Analysis of DNA Methylation in Endometrial Biopsies to Predict Risk of Endometrial Cancer

Abstract

Objective:

To determine whether analysis of methylated DNA in benign endometrial biopsy (EB) specimens is associated with risk of endometrial cancer (EC).

Methods:

We identified 23 women with EBs performed at Mayo Clinic diagnosed as normal (n=14) or hyperplasia (n=9) and who later developed endometrial cancer after a median interval of 1 year. Cases were matched 1:1 with patients with benign EBs who did not develop EC (controls) by histology of benign EB (normal endometrium vs. endometrial hyperplasia without atypia), date of EB, age at EB, and length of post-biopsy follow-up. DNA extracted from formalin-fixed paraffin-embedded tissues underwent pyrosequencing to determine percent methylation of promoter region CpGs at 26 loci in 4 genes (ADCYAP1, HAND2, MME, RASSF1A) previously reported as methylated in EC.

Results:

After pathologic review, 23 matched pairs of cases and controls were identified (14 normal, 9 hyperplasia without atypia per group). Among cases, median time from benign EB to EC was 1 year (range 2 days – 9.2 years). We evaluated 26 CpG sites within 4 genes and found a consistent trend of increasing percentage of methylation from control to case to EC for all CpGs. At the gene-level, mean methylation events of ADCYAP1 and HAND2 in cases were significantly higher than control (p=0.015 and p=0.021, respectively). Though the other genes did not reach statistical significance, we observed an increased methylation trend among all genes. Area-under-curve (AUC) calculations (predicting future development of EC in the setting of benign EB) for ADCYAP1 and HAND2 were 0.71 (95%CI 0.55–0.88) and 0.83 (95%CI 0.64–1, respectively).

Conclusions:

This proof-of-principle study provides evidence that specific methylation patterns in benign EB correlate with future development of EC.

1. Introduction

Endometrial cancer (EC) is the most common gynecologic cancer, currently affecting over 600,000 women in the US alone, with 61,880 new cases and 12,160 deaths estimated for 2019 [1]. Although the majority of patients present with stage I disease and have a good prognosis (> 85% 5-year overall survival), approximately 15% of EC are diagnosed when the tumor has spread outside the uterus (stage III-IV) and have a poor prognosis [2]. Given the increased prevalence in the general population of risk factors for EC such as diabetes and obesity, it’s estimated that the incidence of EC will increase to 82,000 and 122,000 new cases per year in 2020 and 2030, respectively [3]. For this reason, advancements in our ability to identify women at risk of developing EC when disease has not yet developed into invasive carcinoma may enhance early detection and prevention and has the potential to reduce morbidity and societal costs associated with treatment. Although several approaches for EC screening have been proposed, routine screening in women at average risk is not recommended due to the lack of sufficient specificity [4]. A natural starting point for testing an effective risk reduction strategy via early detection is in the cohort of women who have had a benign endometrial biopsy (EB) and subsequently developed EC. With a population-based study, our group previously demonstrated that as many as 28% of EC patients have had a previous benign EB during their lifetime [5]. Consistent with this observation, it is reported that epigenetic factors preceding tumor onset and cancer diagnosis may be important in the development of various cancer phenotypes [6, 7]. Delineation of these mechanisms would provide novel opportunities to develop risk reduction strategies that may have general utility. To target the population of patients with a benign EB, it is important to identify factors that will predict the emergence of disease and would potentially prevent approximately one quarter of all ECs. In this large population of women who had a benign EB, clinical parameters such as obesity, personal history of colorectal cancer, and presence of an endometrial polyp in the EB may also potentially help in the identification of patients at higher risk of developing a future uterine malignancy [5]. However, having objective molecular markers of future development of malignancy in endometrial tissue may provide a better tool for early detection and amelioration strategies. Such markers may also aid the development of new insights in the process of carcinogenesis. While similar approaches have been previously used to predict cancer progression in patients with endometrial hyperplasia, these studies did not incorporate predictive molecular distinctions between benign and malignant endometrium and therefore lacked the discernment desired for prognosticative markers [8, 9].

