of epigenetic age of 0.92 (95%CI: 0.88–0.97, p  = 0.001). Investigating the association between the epigenetic clock and ovarian response, a modest but statistically significant negative association between DNAm age and the number of oocytes retrieved was observed (ρ = -0.258; 95% CI [-0.394, -0.122], p  < 0.001). However, this association was no longer significant in the multivariate model. No significant associations were found between DNAm age or EPA and the number of embryos obtained or oocyte yield (Table S1 ). To further assess the accuracy of epigenetic age and EPA in predicting live birth rates, we assessed the ROC curves (Fig.  1 ), which showed an AUC of 0.652 (95%CI: 0.593–0.711) and of 0.558 (95%CI: 0.496–0.620) respectively. Similar result was found for chronological age, with an AUC of 0.675 (95%CI: 0.616–0.733). The performance of epigenetic age tended to be superior to ovarian reserve variables, as, AFC and AMH had respectively an AUC of 0.637 (95%CI: 0.576–0.698) and 0.603 (95%CI: 0.542–0.665). However, the 95%CIs overlap, hampering robust conclusion. Fig. 1 Receiver operating characteristic (ROC) curve of the live birth prediction model for ART (chronological age, AFC, AMH, epigenetic age, EPA) Receiver operating characteristic (ROC) curve of the live birth prediction model for ART (chronological age, AFC, AMH, epigenetic age, EPA) The possible difference observed was definitely lost when we proceeded to subgroup analyses according to infertility cause (Table  4 ). Table 4 Receiver operating characteristic (ROC) curve of the live birth prediction model for ART according to infertility cause Characteristics n AFC DNAm Age Age acceleration Chronological Age Whole cohort 379 0.637 [0.576–0.698] 0.652 [0.593–0.711] 0.558 [0.496–0.620] 0.675 [0.616–0.733] Unexplained 157 0.663 [0.572–0.753] 0.661 [0.575–0.748] 0.506 [0.413–0.599] 0.780 [0.703–0.857] Endometriosis 43 0.766 [0.624–0.908] 0.643 [0.470–0.815] 0.505 [0.328–0.681] 0.756 [0.610–0.901] Tubal factor 53 0.515 [0.343–0.686] 0.592 [0.422–0.762] 0.604 [0.440–0.769] 0.538 [0.370–0.706] Male factor 78 0.540 [0.409–0.671] 0.747 [0.635–0.860] 0.592 [0.464–0.720] 0.732 [0.616–0.848] Mixed 28 0.614 [0.384–0.843] 0.689 [0.486–0.892] 0.550 [0.329–0.771] 0.669 [0.471–0.868] [95%CI]: 95% confidence interval; AFC: antral follicular count; DNAm age: epigenetic age; ageacc: age acceleration Receiver operating characteristic (ROC) curve of the live birth prediction model for ART according to infertility cause [95%CI]: 95% confidence interval; AFC: antral follicular count; DNAm age: epigenetic age; ageacc: age acceleration We then investigated whether epigenetic age performance was superior to other parameters considering chronological age subgroups. Diving the cohort in 4 groups (≤ 30 n  = 31; 31–35 n  = 120; 36–39 n  = 148; ≥40 n  = 80), we noted that epigenetic age and EPA resulted the predictor with the highest accuracy in the second group (age between 31 and 35 years) (Table  5 ). In this group, AUCs of epigenetic age and EPA were 0.637 (95%CI: 0.519–0.755) and 0.637 (95%CI: 0.523–0.751), respectively. The AUCs of the other variables were less accurate: the AUCs for chronological age, AMH and AFC were 0.561 (95%CI: 0.423–0.699), 0.573 (95%CI: 0.440–0.706), and 0.550 (95%CI: 0.412–0.689), respectively. Considering the other age subgroups, we observed that in the youngest group (≤ 30 years) the most accurate parameter was AFC (AUC of 0.607, 95%CI: 0.356–0.859); AFC was the best predictor also in case of women between 36 and 39 years (AUC of 0.642, 95%CI: 0.546–0.737), while in older population (≥ 40 years), chronological age had the best performance (AUC of 0.625, 95%CI: 0.487–0.763). Table 5 Receiver operating characteristic (ROC) curve of the live birth prediction model for ART according to age Age n AFC AMH DNAm Age Age acceleration Chronological Age ≤ 30 31 0.607 [0.356–0.859] 0.556 [0.262–0.849] 0.511[0.256–0.766] 0.541 [0.281-0.800] 0.593 [0.364–0.821] 31–35 120 0.550 [0.412–0.689] 0.573 [0.440–0.706] 0.637 [0.519–0.755] 0.637 [0.523–0.751] 0.561 [0.423–0.699] 36–39 148 0.642 [0.546–0.737] 0.552 [0.454–0.651] 0.574 [0.475–0.672] 0.571 [0.474–0.669] 0.555 [0.458–0.652] ≥ 40 80 0.596 [0.449–0.742] 0.602 [0.451–0.754] 0.520 [0.361–0.678] 0.541 [0.383–0.698] 0.625 [0.487–0.763] [95%CI]: 95% confidence interval; AFC: antral follicular count; DNAm Age: epigenetic age; AgeAcc: age acceleration Receiver operating characteristic (ROC) curve of the live birth prediction model for ART according to age [95%CI]: 95% confidence interval; AFC: antral follicular count; DNAm Age: