# ENDOMETRIAL CANCER: FROM BIOLOGICAL TO CLINICAL APPROACHES

EDITED BY : Massimo Nabissi, Frederic Amant and Paola Gehrig PUBLISHED IN : Frontiers in Oncology

#### Frontiers Copyright Statement

© Copyright 2007-2019 Frontiers Media SA. All rights reserved. All content included on this site, such as text, graphics, logos, button icons, images, video/audio clips, downloads, data compilations and software, is the property of or is licensed to Frontiers Media SA ("Frontiers") or its licensees and/or subcontractors. The copyright in the text of individual articles is the property of their respective authors, subject to a license granted to Frontiers.

The compilation of articles constituting this e-book, wherever published, as well as the compilation of all other content on this site, is the exclusive property of Frontiers. For the conditions for downloading and copying of e-books from Frontiers' website, please see the Terms for Website Use. If purchasing Frontiers e-books from other websites or sources, the conditions of the website concerned apply.

Images and graphics not forming part of user-contributed materials may not be downloaded or copied without permission.

Individual articles may be downloaded and reproduced in accordance with the principles of the CC-BY licence subject to any copyright or other notices. They may not be re-sold as an e-book.

As author or other contributor you grant a CC-BY licence to others to reproduce your articles, including any graphics and third-party materials supplied by you, in accordance with the Conditions for Website Use and subject to any copyright notices which you include in connection with your articles and materials.

All copyright, and all rights therein, are protected by national and international copyright laws.

The above represents a summary only. For the full conditions see the Conditions for Authors and the Conditions for Website Use. ISSN 1664-8714 ISBN 978-2-88963-049-3 DOI 10.3389/978-2-88963-049-3

#### About Frontiers

Frontiers is more than just an open-access publisher of scholarly articles: it is a pioneering approach to the world of academia, radically improving the way scholarly research is managed. The grand vision of Frontiers is a world where all people have an equal opportunity to seek, share and generate knowledge. Frontiers provides immediate and permanent online open access to all its publications, but this alone is not enough to realize our grand goals.

#### Frontiers Journal Series

The Frontiers Journal Series is a multi-tier and interdisciplinary set of open-access, online journals, promising a paradigm shift from the current review, selection and dissemination processes in academic publishing. All Frontiers journals are driven by researchers for researchers; therefore, they constitute a service to the scholarly community. At the same time, the Frontiers Journal Series operates on a revolutionary invention, the tiered publishing system, initially addressing specific communities of scholars, and gradually climbing up to broader public understanding, thus serving the interests of the lay society, too.

#### Dedication to Quality

Each Frontiers article is a landmark of the highest quality, thanks to genuinely collaborative interactions between authors and review editors, who include some of the world's best academicians. Research must be certified by peers before entering a stream of knowledge that may eventually reach the public - and shape society; therefore, Frontiers only applies the most rigorous and unbiased reviews.

Frontiers revolutionizes research publishing by freely delivering the most outstanding research, evaluated with no bias from both the academic and social point of view. By applying the most advanced information technologies, Frontiers is catapulting scholarly publishing into a new generation.

#### What are Frontiers Research Topics?

Frontiers Research Topics are very popular trademarks of the Frontiers Journals Series: they are collections of at least ten articles, all centered on a particular subject. With their unique mix of varied contributions from Original Research to Review Articles, Frontiers Research Topics unify the most influential researchers, the latest key findings and historical advances in a hot research area! Find out more on how to host your own Frontiers Research Topic or contribute to one as an author by contacting the Frontiers Editorial Office: researchtopics@frontiersin.org

# ENDOMETRIAL CANCER: FROM BIOLOGICAL TO CLINICAL APPROACHES

Topic Editors: Massimo Nabissi, University of Camerino, Camerino, Italy Frederic Amant, KU Leuven, Belgium Paola Gehrig, University of North Carolina at Chapel Hill, United States

Image: Endometrial cancer, New therapeutical options need more research, by Eng. Giacomo Nabissi.

In developed countries, cancer of the endometrium (EC) is the most common cancer of the female reproductive organs, affecting mainly postmenopausal women. The highest incidence of EC is in Northern America and Europe, but its incidence is growing around the world. Several factors influence the risk of developing EC, including: impaired hormone levels, obesity, physical activity, family history, and having been diagnosed with breast or ovarian cancer in the past. For EC, a standard screening test does not exist. It is often diagnosed at stage I/II due to frequent vaginal bleeding, while invasive EC (stage III/IV) is mainly diagnosed in its advanced stage.