DNA methylation of promoter regions of tumor suppressor genes is an important mechanism during carcinogenesis [10], and has been identified in multiple types of cancer and precursor lesions [11, 12]. Aberrant methylation has also been reported to be an early step during carcinogenesis [13], and has a key role in both development and cancer progression. In other cancers, methylated genes have been leveraged as biomarkers for early detection [14], prediction of treatment response [15], and cancer recurrence [16]. Among women with germline DNA mismatch repair mutations, aberrant methylation in the promoter regions of tumor suppressor genes has been detected in benign endometrial tissue before the development of endometrial cancer [13]. Additionally, the use of methylated DNA as a screening tool has the advantage of both high sensitivity and conserved methylation during disease progression [4].

Aberrant methylation of tumor suppressor and promoter genes is detectable in EC precursors and has been shown to distinguish between benign tissue and cancer [17, 18, 19] suggesting that some methylation markers may have value in cancer screening, early detection, and prevention. Furthermore, the possibility to detect methylation markers in the lower genital tract of patients with EC using intravaginal tampons may potentially enable self-collection, thus allowing early detection and prevention in under-served women [18]. Several genes including RASSF1A, ADCYAP1, MME, HAND2, RXFP3, CDKN2A, APC, and MGMT have been previously reported as frequently methylated in EC [17, 19–24]. Additional novel methylated CpG sites in EC are also continuing to be identified in recent and ongoing work [19, 25, 26].

The aim of this study was to test the prognostic efficacy of gene methylation in benign EB for future development of EC. We hypothesized that presumed carcinogenic changes in benign EB years before the histologic demonstration of EC may be predictive of the eventual development of EC. To test this hypothesis, we compared the methylation profile of four genes (RASSF1A, ADCYAP1, MME, HAND2) selected based on previous work. Methylation profiles of these genes were compared between benign EB from patients that later developed EC (cases) and benign EB from patients who did not develop EC (controls).

2. Methods

2.1. Cases and Controls Selection

Using institutional electronic databases, we identified as cases all patients who had a benign EB followed by a diagnosis of EC at our institution between January 1, 1992, and December 31, 2008. Patients with a benign EB who did not develop EC formed the pool of potential controls. Benign EB was defined as normal endometrium or endometrial hyperplasia without atypia. Patients with atypical hyperplasia were excluded because of the high likelihood of concomitant EC as well as the well-documented high risk of developing EC [27].

Cases and controls were matched 1:1 by histology of benign EB (normal endometrium vs. hyperplasia without atypia), date of benign EB (within ±3 years), age at benign EB (within ±8 years), and sample block age. Temporal ranges were not determined a priori but rather follow from case and control availability for this study. Follow-up for each control was at least as much as the time between the benign EB and the diagnosis of EC in the matched case. Applying these criteria, we identified 97 pairs of cases and controls. However, after pathologist review, 74 pairs were excluded due to discordant diagnosis from the one present in the medical records, insufficient tissue available for DNA extraction, or no availability of the tissue. A total of 23 pairs of case/control were included in the study (Supplementary Table 1). From the final 23 cases, 13 had the EC specimen available and were included in the analysis. Other patient variables for which data was analyzed when available for all groups included BMI, race, and diabetes status. All patient data used in this study was subject to an approved Mayo Clinic Institutional Review Board protocol (10–000274).

2.2. DNA Extraction and Bisulfite Modification

The epithelial cells of each formalin-fixed paraffin-embedded (FFPE) tissue section were identified by the pathologist (D.V.) and macrodissected with a 26 gauge needle from FFPE blocks. After deparaffinization, DNA was extracted using the Allprep DNA/RNA FFPE Kit (Qiagen, Hilden, Germany) according to the manufacturer’s protocol. Subsequently, ~1.0 μg of DNA was modified with an EZ-96 DNA Methylation Gold kit (Zymo Research, Irvine, CA) according to the manufacturer’s protocol. DNA was eluted in 30 μL elution buffer for pyrosequencing.