epigenetic age; AgeAcc: age acceleration At last, when we performed an exploratory analysis to assess the accuracy of a multiparametric model in predicting cumulative LBR compared with chronological age alone, we obtained the following combination: Y1 = 5.300 + (− 0,149* age) + (− 0.043*EPA) + (0.021*AFC) Y2 = 5,621 + (− 0,157* age) + (− 0.062*EPA) + (0.109*AMH) As shown in Fig.  2 , in both cases, the multiparametric model with chronological age, EPA and AFC (Y1) or AMH (Y2) slightly improved the predictive accuracy compared with only chronological age: AUC for Y1 was 0.692 (95%CI: 0.637–0.67) and for Y2 was 0.693 (95%CI: 0.635–0.750) compared with the AUC for chronological age of 0.672 (95%CI: 0.613–0.73). Fig. 2 C-statistics testing the multiparametric model to predict live birth. (with AFC on the left panel; with AMH on the right panel) C-statistics testing the multiparametric model to predict live birth. (with AFC on the left panel; with AMH on the right panel)

The project was a single-centre observational prospective study that has been conducted at the Infertility Unit of Fondazione Ca’ Granda, Ospedale Maggiore Policlinico, Milan, Italy. All procedures were in accord with the Helsinki Declaration and all participants provided written informed consent for sample and data use. Ethical approval for the study was obtained from the local Ethical Committee (Comitato Etico Milano 1034_2022bis). Women of reproductive age who underwent IVF treatment, with no age restrictions, were recruited. Inclusion criteria included 1. indication to IVF; FSH < 15 IU/ml; no previous IVF cycles; while exclusion criteria were (1) a severe male factor (< 1 million/mL); and (2) a relevant systemic diseases that may adversely affect the course of pregnancy (e.g. diabetes, uncompensated thyroid disorder, antiphospholipid antibody syndrome). Eligible patients were enrolled at the time of their first fertility clinic visit prior to initiate IVF, and informed consent was obtained prior to the collection of biological samples, from 01 March 2022 to 01 June 2023. Controlled ovarian hyperstimulation (COH) was performed using either a gonadotropin-releasing hormone (GnRH) antagonist or long agonist protocol or flare up, according to patient characteristics and clinician judgment. In the long protocol, GnRH agonist was started in the mid-luteal phase of the preceding cycle to achieve pituitary downregulation before starting stimulation. In the flare-up protocol, low-dose GnRH agonist was started on day 2 or 3 of the cycle to exploit the initial FSH surge, followed by gonadotropin administration. Recombinant FSH or human menopausal gonadotropin (HMG) was administered starting on day 2 or 3 of the cycle in case of GnRH antagonist protocol or flare up. Once at least three follicles ≥ 17 mm was observed, ovulation was triggered with human chorionic gonadotropin (hCG) or GnRH agonist, followed by oocyte retrieval 36 h later. During the study period, cycles could be cancelled if follicular development was abnormal or for personal reasons. Single embryo transfer was performed at the cleavage stage in case of fresh embryo transfer, and at the blastocyst stage in case of frozen embryo transfer. Embryo transfer or cryopreservation was carried out only when the embryo met the minimum quality standards established by the laboratory, based on morphological assessment and according to the criteria of the Istanbul Consensus [ 17 ]. Baseline clinical characteristics and IVF outcome of the selected women were obtained from patients’ charts. During standard blood draws, a single tube of whole blood in additional ethylenediaminetetraacetic acid (EDTA) was collected before the patient underwent ovarian stimulation protocols. They were collected in EDTA-containing tubes and immediately stored at -80˚ C until assayed. Molecular analyses were performed simultaneously after completing patients’ recruitment and follow-up. Once recruitment was completed, the tubes were thawed, and DNA extracted from white blood cells using the DNeasy Blood & Tissue Kit (QIAGEN) [ 18 ]. In brief, after bisulfite conversion and subsequent PCR amplification, the samples underwent pyrosequencing analysis [ 18 ]. DNAm age was calculated considering the methylation pattern of CpG sites at five genes (ELOVL2, C1orf132/MIR29B2C, FHL2, KLF14, TRIM59) [ 17 ]. The “Zbieć-Piekarska2” epigenetic-age model is a simplified epigenetic clock that estimates biological age based on the methylation levels of five specific CpG sites (ELOVL2, C1orf132, TRIM59, KLF14, and FHL2). It was originally developed for forensic