Detection of endometrial cancer via molecular analysis of DNA collected with vaginal tampons.

Early detection approaches for endometrial cancer (EC) are lacking, despite the fact that EC is the most common gynecologic malignancy in the United States and in many other developed countries ( 1 , 2 ). While low-risk, early stage EC has an excellent prognosis with 5-year overall survival (OS) >95%, 5-year OS when diagnosed at stage III or IV is sobering at 68% and 17%, respectively ( 3 ). Additionally, while the incidence of EC is disparately higher among Caucasian women compared with African Americans, mortality is higher among the latter ( 4 ). As similar care has been shown to eliminate racial disparities in outcomes among women with EC, this survival difference may, in part, reflect barriers to accessing health care and resultant delays in diagnosis ( 5 ). Early detection increases the chance of cure ( 6 , 7 ) and may also allow for minimization of surgical intervention and avoidance of costly adjuvant therapy and its related complications. An early detection method for EC may overcome barriers to care and potentially reduce morbidity and mortality related to this cancer and its treatment. The prolonged history of irregular bleeding that precedes diagnosis of most ECs ( 8 ) offers a window of opportunity to identify at-risk women prior to the development of advanced cancer. Up to 25% of women who develop EC are diagnosed with a precursor lesion, endometrial hyperplasia, between 1 to 20 years prior to their EC diagnosis ( 9 , 10 ). Additionally, the average progression time to EC from endometrial hyperplasia ranges from >15 years for the mildest degrees of hyperplasia to 4 years for the most severe lesions ( 11 ), suggesting there is a lengthy interval during which early detection and intervention is possible. Nonetheless, research on early detection methods is minimal ( 12 - 16 ). Approximately 20% of women with EC have Papanicolaou tests that contain cells suspicious for EC ( 17 ), which demonstrates that spontaneous shedding of tumor cells occurs frequently and is detectable with a relatively insensitive test. Furthermore, Kinde et al, demonstrated that 100% of 24 women with EC could be identified by testing their routinely collected liquid cervical cytology samples for a panel of 12 mutations ( 16 ). Prior to that, Fiegl et al demonstrated that DNA from the vaginal pool, collected via tampon, was hypermethylated in 15 women with EC compared to 109 women without EC ( 18 ). Taken together, these studies suggest that detection of EC via minimally invasive methods is a promising approach; however, optimization of markers, standardization of collection methods, and improvement in specificity are needed. Methylated DNA released from necrotic tumor cells is an attractive target and has been detected in a variety of biological fluids, including sputum, serum, peritoneal fluid, stool, nipple aspirates, urine, and vaginal fluid ( 18 - 20 ). DNA methylation assays offer potentially higher sensitivity than cytology as 1) intact cells are not required and 2) the relative stability of DNA offers advantages compared with protein or RNA based tests. A growing number of genes have been identified as methylated in EC and its precursors. RASSF1 is methylated in both primary and recurrent EC, as well as in morphologically normal appearing endometrium adjacent to ECs, irrespective of menopausal status ( 18 , 21 - 23 ); MLH1 methylation occurs in atypical hyperplasia as well as EC suggesting that it is an early event in EC carcinogenesis ( 22 , 24 - 28 ). CDH13 , PR-B, ERalpha-C, CIDEA, HAAO , RXFP3, CDKN2A, PTEN, p16, APC, HSPA2, SOCS2, and MGMT are also hypermethylated in EC ( 18 , 22 , 29 - 33 ) and additional candidate methylation markers are emerging ( 18 , 27 , 33 - 35 ). Pairing innovative, low cost, and minimally-invasive biospecimen sampling techniques with sensitive molecular testing that can be analyzed efficiently using high through-put methods has already led to the development of a novel screening and early detection tool for colorectal cancer ( 20 , 36 ). Minimally-invasive biospecimen collection techniques to sample the endometrium are currently available ( 37 - 39 ). However, a more convenient mode of biospecimen collection that employs a well-accepted hygiene product, the intravaginal tampon, would offer the advantage of patient acceptance and the ability to self-collect specimens; such a tool could be particularly useful for women with barriers to health care access. Towards that end, we assessed methylation of both established and recently identified candidate genes as biomarkers for EC within the vaginal pool, collected via intravaginal tampon, from women with and without EC, to assess the feasibility of such an approach as an early detection method for EC.