2.3. Gene Selection

The four genes (ADCYAP1, HAND2, MME, RASSF1A) we selected to examine in this study were previously reported to be methylated in EC. In particular, ADCYAP1 and MME were identified in a prior discovery effort as the genes with the highest performance in discriminating between endometrial cancer and benign endometrial tissue (AUC=0.86 for both genes) [19]. HAND2 has been reported as methylated in over 90% of EC [22], representing one of the most common molecular alterations in EC. Moreover, HAND2 methylation in premalignant lesions has been demonstrated to be an early event during carcinogenesis [22]. Similarly, RASSF1A is methylated in the majority of EC, as well as in normal endometrial tissue adjacent to EC, suggesting a multifocal pattern of epigenetic alterations associated with carcinogenesis [21].

2.4. Methylation Analysis

Pyrosequencing was performed on bisulfite-treated DNA previously extracted from cases, controls, and EC developed by cases. PCR and sequencing primers for ADCYAP1, MME, and RASSF1A were designed using the PyroMark Assay Design Software (Qiagen, Hilden, Germany; Supplementary Table 2). PCR and sequencing primers for HAND2 were included in a commercially available assay (PyroMark CpG Assay Hs_AC093849.1_03_PM, Qiagen, Hilden, Germany). Target regions of up to 250 bp of the respective sequence of bisulfite-treated DNA resulted in amplification of both methylated and unmethylated regions within the corresponding sequence. The PCR reaction utilized 20 ng of bisulfite-treated DNA template and TaqGold DNA polymerase (Applied Biosystems, Foster City, CA) with the following thermocycling protocol: 10 min at 95°C; 50 cycles of 35 sec at 95°C, 35 sec at 57.5°C, and 1 min at 72°C. Resulting PCR products were then verified for the expected size by gel electrophoresis. All pyrosequencing reactions were carried out on a Biotage PyroMark MD System (Qiagen, Hilden, Germany). Briefly, the incorporated biotinylated primer was immobilized on streptavidin-coated beads to both purify and display the denatured, single-stranded, and biotinylated PCR product. The single-stranded product was then annealed to 0.3uM of sequencing primer complementary to the single-stranded template. This was done at 85°C for 2 min followed by a cooling step to room temperature for 5 min. The pyrosequencing reaction was then performed by the successive addition of nucleotides in a predefined order. Raw data were analyzed using the Pyro Q-CpG 1.0.9 analysis software (Qiagen, Hilden, Germany).

2.5. Statistical Analysis

To assess for differences in patient/tumor characteristics between cases and controls, and the sample characteristics associated with valid CpG results, Student’s t-test and Fisher’s exact test were used for continuous and categorical variables, respectively. Student’s t-test was also used to test for the methylation differences between cases and controls, and between dichotomized time-to-event outcomes. Since the analysis of distribution of time between benign EB and EC identified two groups of subjects, one group diagnosed with EC after a short time interval from the benign EB (interval between 2 and 35 days) and the other group diagnosed with EC after a longer time interval from the benign EB (interval from 150 days to 9.2 years), we used 35 days as a cutoff for the dichotomized time-to-event analysis. A linear regression model with an interaction term was used to test for the effect of biopsy histology on gene methylation and its interaction with the future diagnosis of EC. Distance-based hierarchical clustering with average linkage was performed on the CpG methylation data to visualize the clustering pattern. Distances between samples were calculated based on the root mean squared (RMS) methylation differences, and those between CpGs were calculated as 1 - Pearson’s correlation, both on the CpGs with no missing values. To assess the utility of CpG methylation as a biomarker for future EC development, ROCs (Receiver Operating Characteristic Curve) were created by plotting the sensitivity again using 1 - specificity at different gene methylation cutoffs. Area under the ROC curve (AUC, or c-statistic) and its confidence interval were calculated based on Hanley and McNeil, 1982 [28].

