Epigenetic clock measuring age acceleration via DNA methylation levels in blood is associated with decreased oocyte yield

Abstract

Purpose

To investigate how biologic age (phenotypic age at which your body functions) greater than chronologic age, (age acceleration (AgeAccel)), correlates with oocyte yield.

Methods

Thirty-nine women undergoing ovarian stimulation, inclusive of all infertility diagnoses, were included in this pilot study. Methylome analysis of peripheral blood was utilized to determine biologic age. AgeAccel was defined as biologic age > 2 years older than chronologic age. A negative binomial model was used to obtain the crude association of AgeAccel with number of oocytes. A parsimonious adjusted model for the number of oocytes was obtained using backwards selection (p < 0.05).

Results

Measures of age were negatively correlated with number of oocytes (chronological age Pearson ρ = − 0.45, biologic age Pearson ρ = − 0.46) and AMH was positively correlated with number of oocytes (Pearson ρ = 0.91). Patients with AgeAccel were noted to have lower AMH values (1.29 ng/mL vs. 2.29, respectively (p = 0.049)) and lower oocyte yield (5.50 oocytes vs. 14.50 oocytes, respectively (p = 0.0030)). A crude association of a 7-oocyte reduction in the age-accelerated group was found (− 6.9 oocytes (CI − 11.6, − 2.4)). In a model with AMH and antral follicle count, AgeAccel was associated with a statistically significant 3.3 reduction in the number of oocytes (− 3.1; 95% CI − 6.5, − 0.1; p = 0.036).

Conclusions

In this small pilot study, AgeAccel is associated with a lower AMH and lower oocyte yield providing preliminary evidence that biologic age, specifically AgeAccel, may serve as an epigenetic biomarker to improve the ability of predictive models to assess ovarian reserve.

Introduction

All organisms experience physiologic changes with time; however, discrepancies in the rate and risk associated with senescence are not captured by chronologic age alone and are affected by both genetic and non-genetic factors [1, 2]. A recent review detailed the cellular and molecular hallmarks of aging including genomic instability, telomere attrition, epigenetic alterations, loss of proteostasis, deregulated nutrient-sensing, mitochondrial dysfunction, cellular senescence, stem cell exhaustion, and altered intercellular communication [2]. Researchers have sought biomarkers of aging in order to stratify individual risk associated with this natural process—one such example is measurement of telomere length [1]. Recently, another biomarker, the “epigenetic clock,” has been heralded as even more predictive of chronologic age [3–5]. A variety of clocks have been developed using algorithmic analysis of different specimens (e.g., blood, saliva, and other tissues), numbers of samples (200~8 k), genetic platforms (27 k, 450 k, and DNA sequencing), and CpG sites (as little as 3 and as many as > 500). Correlations to chronologic age have varied from 0.96 to 0.98 with mean absolute deviation from chronologic age ranging from 1.9 to 3.6 and root mean squared error ranging from 3.88 to 4.26 [3, 4, 6, 7]. In the present study, the DNAge^™ clock is based on the 2013 Horvath clock but expands to include additional intergenic sites using targeted sequencing with over 500 loci. In theory, this could capture additional regulatory elements beyond the traditionally selected promoter sites [7].

These clocks are based on expected epigenetic changes in a regulatory process known as DNA methylation, typically a biochemical addition of a methyl group to a cytosine nucleotide followed by a guanine nucleotide [1]. As a result, gene expression may change without a physical change in the DNA sequence. Classically, the function of DNA methylation was thought to be a simple gene silencing mechanism; however, emerging studies suggest that depending on context and genomic location, the regulatory effects are more complex [8, 9]. Certain changes in DNA methylation with aging occur in a predictable manner allowing the methylome pattern to serve as an indicator of biological age [1]. Currently, it is not known if these age-specific changes are causal, resulting in changes in gene expression or simply a consequence of aging; however, deviations between chronologic age and biologic age exist. Biologic age greater than chronologic age is known as age acceleration (AgeAccel). AgeAccel is associated with frailty [10], Alzheimer disease, Down syndrome, HIV infection, Huntington disease, Parkinson disease, obesity, lifetime stress, bipolar disorder [11], lower physical/cognitive function, menopause, incidence of cancer, shorter cancer survival, [12, 13], and all-cause mortality [14–18]. In this study, we investigated the correlation between AgeAccel and oocyte yield at the time of oocyte retrieval in reproductive aged women.

