DNA methylation testing for endometrial cancer detection in urine, cervicovaginal self-samples and cervical scrapes

Endometrial cancer incidence is rising and current diagnostics often require invasive biopsy procedures. DNA methylation marker analysis of minimally- and non-invasive sample types could provide an easy-to-apply and patient-friendly alternative to determine cancer risk. Here, we compared the performance of DNA methylation markers to detect endometrial cancer in urine, cervicovaginal self-samples and clinician-taken cervical scrapes. Paired samples were collected from 103 patients diagnosed with stage I to IV endometrial cancer. Urine and self-samples were collected at home. All samples were tested for nine DNA methylation markers using quantitative methylation-specific PCR. Methylation levels measured in endometrial cancer patients were compared to unpaired samples of 317 healthy controls. Diagnostic performances were evaluated by univariable and multivariable logistic regression analysis, followed by leave-one-out cross-validation. Each methylation marker showed significantly higher methylation levels in all sample types of endometrial cancer patients compared to healthy controls ( P < .01). Optimal three-marker combinations demonstrated excellent diagnostic performances with area under the receiver operating curve values of 0.95 (95% CI: 0.92-0.98), 0.94 (0.90-0.97) and 0.97 (0.96-0.99), for endometrial cancer detection in urine, self-samples and scrapes, respectively. Sensitivities ranged from 89% to 93% at specificities of 90% to 92%. Virtually equal performances were obtained after cross-validation and excellent diagnostic performances were maintained for stage I endometrial cancer detection. Our study shows the value of methylation analysis in patient-friendly sample types


Funding information
Hanarth Foundation; Stichting De Weijerhorst for endometrial cancer detection of all stages. This approach has great potential to screen patient populations at risk for endometrial cancer.

K E Y W O R D S
cervical scrape, DNA methylation, endometrial cancer, urine, vaginal sample What's new?
Endometrial cancer incidence is rising and current diagnostic approaches often require invasive biopsy procedures. Here, the authors compared the diagnostic value of endometrial cancer detection by DNA methylation testing between paired urine samples, cervicovaginal self-samples and clinician-taken cervical scrapes. Endometrial cancer detection in samples collected by home-based methods was excellent and comparable to diagnostic performance in clinician-taken cervical scrapes. The results demonstrated the value of methylation analysis in patient-friendly sample types for detection of endometrial cancer of all stages. The approach has great potential to noninvasively screen patient populations at risk for endometrial cancer.

| INTRODUCTION
Endometrial cancer is the most frequently diagnosed cancer of the female genital tract and the sixth most common cancer in women globally. 1 With its rising incidence worldwide, endometrial cancer accounted for 417 000 new diagnoses and over 97 000 deaths in 2020. 2,3 Early detection is crucial since advanced stage disease has a poor prognosis and a high risk of relapse. 4 In about 90% of cases, postmenopausal bleeding precedes endometrial cancer. 5 Consequently, diagnostic procedures that require a referral for specialized care are indicated for all patients presenting with postmenopausal bleeding, causing discomfort and high healthcare costs. [6][7][8] Yet only 5% to 10% of patients with this common alarming symptom have an underlying malignancy. 9 Another small subset of asymptomatic endometrial cancer patients is detected via a Pap smear obtained during cytology-based cervical cancer screening or other indications. [10][11][12] As cervical cancer screening programs have shifted from cytology to primary testing for human papillomavirus (HPV) in many countries today, the detection of asymptomatic endometrial cancers via Pap testing is declining. 13 Besides the detection of endometrial cancer in Pap smears, recent cytology research demonstrates that endometrial cancer cells are detectable in vaginal samples 14 and urine 15 by shedding through the cervix into the vaginal debris. An added benefit of using minimally invasive types of diagnostic samples, including urine and cervicovaginal self-samples, is that they can be collected at home, which is rather inexpensive and reduces the burden of health care.
As an alternative to cytology, objective biomarker testing on minimally invasive sample types has demonstrated great potential and would be ideal to triage patients with postmenopausal bleeding. DNA methylation signatures in promoter regions of tumor suppressor genes represent a valuable biomarker for the detection of early-stage disease. In the early stages of cancer development, promoter hypermethylation can lead to gene silencing and loss of their tumor suppressive function. 16 Methylation testing does not necessarily require the presence of intact tumor cells for interpretation and is also measurable using tumor-shedded circulating DNA.
Based on our previous studies and literature, nine markers (ADCYAP1, BHLHE22, CDH13, CDO1, GALR1, GHSR, HAND2, SST and ZIC1) are considered to be suitable for detection of endometrial cancer in minimally invasive sample types. [17][18][19][20][21][22][23][24][25] Our study was initiated to evaluate the diagnostic performance of endometrial cancer detection using DNA methylation analysis of these markers in paired urine, cervicovaginal self-samples and clinician-taken cervical scrapes. The cervicovaginal self-sample was also collected at home after urine collection. The clinician-taken cervical scrape was collected in the operating room, prior to surgery. In case the clinician-taken cervical scrape was not collected, the residual cytology sample of the cervical scrape that was taken for clinical diagnostics was used.
2.1.2 | Unpaired minimally-and noninvasive samples from healthy women For comparison, unpaired urine samples, cervicovaginal self-samples and clinician-taken cervical scrapes from healthy female controls were included. Urine controls were obtained through the Urine Controls (URIC) biobank. Controls were selected for eligibility based on a questionnaire in which age, sex and cancer history was documented. Only controls without any cancer history in the past 5 years were included.
A subset of urine samples was previously used and published. 25 Cervicovaginal self-samples and clinician-taken cervical scrapes were derived from leftover material of the Dutch national cervical cancer screening program coordinated by the Dutch National Institute for Public Health and the Environment (RIVM). Selection was based on age and a negative high risk HPV test. Selected controls were similar to cases with respect to age.