Women undergoing clinically-indicated hysterectomy were prospectively enrolled at the Mayo Clinic, Rochester, Minnesota from April 24, 2009 through January 25, 2011. Women eligible for this pilot study included those with a new diagnosis of complex atypical hyperplasia or any histology of biopsy-proven EC. The benign endometrium (BE) control group included women ≥45 years of age undergoing hysterectomy for benign indications. Women were excluded if they had recurrent cancer, known cervical cancer, received neoadjuvant chemotherapy or radiotherapy, prior tubal ligation, endometrial ablation, cervical stenosis, active endometriosis on treatment, or prolapse precluding tampon retention within the vagina. Patient, tumor, and biospecimen collection data were abstracted. This study was approved by the Mayo Clinic and National Cancer Institute Institutional Review Boards. Women were excluded from this analysis if both the tampon and Tao brush were not successfully collected. The biospecimens collected for this study included the vaginal pool (collected via intravaginal tampon), endometrial brushings (collected via Tao brush) ( 37 ), and primary tumor tissue. Patients were instructed to refrain from intercourse and douching for 24 hours prior to surgery. On the day of their hysterectomy, women self-placed an intravaginal tampon that was to be retained within the vagina for at least 30 minutes prior to surgery. After induction of anesthesia, the tampon was removed and placed in phosphate buffered saline (PBS). Subsequently, an endometrial brushing with a Tao brush was performed and the brush tip was removed and suspended in PreservCyt (Hologic; Bedford, MA). Duration of vaginal tampon retention and time to DNA extraction for tampon and Tao brush were recorded. Thirteen candidate biomarkers for methylation analysis were selected from two sources for this study. Genes selected from the gynecologic cancer literature were chosen based on their documented hypermethylation in gynecologic cancer ( RASSF1, HSP2A, HOXA9, CDH13, HAAO, and GTF2A1 ( 18 , 21 - 23 , 29 , 40 - 42 )). Additional genes were identified via methylation profiling utilizing the Illumina GoldenGate Standard Cancer Methylation Panel, which includes 1,505 CpG sites from 807 genes, in 143 ECs from the Polish Endometrial Cancer Study (PECS) study, 69 ECs from the Endometrial Hyperplasia Study (EH), and 63 BE samples from the Benign Reproductive Tissue Evaluation Study (BRTE) study. Details on the discovery effort to identify these abnormally methylated genes and the affected CpG sites have been previously reported ( 34 ). We ultimately selected CpG sites from the eight top candidates from the discovery effort ( ASCL2, HTR1B, NPY, HS3ST2, MME, ADCYAP1, SOX1 , CDH13 ) for further evaluation. All sites had probes that had p values <10 −7 in comparisons of methylation levels in EC versus BE, methylation differences between BE and EC ≥0.4, and mean methylation values in BE <0.2 ( 34 ). DNA was extracted from tampon supernatant using the Roche High Pure Viral Nucleic Acid kit per manufacturer’s protocols after spinning the tampon in a 50mL conical tube atop a removable sterile stainless steel sieve. DNA was eluted in 50 μL of elution buffer and concentration and purity were measured from absorbance at 230 nm, 260 nm, 280 nm on a Nanodrop-2000 spectrophotometer. Prior to DNA extraction from each Tao brush sample, the brush was scraped into the PreservCyt vial to remove remaining cells and material from the brush and discarded. After centrifugation, DNA was extracted from the pellet using the Puregene Tissue Core Kit A (Qiagen) according to the manufacturer’s protocols. DNA was eluted in 80 μL of elution buffer and concentration and purity were measured from absorbance at 230 nm, 260 nm, 280 nm on a Nanodrop-2000 spectrophotometer. DNA underwent bisulfite modification using the EZ-96 DNA Methylation Gold kit (Zymo Research, Irvine, CA) according to manufacturer’s protocols. One μg of DNA was used as input per sample and bisulfite-modified DNA was eluted in 30 μL elution buffer for pyrosequencing. Pyrosequencing was performed separately on tampon and Tao brush DNA. Primers were designed using Pyrosequencing Assay Design Software (Qiagen) ( Supplementary Table 1 ) and the CpG coordinates are listed in Supplementary Table 2 . Pyrosequencing assays were developed for RASSF1A, HSP2A, HOXA9, CDH13, HAAO, and GTF2A1 utilizing previously published primers ( 18 ) or designed based on CG rich sites within the promoter for genes without previously published primers. These assays covered between 4 and 16 CpG sites. Pyrosequencing assays were also developed for ASCL2, HTR1B, NPY, HS3ST2, MME, ADCYAP1, SOX1 , and additional CDH13 sites covering between 2 and 10 CpG sites overlapping with the probes included in the Illumina Golden Gate array ( 34 ). Pyrosequencing assays for SOX1 failed to perform adequately. A target region of up to 250 bp was amplified by PCR using paired primers complementary to the bisulfite-treated DNA sequence for each site and both methylated and unmethylated sequences of selected genes were amplified. Amplification was carried out on 20 ng of bisulfite treated DNA using TaqGold DNA polymerase (Applied Biosystems) under the following conditions: 10 min at 95°C, followed by 50 cycles of 35 sec at 95°C, 35 sec at 57.5°C, and 1 min at 72°C. The resultant PCR products were checked by gel-electrophoresis to confirm the size of the product and rule out the possibility of primer dimers. Pyrosequencing reactions were performed on a Biotage PyroMark MD System (Qiagen) according to manufacturer’s protocols. The incorporated biotinylated primer was immobilized on streptavidin-coated beads to purify and render the denatured, single stranded and biotinylated PCR product. The single-stranded product was then annealed to 0.3 μM of the sequencing primer complementary to the single-stranded template, and placed at 85 °C for 2 min, cooled to room temperature for 5 min. The pyrosequencing reaction was then performed by the sequential addition of single nucleotides in a predefined order. Raw data were analyzed using the provided Pyro Q-CpG 1.0.9 analysis software (Qiagen). Correlation of methylation levels for CpG sites within a gene, as well as association of methylation across genes, were assessed using Spearman correlation coefficients, presented graphically as pairwise comparisons across sites ( Figure 1 ). In addition, principal components were utilized to assess the total variation among CpG sites within a gene ( 43 ). The number of principal components needed to explain 90% of the total variation in the gene was reported along with the percent of the gene’s variation captured by those principal components. Patient and tumor characteristics were tabulated and compared for women with EC and BE by Wilcoxon rank sum and Fisher’s exact tests. The primary analysis was the association of gene-level methylation between EC and BE samples using the mean of the percent methylation for all CpG sites within a gene. Methylation differences between EC and BE were assessed using two-sample t-tests and receiver operating characteristics area under the curves (AUC).

Specimens from 38 patients with biopsy-proven EC and 37 BE controls were prospectively collected. Complete biospecimens (both tampon and Tao brush) were collected from all 38 EC and 28 BE subjects and were included in the subsequent analyses. Patient and tumor characteristics are listed in Table 1 . Among cancer cases, the 2009 FIGO stage distribution was as follows: 26 (68.4%) IA, 6 (15.8%) IB, 3 (7.9%) IIIC, and 2 (5.3%) stage IVB. There was also one case of complex atypical hyperplasia (CAH) which was included in the EC group. Histology distribution was as follows: 1 (2.6%) CAH, 31 (81.6%) endometrioid, 3 (7.9%) serous, and 3 (7.9%) clear cell. The EC group was older and had higher median BMI than women with BE; the rate of hormone replacement use was the same in the two groups ( Table 1 ). Uterine pathology among the BE group included: adenomyosis (n=1; 3.6%); benign endometrial polyp (n=1; 3.6%); leiomyomas alone (n=5; 17.9%) or associated with hyperplasia without atypia (n=1; 3.6%), tubo-ovarian abscess (n=1; 3.6%), benign ovarian mass (n=7; 25%), or benign ovarian mass and endometriosis (n=2; 7.1%). Final diagnoses for the remainder of the BE group included menorrhagia/dysmenorrhea without structural abnormality (n=2; 7.1%) and benign ovarian mass alone with no uterine pathology (n=5; 17.9%). One incidental endocervical adenocarcinoma in situ and 2 unexpected leiomyosarcomas were retained in the control group. The median amount of