Materials and methods

Patients

This pilot prospective cohort study was conducted at a single academic affiliated medical center. The study included women undergoing ovarian stimulation, embryo cryopreservation, and planning preimplantation genetic screening/diagnosis. All ages, infertility diagnoses, and ovum donation were included. The only exclusion criterion was current cancer diagnosis. Stimulation protocols were chosen per discretion of the patient’s primary physician. In both groups, gonadotropin doses were typically 300 IU for follitropin beta and 150 IU for human menopausal gonadotropin. IRB approval was obtained from Stanford University.

Sample preparation

Peripheral blood collection (10 cc) was performed using test tubes with DNA/RNA Shield^™ per manufacture recommendations (Zymo Research^®). Subsequently, Zymo Research^® performed DNA purification (Quick-DNA^™ Miniprep Plus), quality control (NanoDrop, qPCR, and TapeStation), bisulfite conversion (EZ DNA Methylation-Lightning^™), and Simplified Whole-panel Amplification Reaction Method (SWARM^®) using HiSeq 1500 for > × 1000 coverage [7]. Bismark software was used to analyze results of bisulfite treated sequencing. We classified the difference in ages as a measure of AgeAccel (biologic age of > 2 years than chronological age).

Statistics

We described the characteristics of the study cohort overall and by age acceleration status using standard descriptive statistics. T tests or chi-square were used unless the data were not normally distributed, in which case Wilcoxon was utilized. We also described the linear association between the study outcome (number of oocytes), anti-Müllerian hormone (AMH), antral follicle count (AFC), and chronological and biological age using Pearson’s correlation. To estimate the crude and adjusted associations for AMH, AFC, chronological age, and age acceleration with the number of oocytes, a negative binomial model with an identity link was used. For adjusted associations, we fit a fully adjusted model that included AMH, chronological age, and age acceleration and then used backwards selection (p < 0.05) from the fully adjusted model to obtain a simplified adjusted model for the number of oocytes. As this was a pilot study, we were able to reasonably obtain data for a total of 40 patients. Based on this maximum N, an estimated prevalence of 35% for AgeAccel (personal correspondence Zymo Research^®), a mean of 17.4 oocytes (standard deviation of 9.3 for oocyte yield based on historical Stanford data (n = 165)), a type I error of 5%, and 80% power, we estimated a minimum detectable difference of 6.7 in oocyte yield between women with and without age acceleration. All analyses were performed using R version 3.5.50. All testing was 2-sided. A p value of < 0.05 was considered statistically significant.

Results

In a small, heterogeneous population of 39 women, the mean chronological age was 38.0 (SD = 4.5) and mean biologic age was 39.2 (SD = 6.2; Table 1). One sample hemolyzed and was excluded from analysis, as it did not pass DNA quality control testing. Patients were 45% Asian, 42.5% Caucasian, and 12.5% other (Hispanic, Black, Middle Eastern). Identified 17.5% of former smokers are identified, 32.5% reported history of alcohol consumption, and 20% stated they never exercised. A wide array of infertility diagnoses were represented including endometriosis (n = 5), male factor (n = 7), polycystic ovarian syndrome (n = 5), tubal factor (n = 4), and recurrent pregnancy loss (n = 2), with the most prevalent being unexplained (n = 11/44, 25%) followed by diminished ovarian reserve (n = 8/44, 18%). Diagnoses were not mutually exclusive. In both groups, the majority of patients underwent antagonist protocols (87.1%, n = 34) with a smaller number of patients undergoing microdose flare (0.123%, n = 5). When comparing the two groups, the differences ovarian stimulation characteristics were not statistically significant.

Table 1.