3. Results

3.1. Patients and tumor characteristics

DNA was extracted from 59 unique tissue specimens (23 benign EB cases, 23 benign controls, 13 EC from cases). Patient and tumor characteristics are summarized in Table 1. In addition, the two groups were matched for age at biopsy, diagnosis of biopsy, and block age, BMI, race, and diabetes status did not differ between the two groups. In both groups, 14 (60.9%) patients had a benign biopsy, while 9 (39.1%) patients had hyperplasia without atypia. The median time between the benign biopsy and the diagnosis of EC was 360 days, with a range between 2 days and 9.2 years.

Table 1.

Summary results of patient and tumor characteristics. Case versus control comparisons are given where appropriate.

VARIABLES	CASES (N=23)	CONTROLS (N=23)	P-VALUE	EC WITH SPECIME NS AVAILAB LE FROM CASES (N=13)
AGE, YEARS (MEDIAN, IQR)	55 (43–64)	53 (46–62)	0.98	58 (48.5–64)
BMI, KG/M2 (MEDIAN, IQR)	31.8 (25.2–44.8)	27.7 (22.2–38.8)	0.5	31.1 (23.9–43)
RACE (%)			1
CAUCASIAN	21 (91.3)	22 (95.6)		12 (92.3)
UNKNOWN	2 (8.7)	1 (4.4)		1 (7.7)
DIABETES
(N, % YES)	4 (17.4)	3 (13)	1	2 (15.3)
DIAGNOSIS OF BIOPSY (N, %)			1
BENIGN	14 (60.9)	14 (60.9)		-
HYPERPLASIA WITHOUT ATYPIA	9 (39.1)	9 (39.1)		-
BLOCK AGE, YEARS (MEDIAN, IQR)	12.15 (8.3–15.7)	12.24 (9.7–15.7)	0.82	11 (9–13.5)
TIME BETWEEN BENIGN BIOPSY AND EC (DAYS (D), YEARS (Y))			-
MEDIAN (RANGE)	1 Y (2D–9.2Y)	-		-
≤35 DAYS (N, %)	7 (30.4)	-		-
>35 DAYS (N, %)	16 (69.6)	-		-
EC 2009 FIGO STAGE (N, %)			-
IA	21 (91.3)	-		12 (92.3)
IIIC1	1 (4.4)	-		-
IV	1 (4.4)	-		1 (7.7)
EC GRADE (N, %)			-
1	16 (69.6)	-		10 (76.9)
2	4 (17.4)	-		2 (15.3)
3	3 (13)	-		1 (7.7)
HISTOLOGY (N, %)			-
ENDOMETRIOID	22 (95.6)	-		13(100)
SEROUS	1 (4.4)	-		-
MI (N, %)			-
≤50%	21 (91.3)	-		12 (92.3)
>50%	2 (8.7)	-		1 (7.7)

Among the EC diagnoses, the 2009 FIGO stage distribution was: 21 (91.3%) stage IA, 1 (4.4%) stage IIIC1, and 1 (4.4%) stage IV. Histology was endometrioid in 22 (65.6%) patients, and serous in 1 (4.4%) patient. The majority of EC diagnoses also exhibited additional low risk factors, with 87.0% (n=20) classified as Grade 1–2 and 91.3% (n=21) with myometrial invasion ≤ 50%.