DNA collected from the vaginal pool via the tampon was 4.7 μg [IQR 2.5, 10.7 μg] and from the endometrial Tao brush 33 μg [IQR 8.1, 62.6 μg]; both collection techniques yielded sufficient DNA quantities for large-scale methylation analyses via pyrosequencing. Agarose gel electrophoresis and PCR of GAPDH confirmed high quality DNA from both tampon and Tao brush biospecimens. We evaluated methylation within the tampon biospecimens according to patient and tumor characteristics as listed in Table 1 . The strongest associations between gene-level methylation and patient or clinical characteristics were correlation of higher ASCL2 and RASSF1 methylation with greater tumor diameter (R 2 =0.24, p=0.005, and R 2 =0.16, p=0.016, respectively) and lower NPY methylation in patients with advanced stage disease (IIIC/IV) (mean 5.6 ± 2.6%) compared to patients with early stage disease (CAH, IA/IB) (mean 19.9 ± 18.2%, p=0.044). Age, BMI, cancer grade, cancer histology, myometrial invasion, and lymphovascular space invasion were not significantly associated with methylation differences at the gene level. Similar results were found for the Tao brush samples. Median (IQR) time between diagnostic endometrial sampling and tampon biospecimen collection within the EC cohort was 15 ( 12 , 20 ) days. The duration of time between diagnosis and vaginal pool sampling was not significantly related to methylation percentage. Additionally, the duration of tampon retention within the vagina did not influence methylation percentage (all R 2 <0.08) or DNA yield (all R 2 <0.08). Median (IQR) tampon retention time for the EC group was 111 (90.5, 163.5) minutes and 98 (90.3, 115) minutes for the BE group (p=0.91). Methylation results across all 12 genes among tampons collected from the 3 women with unexpected neoplasias (2 leiomyosarcomas and 1 endocervical adenocarcinoma in situ) were similar to those from women with a benign final pathology diagnosis. Among the 12 genes, 97 CpG sites were evaluated via pyrosequencing, ranging from 2-16 sites for each gene. Site-specific methylation differences between EC and BE are listed in Supplementary Table 3 . There was strong correlation in methylation levels across CpG sites within a gene, with the exception of GTF2A1 ( Figure 1 , Tables 2 and 3 ). Of note, none of the 10 CpG sites within GTF2A1 exhibited significant differences in methylation between EC and BE within tampon or Tao brush specimens ( Supplementary Table 3 ). In gene-level analyses, significantly higher methylation (p<0.05) was observed in tampon specimens from EC compared to BE for 9 of the 12 genes studied (ADCYAP1, ASCL2, CDH13, HS3ST2, HTR1B, MME, HAAO, HOXA9, and RASSF1 ) ( Table 2 ). The genes with the greatest differences in mean methylation between EC and BE were HTR1B (23.8 ± 14.9 vs. 10.4 ± 4.2, p<0.0001, AUC=0.82), HOXA9 (21.5 ± 21.8 vs. 6.0 ± 4.8, p=0.0001, AUC=0.74), and RASSF1 (21.3 ± 18.1 vs. 6.8 ± 5.3, p=0.0001, AUC=0.75) ( Figure 2 ). HTR1B , HOXA9 , and RASSF1 had similar patterns of methylation in EC and BE Tao brush samples, with slightly higher absolute methylation percentages in Tao brush samples compared to the tampon samples ( Figure 2 ). While 6 genes identified previously via profiling ( ADCYAP1, ASCL2, CDH13, HS3ST2, MME, and NPY) demonstrated comparably high levels of association (AUCs ranging from 0.66-0.70) between methylation and EC status, HTR1B had the highest AUC of all genes tested at 0.82 ( Figure 3a ). Among the genes reported in the literature, HAAO, HOXA9, and RASSF1 had AUCs comparable to those identified in our biomarker discovery work, ranging from 0.68-0.75, while HSP2A and GTF2A1 showed no association between methylation and disease status ( Figure 3b ). Analyses of Tao brush DNA demonstrated higher levels of methylation compared to tampon analyses ( Figure 2 , Supplementary Table 3 ); however differences in methylation levels between EC and BE were similar in the vaginal tampon and Tao brush samples ( Figure 2 ). Among genes identified by profiling, ADCYAP1 had the greatest increase in methylation percentage among EC (37.9 ± 20.0) compared to BE (11.6 ± 6.5) with an excellent AUC of 0.90; among genes previously reported, RASSF1 performed the best with mean methylation percentage of 26.0 ± 14.7 among EC vs. 10.5 ± 4.6 among BE (AUC=0.82) ( Table 3 ).