Demographic characteristics and IVF outcomes stratified by age acceleration status

Characteristic	Age acceleration		p value
	No (N = 24) Antagonist: 22 Microdose: 2	Yes (N = 15) Antagonist: 12 Microdose: 3
Biologic age (years), (IQR), min. age: 27, max age: 54.3	35.0 (33.0, 41.4)	44.5 (41.8, 46.3)	< 0.0010
Chronologic age (years), (IQR), min. age: 25.8, max age: 45	38.3 (33.8, 40.9)	39.7 (36.9, 42.3)	0.29
Caucasian % (n)	41.7 (10)	42.9 (6)	1.00
% nulligravid	25.0 (6)	50.0 (7)	0.23
% nulliparous	50.0 (12)	71.4 (10)	0.34
BMI (kg/m2) (IQR)	25.7 (23.0,27.9)	25.6 (24.1, 30.4)	0.53
AMH (ng/mL) (IQR)	2.29 (1.51, 4.54)	1.50 (0.440, 2.96)	0.053
AFC (# 2–10 mm) (IQR)	14.5 (11.0,18.0)	8.00 (6.50, 9.50)	0.0050
Duration of stimulation (d) (IQR)	10.0 (9.0, 12.0)	10.0 (9.0, 11.5)	0.98
Total gonadotropin dose (IU) (IQR)	4050 (3600, 4890)	4050 (3830, 4500)	0.84
Peak E2 level (pg/mL) (IQR)	2170 (1700, 3004)	1750 (1190, 2430)	0.097
Number of oocytes (IQR)	14.5 (8.75, 18.8)	6.00 (3.50, 9.00)	0.0020
Number of metaphase II (IQR)	7.00 (6.00, 13.5)	3.00 (2.00, 5.75)	0.0020
Number of 2PN (IQR)*	7.00 (4.00, 12.5)	3.00 (2.00, 5.75)	0.0070
Number of blasts biopsied (IQR)	4.00 (2.00, 7.25)	1.00 (0.500, 3.00)	0.011
Number of euploid blasts (IQR)	1.00a (0.00, 3.00)	1.00b (1.00, 2.00)	0.87
Euploidy rate (IQR)	0.220 (0.00, 0.575)	0.500 (0.367, 0.944)	0.032

Data are presented as median with IQR. T test for means, chi-square for categorical and Wilcoxon for medians. IQR, interquartile range; PN pronuclear

^aTwo patients had no embryos and nine had zero euploid

^bFour patients had no embryos and two had zero euploid

Chronological and biologic ages were highly correlated (Pearson ρ = 0.81). Measures of age were negatively correlated with number of oocytes (chronological age Pearson = − 0.45, biologic age Pearson ρ = − 0.46) and AMH was positively correlated with number of oocytes (Pearson ρ = 0.91). Thirty-eight percent of patients were classified as having AgeAccel (n = 15, mean: 4.97 years, range: 2.4–9.3 years). Significantly, patients with AgeAccel were noted to have lower AMH values (1.29 vs. 2.29 ng/mL (p = 0.053)), lower oocyte yield 5.50 vs. 14.50 (p = 0.0020)), lower AFC (p = 0.050), fewer mature follicles (> 15 mm; p = 0.0050), lower numbers of metaphase II eggs (MIIs) and two-pronuclear zygote (2PNs) (p = 0.0020 and p = 0.0070, respectively), and lower number of blastocysts biopsied, independent of ploidy status (p = 0.011; Table 1). The number of euploid blasts ranged from 0 to 8 in both groups with a corresponding median of 1.00. The euploidy rate was noted to be higher in the age-accelerated group and reached statistical significance (p = 0.0324). A crude association of an approximate 7-oocyte reduction in the age-accelerated group was found (− 6.9 (− 11.6, − 2.4)). AFC was highly collinear with AgeAccel and there was little overlap in distribution; therefore, it was excluded from the adjusted model.