purposes due to its practicality, cost-effectiveness, and relatively good accuracy across tissues. In our study, we adopted this model as it allows DNAm analysis through pyrosequencing, suitable for clinical settings and able to capture biologically relevant age-related variation. Importantly, our laboratory had previously adopted this model in other studies [ 16 ], developing strong technical confidence and expertise in its implementation, which ensured reproducibility and reliability of the results in this context. Epigenetic age (Y) was calculated as follows: Y = 3,26847784754751817 + 0,465445549010653 methC7_ELOVL2 − 0,355450171437202 methC1_C1orf132 + 0,306488541137007 methC7_TRIM59 + 0,8326844353523238792 methC1_KLF14 + 0,237081243617191 methC2_FHL2 The sample size was based on the previous study that found a difference of 1.2 years in epigenetic age between ART women who did or did not succeed [ 16 ]. Estimating that approximately one third of the patients will be successful in the procedure, in terms of live birth, with type I of 0.05 and a power of 80%, 400 women had to be recruited. The main outcome was the cumulative LBR over three oocyte retrieval (thus including also pregnancies obtained from cryopreserved gametes or embryos). In our centre, this outcome is typically considered the standard measure, as it provides a more comprehensive assessment of a patient’s chance to achieve a live birth. Live birth was defined as the delivery of at least one viable newborn after 24 gestational weeks. A descriptive analysis of the baseline characteristics of the population was carried out. Where the data had a normal distribution, the mean value and standard deviation were reported, otherwise median and quartiles. For qualitative variables, the count and percentage were reported. Statistical analyses involved the application of Student’s t test, Mann-Whitney and Fisher’s exact test, performed with SPSS 27.0 software. To obtain an estimate of biological ageing independent of chronological one, a measure defined as age acceleration was used. To calculate this, a linear regression model was applied with chronological age as the independent variable and epigenetic age as the outcome. The residual of this statistical model (e.g. the difference between the observed value of biological age and that predicted by the model) constitutes the age acceleration due to epigenetic effects. If the epigenetic age was greater than the chronological age, the age acceleration had a positive value expressed in years, negative if opposite. By constructing a receiver operating characteristic (ROC) curve, it was possible to observe the predictive value of epigenetic age in relation to the outcome of the IVF procedure. Using a univariate linear regression model, it was assessed whether age acceleration and epigenetic age were associated with the outcome of the IVF procedure. Again, using a univariate linear regression model, it was assessed whether it could be informative on ovarian cycle response. In case of significant differences in baseline characteristics, a multivariate logistic regression model was adopted to obtain adjusted measures of association between estimated epigenetic age and success rate (LBR). Exploratory analysis stratifying for cause of infertility and age interval were then led. Finally, we developed a multiparametric model incorporating the most informative factors (chronological age, epigenetic age acceleration (EPA), and ovarian reserve markers) along with their respective coefficients (beta). We then assessed its accuracy in predicting whether a woman would achieve a live birth after IVF. A logistic regression was applied using Y, the variable resulting from the combination of all factors in the model, as the sole predictive variable. The resulting value was: Y1 = b0 + (B 1 * chronological age) + (B 2 * EPA) + (B 3 * AFC) Y2 = b0 + (B 1 * chronological age) + (B 2 *EPA) + (B 3 * AMH) We applied the c-statistics, evaluated using the area under the curve (AUC), to assess the effectiveness of the prediction models. A live birth was predicted when the calculated probability from the logistic regression was greater than 0.5. The models’ performance was compared by analysing the predictive probabilities of a live birth generated by this multiparametric model with the one provided by the chronological age alone. The accuracy was then evaluated in the predefined age subgroups. A binomial distribution model was used to calculate 95% of confidence interval of proportions. P values below 0.05 were considered statistically significant.