| Tissue samples
Methylation markers GHSR, SST and ZIC1 were previously discovered for cervical cancer detection, 26 but also appeared diagnostically relevant for endometrial cancer detection in urine. 25 To verify that increased methylation levels originate from the endometrial tumor, formalin-fixed paraffin-embedded (FFPE) tissue specimens of a subset of endometrial cancer patients from the SOLUTION1 study were also tested. FFPE tissues of normal endometrium were collected from patients with early-stage ovarian cancer without metastases to the endometrium who underwent a surgical staging procedure. 27 The age of selected controls was within the same age range as the cancer patients.

| Sample processing
Urine of endometrial cancer patients and healthy female controls were collected from home in collection tubes containing 0.6 M ethylenediaminetetraacetic acid (EDTA; final concentration: 40 mM) to maintain DNA quality during transport, following a previously validated storage and collection protocol. 28 In a previous feasibility study analyzing different urine fractions for optimal endometrial cancer detection using methylation markers, full void urine was shown to perform best and therefore used in the current study. 25 The cervicovaginal self-samples were collected using a dry-brush device (Evalyn Brush, Rovers Medical Devices, Oss, The Netherlands).
After collection of the urine and cervicovaginal self-sample, these samples were sent together within 72 hours by regular mail to the Pathology department of Amsterdam UMC, location VU University Medical Center, and processed directly upon arrival. Urine was stored at À20 C, and the dry brush was placed in 1. 5

| DNA methylation analysis using quantitative methylation specific PCR
Promoter hypermethylation of the ADCYAP1, BHLHE22, CDH13, CDO1, GALR1, GHSR, HAND2, SST and ZIC1 genes was tested in three multiplex assays by quantitative methylation-specific PCR (qMSP) using 50 ng of bisulfite-converted DNA. Primer and probe sequences were described before, 26,29 or are available upon request. Each assay also targets the reference gene ACTB for quantification and quality control. To ensure sufficient sample quality, samples with a Cycle threshold (C t ) value for ACTB ≥ 32 were excluded from further analysis. Methylation levels were determined using the comparative Ct method using the following formula: 2 -(Ct marker À Ct ACTB) Â 100. The discriminatory power of the qMSP assays was verified by testing tissue specimens of a subset of endometrial cancer patients included in the SOLUTION1 study and normal endometrial tissue specimens as controls.

| Data analysis
Only complete sample sets with valid DNA methylation test results (ACTB < 32) from endometrial cancer patients were included (eg, of cases with an invalid urine sample, also the self-sample and scrape were removed from the analysis). Methylation levels were expressed as 2log-transformed C t ratios. Differences in DNA methylation levels between endometrial cancer patients and controls were visualized using boxplots and tested for statistical significance using the Mann-Whitney U test. To assess the correlation of DNA methylation levels between paired sample types, the Spearman's rank correlation was used. The Spearman's rank correlation coefficient (r) was interpreted as poor (r ≤ .19), fair (r = .20-.39), moderate (r = .40-.59), strong (r = .60-.79) and very strong (r ≥ .80). 30 The diagnostic performance of individual methylation markers was evaluated by univariable logistic regression analysis in which the predicted probability was calculated for each sample. The predicted probability (a value ranging from 0 to 1) represents the probability for the presence of endometrial cancer.
Optimal three-marker combinations were formed for each sample type using multivariable logistic regression analysis with backward selection. The performance of the individual markers and optimal three-marker panels was visualized using receiver operating characteristic (ROC) curves, including the area under the ROC curve (AUC) with corresponding 95% confidence intervals (CIs). Sensitivities and specificities were based on the Youden's Index (J)-threshold. Diagnostic performances of each marker and three-marker panels for the detection of early-stage endometrial cancer was evaluated in a subanalysis in which only stage IA and IB cancers were taken along in the univariable and multivariable regression analyses which were performed as described above. The diagnostic performances of each marker and the three-marker panels were assessed outside the set by leave-one-out cross-validation (LOOCV). Predicted probabilities of the individual markers and optimal three-marker panels were also plotted individually in a heatmap format to illustrate differences between the sample types, histological subtypes and the potential added value of the marker combination.