In a model shown in Table 2, chronologic age, AMH, AFC, and AgeAccel were associated with oocyte yield, with chronologic age being the least significant. A 5-year increase in chronologic age was associated with an ~ 4 oocyte reduction in yield. A 1-ng/mL increase in AMH was associated with a 3.2 increase in number of oocytes (95% CI 2.1, 4.4; p < 0.0010) while AgeAccel was associated with a statistically significant 3.3 reduction in the number of oocytes (95% CI − 6.3, − 0.6; p = 0.0030). Note that we did not include AFC in adjusted models due to high collinearity with age acceleration. For both AMH and chronological age, we found no evidence to indicate anything more than a linear association (i.e., quadratic or other non-linear functional form) provided an adequate model fit. Significantly, if a stricter definition or AgeAccel was used then the number in the AgeAccel group thins quickly (e.g., 12 and 6 cases of AgeAccel if using > 3 and > 5, respectively) and leads to uncertain and unstable estimated associations.

Table 2.

Crude and adjusted associations of chronologic age, AMH, and age acceleration with oocyte yield

Variables and prediction measures	Crude b (95% CI), p value	Fully adjusted b (95% CI), p value	Adjusted, standard selection b (95% CI), p value
AMH (ng/mL)	4.2 (3.1, 5.3), < 0.0010	3.2 (2.1, 4.4), < 0.0010	3.4 (2.3, 4.6), < 0.0010
Chronologic age (years)	− 0.8 (− 1.4, − 0.2), 0.011	−0.3 (−0.6, 0.1), 0.15	Excluded under backwards selection
AgeAccel	− 6.9 (− 11.5, − 2.6), < 0.0010	− 3.3 (− 6.3, − 0.6), 0.0030	− 3.3 (− 6.5, − 0.3) 0.0040

Discussion

The primary finding of this small pilot study is that a serum assessment of biologic age is correlated with ovarian response as measured by oocyte yield. Specifically, increased biologic age compared to chronologic age, AgeAccel, is associated with lower AMH levels and lower oocyte yield at the time of retrieval. These results from a pilot suggest that biologic age, specifically AgeAccel, may serve as an epigenetic biomarker to improve ability of predictive models to assess ovarian reserve. Given the significance of ovarian assessment not only for reproductive but also cardiovascular, skeletal, mental, and overall health, epigenetic clocks are a possible new approach to traditional ovarian reserve testing [19].

Chronologic age is heralded as the most significant predictor of decline in reproductive capacity; however, much variability exists [20]. Various surrogate markers are used (such as follicle stimulating hormone (FSH), AMH, and AFC) with known limitations leading to an incomplete clinical picture [21]. Overall, these markers are most commonly correlated with oocyte quantity and not necessarily quality; however, the two are interrelated [19]. As with any screening test, results of ovarian reserve testing must be considered in a patient-specific context and may provide more information in composite when compared to a single stand-alone test [19]. Without a larger sample size, it is hard to predict if AgeAccel is superior to traditional markers; however, these preliminary data provide evidence that clinicians may be able to glean additional or different information by utilizing biologic age.

Historically, basal FSH was among the first markers to have a demonstrated association with ovarian aging as elevations were noted with increasing age and diminishing reserve. By convention, assays are performed on day 3 with various cut offs described in literature and an acknowledgement of both interpersonal and intercycle variability. Perhaps the greatest limitation with regard to use in relation to procreative testing is the late onset of decline as 75% of women aged 40–44 still have normal levels (< 10 mIU/mL) [19]. Analysis of FSH and failure to conceive has shown variable specificity (50–100%) and lower sensitivity (3–65%) [22, 23]. As a result of these limitations, this test has fallen out of favor in exchange for AMH and AFC [19, 24].