The present study aimed to assess whether epigenetic clocks could serve as a predictor of IVF success in terms of live birth, compared with common ovarian reserve markers and chronological age. According to our results, epigenetic age can predict IVF success, when corrected for traditional ovarian reserve variables. Although the absolute predictive performance of epigenetic age was modest, the adjusted OR of 0.91 per year highlights a biologically relevant association between accelerated epigenetic aging and reduced IVF success. This effect was independent of traditional markers such as AFC, reinforcing the hypothesis that epigenetic clocks may reflect oocyte competence or systemic reproductive aging not captured by existing metrics. As the association remained significant even after adjusting for both baseline ovarian reserve and IVF response, a robust link between the epigenetic profile and reproductive outcomes could be speculated. On the other hand, the clinical relevance of our findings is debatable. The improvement of the predictive capacity is too modest to advocate its use in clinical practice. The c-statistics analyses showed only a mild improvement in performance with the inclusion of epigenetics information. While this does not yet justify routine clinical application, it supports the rationale for further studies aimed at refining epigenetic clocks specifically designed for fertility and for exploring their integration into multiparametric predictive models. Our findings reinforce and extend those of our previous pilot study [ 16 ]. In that analysis, based on a smaller sample size ( n  = 181 women) and on a restricted chronological age range (37–39 years), we already pointed out that the epigenetic age, derived from the same epigenetic clock, was significantly different in women who succeeded or failed in achieving a full-term pregnancy. Specifically, epigenetic age was significantly lower in the participants who had a live birth compared to those who did not (36.1 ± 4.2 and 37.3 ± 3.3 years, respectively, p  = 0.04), with an adjusted OR for live birth per year of EPA of 0.91 (95%CI: 0.83–1.00, p  = 0.048). The primary strength of the present research lies in its prospective design, with a broader age range and larger population sample, yielding meaningful insights that strengthen the concept that this line of research may be fruitful. Indeed, it is possible that the performance of the tested epigenetic clock could improve in some indications. While the overall cohort size was sufficient to detect significant associations, subdividing the population by infertility cause resulted in smaller sample sizes per group, preventing us to detect meaningful differences. The apparent loss of predictive accuracy in subgroup analyses could be explained by the reduced statistical power rather than a lack of association in specific infertility categories. Future studies should aim to include larger cohorts for each infertility cause to better assess the role of epigenetic age in relation to specific pathogenic mechanisms. Not only, but infertility causes are often heterogeneous and overlapping, and clinical classifications may not fully reflect the underlying biological complexity. It is plausible that the biological processes captured by epigenetic age could cut across traditional diagnostic categories, thus reducing the discriminatory power within narrowly defined subgroups. This aspect could be particularly compelling in the “idiopathic” infertility, where unrecognized abnormalities that result in elusive genetic and/or epigenetic changes of underlying subtle defects in gamete function, embryo development, or implantation have already been suggested as a potential causative factor [ 19 , 20 ]. The most attractive group could be women older than 35 for whom the diagnostic work-up failed to identify any cause of infertility [ 3 ]. In our exploratory analysis, epigenetic aging appeared to have greater predictive power than other variables in women aged 31–35 years. Although the analysis was impaired by the small sample size, further analysis should be encouraged to disentangle the potential role of lifestyle and modifiable factors in reproductive aging. Indeed, while lifestyle interventions are commonly advised, no direct causal link has been established between epigenetic age and infertility. Epigenetic aging may mediate broader biological disruptions (i.e. oxidative stress, chronic low-grade inflammation) that are increasingly recognized as contributors to reproductive decline. However, caution should be adopted in data interpretation as the observed predictive performance in this age subgroup could also be influenced by different distribution of infertility causes. This cannot be fully disentangled at this point, given the exploratory nature of this analysis. Our findings add to the growing body of literature postulating that epigenetic aging may reflect biological processes involved in reproductive aging, although the causal role of DNA methylation changes remains to be established. While in the pilot study of Monseur et al., a correlation was reported between epigenetic clock, AMH and oocytes yield [ 21 ], in the current one, we confirmed and strengthen the preliminary impression that the association between epigenetic age and IVF success is independent of ovarian reserve biomarkers as it remained statistically significant adjusting for