| Study population and characteristics
A total of 158 patients with histologically confirmed endometrial cancer were included in the SOLUTION1 study. For various reasons, mostly because not all three sample types were available (n = 40), cases were excluded, resulting in a final study population of 103 endometrial cancer patients ( Figure S1). Within this group, a paired urine, cervicovaginal self-sample and clinician-taken cervical scrape of each case were available for methylation analysis. Unpaired samples of control women (n = 317) were used for comparison. Clinical characteristics of endometrial cancer patients and controls with valid qMSP results are depicted in Table 1. Additionally, FFPE tissue was collected from endometrial cancer cases of various histological subtypes (n = 33) and healthy endometrium (n = 15).  between the different sample types are presented in Table 2. Correlation coefficients between all markers are illustrated in Figure S3, showing that the majority of markers correlated highly between urine and self-samples while less correlation was seen when comparing urine and self-samples with cervical scrapes.
3.3 | Performance of DNA methylation analysis for endometrial cancer detection Marker panels allowed endometrial cancer detection with increased sensitivity, without a major impact on specificity. This was especially the case in cervicovaginal self-samples and clinician-taken cervical scrapes. The sensitivity and specificity of single genes in urine ranged from 34% to 87% and 84% to 98%, respectively, and the marker combination yielded a sensitivity and specificity of both 90%. For cervicovaginal self-samples, the sensitivity and specificity of single genes ranged from 28% to 78% and 85% to 95%, respectively, while the marker combination revealed a sensitivity of 89% and specificity of 92%. Similarly, for the clinician-taken cervical scrapes, the sensitivity and specificity of the single genes ranged from 44% to 87% and 67% to 93%, respectively, while the marker combination sensitivity was 93% with a specificity of 90%. Sensitivities and specificities were calculated based on the maximal Youden's Index (J) threshold (Table S1).
The diagnostic performance of individual markers and marker panels were validated by LOOCV, which yielded virtually equal AUC F I G U R E 1 DNA methylation levels of ADCYAP1, BHLHE22, CDH13, CDO1, GALR1, GHSR, HAND2, SST and ZIC1 in urine, cervicovaginal self-samples and clinician-taken cervical scrapes from healthy female controls (n = 100; n = 107; n = 110, respectively) and endometrial cancer patients (n = 103). DNA methylation levels are shown by the 2logtransformed Ct ratios. Boxplots illustrate medians with lower and upper quartile and range whiskers. Outliers are indicated with black circles. A P-value of .05 was considered statistically significant. **P ≤ .01; ***P ≤ .001; ****P ≤ .0001. C t , cycle threshold; EC, endometrial cancer.
values (Table 3), sensitivities and specificities (Table S2) for the single markers and optimal three marker combinations. Additionally, the performance for early-stage endometrial cancer detection was assessed by performing a subanalysis including only stage I endometrial cancers (n = 72). This revealed nearly equal diagnostic performances for both the individual markers and marker panels in urine, cervicovaginal selfsamples and clinician-taken cervical scrapes, which were also validated by LOOCV (Table S3).