This proof of principle study demonstrates that EC-specific DNA methylation markers are detectable in lower genital tract samples collected via intravaginal tampon. Our results are supported by prior studies showing detection of EC markers using DNA methylation to test tampons ( 18 ) and mutation analysis of liquid-based cervical cytology samples ( 16 ). Specifically, our study shows that women with EC have higher gene-level methylation of both well-studied and recently identified markers in both vaginal tampon samples as well as endometrial brushings compared to women with BE. A collection method that is easily performed with a familiar sampler, such as a tampon, can potentially enable self-collection and delivery to a testing laboratory by mail. This approach could extend access to EC early detection methods into settings with limited resources secondary to socioeconomic status, geography, or cultural barriers. Among women who are traveling to distant referral centers for consultations, obtaining assay results in advance of visits could enable appropriate triage to gynecologic oncology or general gynecology clinics. In addition, the potential to use serial collection of tampons in research and clinical management may have value, especially among women at increased risk for EC secondary to high penetrance germline mutations, a history of endometrial hyperplasia, or treatment with tamoxifen. Given the rising incidence of EC in the United States and elsewhere, disparities in health care access, and the high mortality associated with metastatic disease ( 7 , 44 , 45 ), the development of an early detection approach has the potential to streamline management of at-risk women. In addition, recognizing that metrorrhagia and postmenopausal bleeding are common and disturbing symptoms, the ability to rule out EC can provide reassurances and limit the need for repeated biopsies. Previous studies have successfully exploited minimally-invasive collection techniques and highly sensitive DNA methylation analyses to detect cancer at other organ sites. DNA methylation changes that occur in colorectal cancer ( 20 , 36 ) and upper gastrointestinal (GI) cancer ( 46 , 47 ) can be detected in stool. In fact, Imperiale, et al recently demonstrated that multi-target stool DNA testing is a highly sensitive screening test for colorectal cancer that carries a detection rate similar to that of colonoscopy ( 36 ). This novel test has the potential to substantially impact colon cancer screening and supports momentum to develop novel and minimally-invasive screening tests for other cancers. In this study, we validated methylation markers previously identified as hypermethylated in EC compared with BE ( 18 , 21 - 23 , 29 , 34 , 40 ), and demonstrated detection of these markers in the downstream effluent of the female reproductive tract. One of these markers, HS3ST2 , was identified in our original discovery efforts ( 34 ) and has recently been validated as an EC biomarker in a cohort of women with endometrial biopsies showing atypical hyperplasia ( 48 ). In addition, we identified a novel gene methylated in EC, HOXA9 , which is also highly methylated in the vaginal pool of women with EC. HOXA9 has previously been reported as methylated in BE of women presenting with ovarian cancer ( 42 ) and was included in our panel of genes given its documented hypermethylation associated with gynecologic cancers. Interestingly, HOXA9 was shown to be methylated in the vaginal pool as frequently as methylation of RASSF1 , a gene methylated in nearly 90% of ECs ( 21 ). Not surprisingly, there was very low GTF2A1 methylation in EC and BE in both the vaginal pool as well as in endometrial brushings. GTF2A1 has previously been shown to be methylated in ovarian cancer ( 41 ) and was included in our analyses as a negative gynecologic cancer biomarker control. The histologic subtypes and stage distribution of ECs in this study reflect our clinical population, consisting predominantly of stage I endometrioid carcinomas. While precursor lesions, the spectrum of DNA mutations, and clinical behavior vary between type I and type II EC, we did not observe differences in DNA methylation patterns between the two types in this limited dataset. In larger sample sizes, RASSF1 methylation has also been shown to be similar in type I compared to type II EC; however, CDH13 has been shown to be more frequently methylated in low-grade type I compared to type II EC ( 40 ). Given the inclusion of only 6 type II cases in our study, further investigation into methylation differences between type I and type II EC is warranted. On final pathologic diagnosis, three non-endometrial uterine neoplasias were identified within subjects from the BE group: two women with leiomyosarcoma who underwent hysterectomy for presumed symptomatic leiomyomas and one woman with cervical adenocarcinoma in situ. Tampons and brushes from these three subjects demonstrated methylation patterns indistinguishable from other women with BE, suggesting that methylation markers may be specific for EC rather than a global indicator of gynecologic neoplasia. While this