Unlike FSH, AMH levels naturally decline as early as 15 years prior to menopause. Typically, an abnormally low level is defined as below 1 ng/mL regardless of when collected during menstrual cycle. Debate surrounding random, noncyclic variations of AMH still exists that is contrary to the historic notion that levels are stable throughout cycle [24]. In spite of this, AMH is often heralded as the most sensitive test with reliable prediction of ovarian response and timing of menopause. Additionally, AMH has an inverse correlation with chronologic age. Unlike FSH levels, AMH can be useful in predicting the risk of ovarian hyperstimulation. Despite these advantages, this marker also has poor prediction of clinical pregnancy with specificity ranging from 34.4 to 86.2% and sensitivity even lower (26–78.5) [24]. Meta-analyses have reported the area under the curve (AUC) for AMH for live birth as 0.61 [25]. Significantly, patients with AMH levels as low as < 0.16 ng/mL have overall live birth rate per cycle as high as 9.5% [26]. These are among some of the findings that have led clinicians and researchers to consider AMH to be a marker for oocyte quantity but not necessarily quality. Other limitations include lack of standardization and intra/interassay discrepancies as well as behavioral factors that may affect levels such as use of oral contraceptives, smoking, and diet [24].

The number of visualized follicles (roughly 2–10 mm) within the ovary, AFC, has been correlated with ovarian reserve with histologic confirmation [19, 27]. Patients with infertility with a lower AFC are more likely to have an under response to treatment. Similar to AMH, these qualities make AFC a good predictor of response to ovarian stimulation; however, there is also intercycle and interpersonal variability that can be further exacerbated among patients with higher BMIs [24]. Like the other two tests discussed, AFC is a good marker for ovarian response but only has specificity for nonpregnancy ranging from 64 to 98% and is even less sensitive (7–34%) [22].

As previously described in a review of ovarian reserve testing, multiple analyses have demonstrated AMH and AFC’s superiority to FSH, with AMH often preferred given a trend of not only prediction of oocyte yield but also the lack of requirement for sonographers and ultrasonography equipment [24]. Even with a strong correlation to ovarian reserve and reliable prediction of ovarian response, low AMH levels have not been shown to be correlated with decreased natural cognception outcomes. This has led to debates surrounding its use in the general population. Ideally, ovarian reserve testing could identify patients at elevated risk of early reproductive senescence; however, there is no consensus on whether the currently available markers are useful for screening the population-at-large [24]. Given that both the average age of first birth (26.8 years in 2017) as well as the birth rate among those aged 40–44 has been increasing, the ability to asses fertility status is becoming more and more clinically relevant [28]. In the present study, we investigated a new molecular candidate with testing that is noninvasive, affordable, and rapidly interpretable that may improve predictive ability of existing ovarian reserve testing.

Epigenetic clocks, as measured by changes in DNA methylation patterns, are gaining traction as biomarkers for aging [4]. A review of the literature demonstrates a multitude of conferred risk for AgeAccel including earlier age at menopause [29]. Given this association, we hypothesized an association with earlier reproductive senescence as measured by oocyte yield at the time of retrieval.

As in previous studies, biologic age was strongly correlated with chronologic age. As expected a priori, AMH levels decreased with increasing chronologic age. AgeAccel was associated not only with decreased AMH but also a statistically significant reduction in oocyte yield.

Our pilot study is noted to have significant weaknesses most notably a small sample size. We did not control for BMI, cycle number, or stimulation type or adjust for male factor characteristics such as parameters measured during semen analysis, which would have resulted a more complete model. Lastly, this patient population was primarily higher socioeconomic status. A larger sample size would likely demonstrate more overlap in AFC, and thus should be included in a future model. Without AFC in the model, the authors are unable to determine if biologic age provides additional or different clinical information. More participants would address the issue of collinearity, allow for a stricter definition of AgeAccel, and may increase sensitivity of identifying high-risk populations of early reproductive senescence. Despite prospective nature of this study, we are unable to elucidate whether associations with AgeAccel are causal or consequential.