− specificity
The application of previously discovered methylation markers in endometrial carcinomas of nonendometrioid histologies has remained largely unexplored, as previous studies included mostly endometrioid carcinomas. Even though nonendometrioid carcinomas are rare, early detection of this aggressive subtype is critical as they have a higher risk to metastasize and a substantially worse prognosis. 38 Our study revealed differences in DNA methylation changes between endometrioid and nonendometrioid cancers. While endometrioid carcinomas showed increased methylation of all methylation markers, the nonendometrioid carcinomas showed particularly increased CDO1 and GHSR methylation, followed by BHLHE22. Interestingly, CDO1 and GHSR are also known as pan-cancer markers, as they are described to be highly methylated in many human cancers. 39,40 Although their performance is excellent, they are probably not specific for endometrial cancer only and combining them with endometrial cancer-specific markers might be valuable for endometrial cancer-specific test development. Methylation of BHLHE22 and CDO1 in nonendometrioid cancers has previously been reported by Huang et al 17  DNA methylation analysis for endometrial cancer detection offers a sensitive molecular test, applicable to both minimally-and noninvasive sample types. This easy-to-apply approach offers the potential to reduce the number of biopsy procedures, thereby reducing costs and easing pressure on the healthcare system. The cervicovaginal selfsampling device provides a home-based sampling method which is introduced in the Dutch cervical cancer screening program to increase screening participation. 41 Logistics around transport and sample processing of this sample type is already in place in diagnostic laboratories in The Netherlands, which eases its implementation for endometrial cancer diagnostics. Urine is another attractive diagnostic sample type for the detection of endometrial cancer. 25 This liquid biopsy has gained more interest because it is easy to obtain and preferred by women over other sample types. 42 Apart from locally shedded cellular tumor-DNA, urine also contains transrenally excreted tumor-derived cell-free DNA which poses an additional advantage. 14,43,44 DNA methylation testing on patient-friendly samples types may contribute to the timely detection of endometrial cancer in patients with symptoms of postmenopausal bleeding. Moreover, this method is promising to screen asymptomatic women at risk for endometrial cancer (ie, women with inherited cancer syndromes, such as Lynch or Cowden) which are currently intensively screened using repeated endometrial biopsies. This approach may reduce the number of invasive procedures within these patient groups and prioritize the use of resources for patients in greater need in times of scarcity. Methylation testing in patient-friendly samples could also be valuable for recurrence detection after curative intent treatment, as recently explored in plasma by Beinse et al. 45 To assess the clinical applicability of this approach for abovementioned purposes, DNA methylation testing needs to be further validated on samples of patients presenting with postmenopausal bleeding with varying final diagnoses (ie, including women without abnormalities, benign endometrial conditions and cancer) or patient populations with an increased risk of developing endometrial cancer.
During finalization of the current article, the clinical utility of simple methylation-based tests in self-collected samples for endometrial cancer detection was also evaluated by Herzog et al. 35 Methylation levels of two regions of the GYPC gene and the ZCAIN12 gene allowed endometrial cancer detection in cervical scrapes, vaginal swabs and self-collected cervicovaginal samples with high accuracy in a cohort of women presenting with postmenopausal blood loss. Of note, the specificity in clinician-collected cervical scrapes was substantially lower (76%) as compared to present findings (92%). Nevertheless, their results are complementary to ours and independently exemplify the potential of epigenomic testing in self-collected samples.
Our study is limited by the fact that the distribution of histopathological subtypes included in our study does not reflect the natural prevalence. The inclusion of patients diagnosed with endometrial cancer mainly occurred in tertiary care cancer centers treating high grade cancers and rare histopathological subtypes. It is, however, worth noting that early detection of high grade nonendometrioid cancers is of utmost importance given their worse prognosis. 38 Samples were not collected at first clinical presentation but after endometrial cancer diagnosis was made based on a pipelle or hysteroscopic biopsy. This order would be different when home-based sampling would be applied in clinical practice. Sample collection after endometrial biopsy might have facilitated the release of tumor DNA into the urine or vaginal fluid. Yet, the influence of biopsy procedures on the presence of tumor DNA in self-collected specimen is most likely limited as the median time between biopsy and self-sampling was 37 days. In some excluded cases the complete carcinoma was biopsied with no or minimal residual cancer being found during histopathological evaluation of the uterus. It is conceivable that DNA methylation testing will be even more accurate when used at first clinical presentation. Finally, this comparative study had no access to paired samples from controls and did not include controls with postmenopausal bleeding symptoms or benign endometrial conditions. Even though others have shown that most of the markers tested in our study enable discrimination between benign endometrial pathology and cancer 17,18,20,21,23,24,32 , this was not validated in the current study.
Strengths of the current study are that nine DNA methylation markers, originated from different discovery screens, were tested on a large series of 626 samples. Over a 100 patients diagnosed with endometrial cancer were included, encompassing the full and heterogeneous range of endometrial cancer histotypes, grades and FIGO (2009) stages. Methylation marker assays were multiplexed to measure the methylation levels of three genes and a reference gene within the same reaction, without loss of PCR efficiency, to reduce hands-on time, costs and the amount of DNA needed. Self-samples and urines were collected at home, which is an appropriate setting to evaluate the use of home-based sampling for endometrial cancer detection. The collection of paired sample types allowed a comprehensive comparison of their performance.
Our study demonstrates that DNA methylation testing allows endometrial cancer detection with high sensitivity and specificity using a three-marker panel of methylated genes in patient-friendly sample types that can be collected at home. Following validation in additional cohorts, including individuals presenting with postmenopausal bleeding and asymptomatic women at-risk for endometrial cancer, methylation testing could be valuable as a preselection method to inexpensively determine who needs to undergo invasive endometrial tissue sampling and facilitate timely diagnosis.