study was not designed to measure patient acceptance of the collection technique, all women enrolled in both groups self-placed the tampon and recovery of ample amounts of DNA from the vaginal pool was demonstrated. The potential of combining self-collection with sensitive molecular testing for disease detection is also supported by a meta-analysis of human papilloma virus (HPV) testing in which analyses of clinician collected samples showed only modestly better performance compared to those that were self-collected ( 49 ). As such, the utilization of the tampon as a biospecimen collection device carries promise as a means of self-sampling. Despite the encouraging results of this study, we recognize that further refinement is necessary to develop a clinically useful early detection assay for EC. Although there was significantly higher methylation among the EC group compared to the BE group, there remain a subset of ECs whose methylation overlaps with the distribution of BE methylation. Establishment of an informative panel of markers that yields a high negative predictive value is critical. Similar to GI cancer screening, a multi-target DNA panel is likely to be necessary to optimize test sensitivity. However, in contrast to the colon, obtaining endometrial biopsies is considerably easier, and therefore, we envision tampon testing as an intermediate step in management of at-risk women. Fortunately, methylation among CpG sites was consistent within each candidate gene that discriminated between EC and BE; single CpG sites per gene could therefore be utilized to simplify the DNA analyses. At present, additional discovery efforts using more global methylation analyses are underway to identify other novel candidate biomarkers that may better separate EC from BE. In addition, DNA mutations detected in the lower female genital tract have also been shown to discriminate between EC and BE ( 16 ),somatic mutations in genes, such as TP53, PIK3CA, and PTEN, are well described in EC ( 40 , 50 - 52 ), and further discovery utilizing next generation sequencing is underway in our sample set. As such, DNA methylation and mutation analyses could complement one another in the detection of EC. A major strength of this study is the inclusion of genes that are well-documented as methylated in EC ( 18 , 21 - 23 , 29 , 40 ) as well as more recently defined, highly methylated genes validated in independent sets of EC and BE tissues ( 34 ). DNA biomarkers are attractive for designing a self-collected test because it may enable at-home collection and mailing to a testing laboratory. In addition, DNA aberrations can be assessed with high throughput methods allowing rapid assessment of multiple samples simultaneously, the cost of methylation and mutation analyses are progressively decreasing, and the increasing number of certified laboratories that can perform molecular analysis with rigorous quality assurance would support the clinical use of self-collected tampons, once a standardized approach is validated. A notable limitation of this study is that all ECs had biopsy-proven disease via pre-study endometrial sampling. As such, while the median time between diagnostic endometrial sampling and vaginal pool collection was greater than two weeks and methylation percentages were not associated with the length of time between sampling and tampon collection, the impact of biopsy on tumor cell and cell-free DNA shedding is unknown. Spontaneous cancer cell shedding has been demonstrated to occur in EC ( 17 ) and all subjects with EC in our study presented with abnormal vaginal bleeding, which suggests ongoing spontaneous tumor DNA shedding. Nonetheless, interpretation of our study and previous reports are limited by collection of samples following prior endometrial instrumentation ( 16 , 18 ). Further investigation of the degree of DNA methylation occurring within the vaginal pool of prebiopsied symptomatic and high-risk asymptomatic women is warranted to further assess spontaneous DNA shedding from EC. Additionally, the impact of endometrial sampling on vaginal pool DNA methylation is warranted and prospective biospecimen collections in these patient groups are currently underway. Ultimately, determination of whether vaginal pool methylation changes can be identified in the general female population to identify those without risk factors for EC but at risk for EC as well as a cost-analysis of screening the general population are warranted. In summary, this proof of principle study demonstrates that self-collection of the vaginal pool via intravaginal tampon in combination with sensitive molecular analyses may enable detection of EC. Future directions include extension of DNA methylation marker analyses to additional novel genes, inclusion of EC mutations that could complement and improve EC detection sensitivity, and analyses of DNA collected via tampons prior to endometrial biopsy among high risk women. This approach may yield an early detection test for EC, which could enable effective treatment with reduced morbidity and improve mortality from the disease, particularly among under-served women.