3.2. DNA yield and factors associated with successful methylation measurement

The median amount of DNA extracted was 142.3 ng/μL (IQR 42.5 ng/μL, 364 ng/μL). Among the 4 genes, 26 CpG sites were analyzed by pyrosequencing: 3 CpG sites in the MME promoter, 4 CpG sites in HAND2, 4 in ADCYAP1 and 15 in RASSF1A. Overall, 51% of the CpG sites had successful methylation measurement. Only 5 (6.5%) tissue samples did not have successful methylation measurement on any single one of the 26 CpG sites. Each sample had a median of 11 CpG sites (range 0–26) with valid results. Overall, 24 of the 59 (41%) samples had valid results on more than half of the CpG sites evaluated (more than 13 out of 26). We analyzed the possible reasons for the invalid results (Table 2) and we observed that age of the paraffin embedded tissue (more than 15 year old) was associated with a higher probability of pyrosequencing failure. Low DNA quantity predicted invalid results as well (p=0.01). The invalid results were also not randomly distributed: pyrosequencing tended to fail for the entire gene with the ADCYAP1 gene having the highest percentage of valid coverage, followed by HAND2, MME and RASSF1A (Supplementary Figure 4).

Table 2.

Percentage of valid results among variables is given. A significant difference was found in comparing block age, with ≥ 15 year blocks showing an increase in invalid results. (*t test is applied to the log scale of the concentration)

Variable	Number of the CpG Sites with valid results		P-value
	<50% (13 CpG Sites) (N=35)	≥50% (13 CpG Sites) (N=24)
DNA Concentration (ng/ul) mean, SD	291 (570)	574 (768)	0.01*
DNA Concentration (ng/ul)			0.06
<100	17 (48.6%)	5 (20.8%)
≥100	18 (51.4%)	19 (79.2%)
Type of the Sample			0.39
Control	15 (43%)	8 (33%)
Case	11 (31%)	12 (50%)
Cancer	9 (26%)	4 (17%)
Block age mean years (range)	12.3 (3.22–18.7)	10.2 (2.73–15.7)	0.07
Block age			0.02
<15 years	21 (60%)	22 (91.7%)
≥15 years	14 (40%)	2 (8.33%)

3.3. Comparison of Methylation Profile between Cases and Controls

In total, among the 4 selected genes we analyzed 26 CpGs (range 3–15 CpGs) using pyrosequencing. Hierarchical clustering based on the methylation profiles revealed interesting clustering patterns: CpGs from the same gene have similar methylation patterns across the samples and thus were tightly clustered by gene while cancer samples tended to cluster separately from the controls (Supplementary Figure 4). In CpG-level analysis of all 4 genes, a consistent trend of increasing methylation was observed from control EBs to case EBs to cancer tissue itself (Supplementary Figure 1). This was consistent with additional data showing all CpGs being positively correlated, especially those within the same gene (Supplementary Figure 2). Due to the high correlation of CpG methylation within the same gene, we averaged the CpG methylation across genes and summarized the results at the gene-level. Consistent with CpG-level analysis, at the gene-level analysis we observed a consistent trend of higher methylation in controls to cases to cancers for all 4 genes analyzed (Figure 1). The mean methylation levels of ADCYAP1 and HAND2 were significantly different between cases and controls (31.7% ± 18.6 vs. 19.9 ± 6.8, p=0.015, and 21.8% ± 15.1 vs. 8.5 ± 4.6, p=0.021, respectively) (Table 3).

Figure 1.

Boxplots of methylation percentages in controls, cases, and EC for the 4 genes analyzed. A consistent trend is shown of higher methylation from controls to cases to EC developed by cases.

Table 3.

Mean methylation levels of ADCYAP1, HAND2, MME, and RASSF1. Respective comparison statistics between controls, cases, and EC cases are also given.