baseline characteristics [ 16 ]. Not at last, epigenetic mechanisms might hold relevance as a predictor only in infertile populations, where the processes driving reproductive aging are more likely to be shaped by epigenetic impairments. Intriguingly, recent research based on a registry of 1,657 couples who naturally conceived pregnancies, has in fact denied any association between DunedinPACE clock and fecundability in couples with naturally conceived pregnancies [ 22 ]. This data suggests that epigenetic impairments may be limited in populations with preserved fertility, while they could serve as a targeted tool for assisted reproductive settings. In the present study, we employed a simplified epigenetic clock. Even if we adopted an already validated one, we recognize that it is based on a simple method of sequencing (pyrosequencing) and examines only a limited number of genes (five). We could have chosen more comprehensive and precise approaches, such as those relied on next-generation sequencing which assess thousands of genes [ 14 ]. It could be reasonable to expect that the adoption of more sophisticated second-generation epigenetic clocks (e.g. GrimAge) could offer promising advantages [ 23 ]. To note, the Zbieć-Piekarska2 epigenetic clock was not specifically validated in women of reproductive age, but developed on a broad age range (2–75 years) and in a mixed-gender population. This methodological aspect may affect the precision of the model in our study population. However, this limitation is not unique to the Zbieć-Piekarska model, as none of the most commonly used epigenetic clocks to date have been optimized for fertility-specific settings. Therefore, the recent trend to develop surrogate clocks for specific clinical outcomes (e.g. life span, age-related diseases such as cancer and cardiovascular events) encourage the advance of fertility-specific epigenetic clock. Besides, a crucial consideration is in fact the use of epigenetic clocks based on peripheral blood, which lies on the “a priori” assumption that somatic tissue reflects the ovarian function. However, recent findings have already cast doubt on this conjecture [ 24 ]. Morin et al. observed that cumulus cells, a key component of ovarian follicles, displayed a significantly younger epigenetic age compared with blood cells and similar results were reported in a larger cohort of infertile women [ 24 ]. Indeed, in a subgroup analysis, Hanson et al. found that white blood cells were epigenetically older in poor ovarian responders [ 23 ] and similar data were described in Olsen et al. study, where they stratified population by ovarian reserve variables [ 25 ]. Although these observations might have limited direct applications in IVF field, since ovarian cells are available only after the IVF treatment, they underline the relevance of studying the methylation trajectory in ovaries compared with somatic cells. In other words, there could be a real danger that we are just applying the wrong model to the right question, and we should start again from analysing the whole genome and epigenome of the peripheral and the ovarian compartment to identify the most predictive CpG sites for reproductive outcomes. Some additional limitations of our study must be mentioned. First, the most suitable outcome would be the cumulative live birth after the whole IVF journey. Some women failing to conceive in their first attempt could succeed in the following ones. The inclusion of these cases among those who fail with IVF at first attempt may dilute the findings and underestimate the performance of the tested clock. Second, some women may have barriers to conception that cannot be overcome with IVF (such as non-receptive endometrium). Again, this may negatively affect the performance of the test. In conclusion, as a proof-of-concept, our study supports further investigation and the future development of more refined, fertility-specific epigenetic clocks that could eventually inform personalized reproductive strategies. While the modest predictive improvement combined with associated costs and technical complexities do not currently encourage its routine implementation in clinical IVF practice, our data proved that epigenetic age and acceleration can play a role in predicting IVF success, even if corrected for traditional ovarian reserve biomarkers. Epigenetic clocks could complement traditional ovarian biomarker, especially if integrated in a composite predictive model which includes multiple factors (AFC, AMH, epigenetic age), once adopted the correct epigenetic model for fertility scope. This preliminary observation could pave the way for future clinical applications, where epigenetic screening could become an integral part of fertility assessment contributing to enhance risk stratification. Given these promising results and acknowledging the limitations of our project (e.g. the epigenetic model chosen, the assumption that ovarian ageing mirrors peripheral blood ageing), it becomes clear that future research should focus on refining epigenetic clocks specifically designed for reproductive health. Large-scale genome-wide studies will be necessary for the identification of CpG sites which are most predictive of ovarian ageing. Ultimately, such efforts could lead to the development of fertility-specific epigenetic models, offering a more precise and individualized approach in the reproductive field.