There are strengths to the present study including a prospective design with a wide range of chronologic ages represented in an unselected, heterogeneous population. Biologic age demonstrated a strong correlation with chronologic age and, as expected, AMH was inversely correlated with oocyte yield. Our findings are similar to our hypotheses a priori as strong correlation of chronologic age and biologic age in reproductive aged women had previously been described in addition to an association of age acceleration with earlier age of menopause [30, 31]. AgeAccel defined as a discrepancy of greater than or equal to 2 years of age was chosen as a meaningful difference given previous reports of an intrinsic median absolute age error of 1.9 years [7]. It is notable that euploidy rate was actually higher in the age-accelerated group. The authors suspect that this could have also been affected by an uneven distribution in the number of oocytes, blastocysts biopsied, and euploid blasts with higher numbers of euploid in the non-accelerated group despite lack of statistical significance. For future analysis, it may be useful to categorize response according to number of oocytes retrieved. Notably, when responders were separated into categories of poor (1–3), suboptimal (4–9), normal (10 to 15), and high (> 15), the different frequencies of distribution were noted to be statistically significant with more normal and high responders in the non-accelerated group (data not shown). Future studies with a larger sample size should experiment using different AgeAccel thresholds as well categorization based on response to possibly identify higher risk populations.

Researchers and clinicians alike could benefit from a surrogate biomarker that better captures the physiologic decline in fertility seen with chronologic age. DNA methylation, as measured by epigenetic clocks, may serve as a useful serum assessment in patients undergoing fertility evaluation. The DNAge^™ test allows for simple intervention that could be incorporated into ovarian reserve assessment at time of consultation with reproductive endocrinologists. A serum myDNAge^® test is currently commercially available at a cost ($299) that is comparable other standard ovarian reserve testing (FSH: $95–124; AFC: $76–95; AFC: $300–500); however, clinicians should exercise caution in interpretation of age acceleration results or early adoption of this assessment until larger studies are performed [24, 32].

Like all clinicians, obstetricians and gynecologists routinely see patients of similar chronologic age with discordant susceptibility to chronic disease and other comorbidities. The epigenetic clock may serve as a unique way to identify age accelerated patients under the age of 35 at higher risk for pathologies typically associated with advanced maternal age such as spontaneous abortion, ectopic pregnancy, congenital malformations, hypertensive disorders of pregnancy, gestational diabetes, placental anomalies, multiple gestation, pre-term delivery, low birth weight, and maternal as well as perinatal/neonatal mortality [33].

Epigenetic changes can be reversible, making regulatory enzymes such as DNA methyltransferases potential pharmacologic targets. Prior research has focused more on histone deacetylase inhibitors (HDACIs) in the treatment of endometriosis but interestingly noted synergistic cross talk with DNA demethylation [34]. DNA methylation inhibitors such as azacytidine and decitabine are less studied given side effects (cytopenia and gastrointestinal toxicity) as well as an injectable route of administration [34]. While these agents or others could be used to induce changes in methylation status, it is possible there would be no effect on an individual previously designated as AgeAccel.

Methylome analyses have often been performed using microarray platforms. As the complex role of epigenetic regulation continues to be elucidated, DNA sequencing, as utilized in the present study, may become a more standard approach for analysis [3, 4, 6, 7]. Arrays, while less expensive and requiring less complicated preparation, suffer from a fundamental design bias, as results only capture those regions for which primers were designed. As our understanding of DNA methylation effects on gene regulation depending on context and location expands, sequencing will likely become increasingly important [35].

Patients of a similar chronologic age do not have the same risk profile for chronic disease or reproductive senescence. In this small pilot study, we performed a prospective, serum assessment using an epigenetic clock in women of reproductive age pursuing fertility treatment. Methylome analyses to obtain proprietary biologic age, DNAge^™, correlated with chronologic age in reproductive age women. AgeAccel, as measured in peripheral blood, is associated with decreased oocyte yield. This preliminary evidence suggests an association that lays the groundwork for additional studies to confirm whether AgeAccel has clinical role in risk assessment through incorporation in ovarian reserve testing. Specifically, future research with a larger sample size that controls for important covariates in predictive models including AgeAccel may better identify patients at higher risk for earlier reproductive senescence as well as morbidities traditionally associated with advanced maternal age.

Funding information

Zymo Research provided discount analyses of blood samples.

Compliance with ethical standards

Study was approved by the IRB of Stanford University.

Conflict of interest

The authors declare that they have no conflict of interest.