		Controls (N=23)		Cases (N=23)				EC cases (N=13)
	CpG positions (N)	Mean methylation (%)	SD	Mean methylation (%)	SD	P-value	c-stat	Mean methylation (%)	SD	P-value
Gene						(t test)	(95% CI)			(Anova)
ADCYAP1	4	19.9	6.8	31.7	18.6	0.013	0.71	32.1	13.1	0.018
							(0.88 – 0.55)
HAND2	4	8.5	4.6	21.8	15.1	0.021	0.83	25	11.8	0.01
							(1 – 0.64)
MME	3	9.2	5.3	15.8	17.3	0.27	0.65	25.4	14.5	0.108
							(0.87 – 0.43)
RASSF1	15	6.65	2.43	19.4	21	0.11	0.66	38.5	30.3	0.04
							(0.91 – 0.4)

Analysis of the utility of the four genes as a biomarker for predicting future development of EC showed that both ADCYAP1 and HAND2 can be predictive of EC, with AUC=0.71 (95%CI 0.55–0.88) and AUC=0.83 (95%CI 0.64–1), respectively (Figure 2).

Figure 2.

Receiver operating characteristic (ROC) curve analyses of ADCYAP1 and HAND2. Areas under curve (AUC) for prediction of future development of EC are also shown for ADCYAP1 and HAND2.

3.4. Factors Associated with Methylation

3.4.1. Histology and future outcomes associated with benign endometrial biopsy

Though the histological type of the benign EB (i.e. normal endometrium vs. endometrial hyperplasia without atypia) was not a confounding factor, we evaluated its effect on methylation and the interaction with future diagnosis of EC. When we assessed the effect of the different histology of benign EB (normal endometrium vs. hyperplasia without atypia) on methylation, we observed that normal endometrium has significantly lower methylation than hyperplasia without atypia (p=0.043, 0.074, 0.037 and 0.067 for ADCYAP1, HAND2, MME and RASSF1 respectively). The normal endometrium also had less variability in methylation than hyperplasia without atypia (Figure 3). When the analysis was stratified by future diagnosis of EC (cases vs. controls), we observed that the difference was more prominent for cases compared to controls (Figure 3). Interaction analysis revealed that there were significant interaction effects between histological type and future diagnosis of EC for ADCYAP1 and HAND2 (p=0.012 and 0.018, respectively).

Figure 3.

Boxplots of methylation percentages in controls and cases stratified by diagnosis of benign EB (normal endometrium vs. hyperplasia without atypia).

3.4.2. Time between benign endometrial biopsy and cancer diagnosis

As the time between the benign EB of cases and the diagnosis of EC varied largely in our cohort (range from 2 days to over 9 years), we next sought to determine whether there was a correlation between methylation and the time from benign EB and EC. Although the analysis was unpowered to detect small differences, we were not able to detect any significant correlation (Spearman correlation, p-values >0.05). Moreover, when we further dichotomized the time-to-event variable using 35 days as a cutoff, we identified a trend of higher methylation associated with short time between benign EB and EC diagnosis. Due to the large variability in the short time cohort, we were unable to achieve statistical significance (two-sample t test > 0.05) (Supplementary Figure 3).

4. Discussion

In this study, we demonstrate that genes methylated in EC can be identified in benign EB up to potentially 9.2 years prior to the diagnosis of EC. Specifically, ADCYAP1 and HAND2 had significantly higher methylation in benign EB of women who developed EC compared to women who did not develop EC.

While the long term efficacy of EC prediction based on molecular markers remains a challenging task, we demonstrate here that long-term retrospective studies examining molecular predictors has utility in continuing work developing prognostic molecular assays. This study is the first to examine long-term predictive potential of methylation markers in benign EBs. The results of this study also complement previous research that has identified molecular changes such as microsatellite instability, DNA mismatch repair, and aberrant methylation of selected tumor suppressor genes in both cell lines [23] and in benign lesions several years before the development of EC among patients with Lynch syndrome [13]. However, although patients with Lynch Syndrome exhibited a progressively increasing average number of methylated loci from benign endometrium to simple hyperplasia to complex hyperplasia (with and without atypia) both in patients who subsequently developed EC and who did not develop EC, no difference in methylation levels were observed between the two groups. This inconsistency with our results could be explained by the different study design, the lack of matching between the two groups, and by the fact that both groups used included only MMR carriers.