Infertility has become a growing public health concern, affecting approximately 48 million couples and 186 million individuals worldwide [ 1 , 2 ]. Infertility is not only widespread but also progressively evolving, needing timely intervention to optimize therapeutic outcomes. Despite the significant improvements of assisted reproductive technologies (ART) [ 3 ], the success rate remains modest: only one-third of in vitro fertilization (IVF) cycles result in live births and success linearly declines with age [ 4 ]. Natural fertility declines with age, especially after 35 years [ 4 ], making the issue challenging in high-income countries, where childbearing is often delayed [ 5 ]. Even if age and ovarian reserve are recognized as key determinants of IVF outcomes [ 6 , 7 ], traditional markers of ovarian reserve, such as anti-müllerian hormone (AMH) and antral follicle count (AFC), though valuable, often fail to accurately predict IVF live birth rate (LBR). This is largely because IVF success depends more on oocyte quality than quantity, and current clinical variables lack precision in assessing this critical factor [ 7 – 9 ]. The lack of accurate measure of oocyte quality highlights the ongoing challenge to define the “good” IVF success predictor [ 10 ] and spurs interest in aging biomarkers within reproductive medicine [ 11 ], which might reveal to be crucial to navigate the complex landscape of infertility and to develop more effective interventions [ 12 ]. In recent years, the concept of fertility as a ‘sixth vital sign’ or as a ‘proxy for overall health’ has gained traction [ 11 , 12 ], suggesting that fertility may reflect a broader set of factors, including genetic predisposition, environmental exposures, and lifestyle influences. Within this context, epigenetic mechanisms, namely epigenetic clocks, are emerging as promising biomarkers of biological age, offering more accuracy than chronological age alone [ 13 ]. Epigenetic clocks, based on DNA methylation (DNAm) through machine learning models [ 14 ], have revolutionized aging research Since 2012, numerous models have been proposed for different biological and clinical purposes, and are now beginning to be explored also in the reproductive field [ 15 ]. However, none has yet been specifically developed and validated for this context [ 15 ]. In a recent pilot study of 181 infertile women aged 37–39 years, we observed that those who delivered after ART were epigenetically younger than those who did not (respectively: 36.1 ± 4.2 and 37.3 ± 3.3 years, p  < 0.04). As the difference remained statistically significant ( p  = 0.028), even adjusting for AFC and follicular stimulating hormone (FSH), these findings suggest that epigenetic clocks could glean additional information to the traditional ovarian reserve biomarkers [ 16 ]. Although lifestyle changes are commonly recommended in reproductive care, the link between epigenetic aging and infertility appears to be indirect, likely mediating systemic processes such as inflammation and oxidative stress [ 11 , 12 ]. Building on our previous results, the present study seeks to investigate the potential role of epigenetic clocks in predicting IVF success, in terms of live birth rate, in a broader and unselected population. By examining the relationship between epigenetic age and live birth rate, we aim to improve the precision of fertility assessments and to contribute to the development of tailored treatments that could enhance patient care and optimize reproductive outcomes.

Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1