In our cohort, 30.4% (n=7) of the cases were diagnosed with EC within the next 35 days from the benign EB, thus suggesting that EC was either already present at the time of the biopsy in other areas within the uterus or there was a misdiagnosis. Although not statistically significant, the observed trend of association between short time-to-EC (<35 days) and higher methylation could also be explained by a field-effect of methylation. A similar field-effect consisting in the presence of methylation in the normal endometrium adjacent to and in other sites within the uterus containing EC has already been reported in low risk EC [21] and other cancers such as breast and gastric [29, 30]. Therefore, it is possible that increased methylation could be used as a factor in assessing diagnostic confidence and for flagging false negative biopsies. Additional studies assessing the methylation profile of morphologically normally appearing endometrium within a uterus containing EC are thus warranted.

The present study also has limitations that should be addressed by future studies. First, the majority of cases included in our cohort developed low-risk endometrioid EC, with only 1 case (4.4%) of serous EC. For this reason, our results do not currently apply to type II EC. Type II EC is also known to be associated with TP53 mutations whereas Type I is not [31]. The extent to which aberrant methylation may therefore be used as a predictive factor associated with type II EC and whether methylation associations significantly diverge for Types I and II EC are at this point not well understood. Inquiries into these differences are thus merited and may also highlight mechanistic divergence pertaining to causality and development between EC types. Second, pyrosequencing did not provide successful methylation measurement for approximately 50% of the CpGs measured, mainly because of the age of preservation in paraffin and the small amount of tissue available in the EB. However, as reported, the analysis using only the non-missing values confirmed that all the CpGs evaluated clustered by gene. The continued development and improvement of DNA methylation microarrays with extensive genomic coverage [32. 33, 34] will allow the identification of more aberrant methylated genes and provide reliable and reproducible evaluation of the methylation pattern using FFPE tissue. Additionally, the continuing development of whole-methylome sequencing in biomarker studies for various cancers may also be applicable in developing EC prediction assays [40]. Third, our small sample size did not allow an accurate assessment of the utility of CpG methylation as a biomarker since the confidence intervals of ROCs were too wide. Neither did it allow investigating whether combining clinical variables such as BMI and diabetes status could significantly improve sensitivity and specificity of DNA methylation in predicting EC. The usefulness of clinical variables in improving the accuracy biomarkers in predicting EC has been described for BMI, sonographic endometrial thickness, and postmenopausal bleeding [41, 42]. Additionally, recent work has also suggested an evaluative role of the microbiome in the development of EC [35]. As suggested by European [36] and US [37] research agencies, future multidisciplinary research studies should combine clinical data, epigenomics, microbiome analysis, and other omics-based technologies, to develop and validate an individualized cancer prevention and screening program for gynecologic cancer.

In summary, this proof-of-principle study shows that molecular changes such as methylation could be potentially used to predict EC in women with benign EB. Each year approximately 5% of women in the US seek medical attention due to abnormal uterine bleeding that may require an endometrial biopsy to rule out the presence of endometrial malignancy [38], however, only approximately 10% of them are diagnosed with endometrial cancer [39]. For this reason, the identification of molecular markers such as methylation of selected genes can potentially provide an effective screening tool for predicting which of the remaining 90% of patients will develop endometrial cancer. This will lead to effective and individualized strategies of cancer-prevention. In addition, this approach could be potentially improved by focusing on other categories of women at high risk of EC, including but not limited to obese, diabetic, and postmenopausal women. Multifaceted prediction including these factors coupled with gene methylation status may indeed be necessary to capture the complex subtleties in EC tumorigenesis, with such approaches likely being necessary for enhanced and robust prediction of future EC.

Highlights.

• Examined DNA methylation in endometrial biopsies (EB) for efficacy of prognostication of future endometrial cancer (EC).

• Found that increased DNA methylation of ADCYAP1 and HAND2 in non-malignant EB correlate with future development of EC.

• Suggest that methylation changes warrant further testing as biomarkers to identify women at elevated risk for developing EC.