Despite the different clinical improvements when EC is progressed to the advanced/ metastatic stage, the patients still have poor prognoses and unsatisfactory outcomes with conventional chemotherapy. The need for precision drugs is underscored by the limited number of options these patients have. So, the introduction of new techniques and the discovery of new biological characteristics, such as identification of transcriptome, proteome and metabolomics profiles, and new potential diagnostic/ prognostic biomarkers are fundamental to improve our knowledges on EC.

The scope of this Research Topic is to welcome articles and reviews from clinicians and scientists around the world in order to give an updated vision of the recent clinical insights, technical advances, and cellular and molecular targets related to endometrial cancer.

Dedicated to Mr. Dario Conti, which supported endometrial cancer research.

Citation: Nabissi, M., Amant, F., Gehrig, P., eds. (2019). Endometrial Cancer: From Biological to Clinical Approaches. Lausanne: Frontiers Media. doi: 10.3389/978-2-88963-049-3

# Table of Contents

*06 Endometrial Cytology as a Method to Improve the Accuracy of Diagnosis of Endometrial Cancer: Case Report and Meta-Analysis* Qing Wang, Qi Wang, Lanbo Zhao, Lu Han, Chao Sun, Sijia Ma, Huilian Hou,

Qing Song and Qiling Li *15 An Efficacious Endometrial Sampler for Screening Endometrial Cancer* Lu Han, Jiang Du, Lanbo Zhao, Chao Sun, Qi Wang, Xiaoqian Tuo, Huilian Hou, Yu Liu, Qing Wang, Qurat Ulain, Shulan Lv, Guanjun Zhang, Qing Song and Qiling Li

*22 A Novel Approach to Preoperative Risk Stratification in Endometrial Cancer: The Added Value of Immunohistochemical Markers* Vit Weinberger, Marketa Bednarikova, Jitka Hausnerova, Petra Ovesna,

Petra Vinklerova, Lubos Minar, Michal Felsinger, Eva Jandakova, Marta Cihalova and Michal Zikan


Neil A. J. Ryan, Niall J. Davison, Katherine Payne, Anne Cole, D. Gareth Evans and Emma J. Crosbie

*55 Assessing the Role of Selenium in Endometrial Cancer Risk: A Mendelian Randomization Study*

Pik Fang Kho, Dylan M. Glubb, Deborah J. Thompson, Amanda B. Spurdle and Tracy A. O'Mara on Behalf of the Endometrial Cancer Association Consortium


Michiel Remmerie and Veerle Janssens


# Endometrial Cytology as a Method to Improve the Accuracy of Diagnosis of Endometrial Cancer: Case Report and Meta-Analysis

Qing Wang1,2†, Qi Wang1,2†, Lanbo Zhao<sup>3</sup> , Lu Han1,2, Chao Sun1,2, Sijia Ma1,2, Huilian Hou<sup>4</sup> , Qing Song<sup>1</sup> and Qiling Li 1,2 \*

<sup>1</sup> Center for Single-Cell Biology, First Affiliated Hospital, Xi'an Jiaotong University, Xi'an, China, <sup>2</sup> Department of Obstetrics and Gynecology, First Affiliated Hospital, Xi'an Jiaotong University, Xi'an, China, <sup>3</sup> Guipei 77, Health Science Center, Xi'an Jiaotong University, Xi'an, China, <sup>4</sup> Department of Pathology, First Affiliated Hospital, Xi'an Jiaotong University, Xi'an, China

#### Edited by:

Massimo Nabissi, University of Camerino, Italy

#### Reviewed by:

Johanna Pijnenborg, Radboud University Nijmegen Medical Centre, Netherlands Gökhan Açmaz, Erciyes University, Turkey

> \*Correspondence: Qiling Li liqiling@mail.xjtu.edu.cn

†These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Women's Cancer, a section of the journal Frontiers in Oncology

Received: 15 September 2018 Accepted: 21 March 2019 Published: 24 April 2019

#### Citation:

Wang Q, Wang Q, Zhao L, Han L, Sun C, Ma S, Hou H, Song Q and Li Q (2019) Endometrial Cytology as a Method to Improve the Accuracy of Diagnosis of Endometrial Cancer: Case Report and Meta-Analysis. Front. Oncol. 9:256. doi: 10.3389/fonc.2019.00256 More and more researchers have reported that dilatation and curettage (D&C) or Pipelle had low accuracy, high misdiagnosis, and insufficient rate. Endometrial cytology is often compared with histology and seems to be an efficient method for the diagnosis of endometrial disorders, especially endometrial cancer. We report a case of misdiagnosed endometrial cancer by D&C, but with a positive cytopathological finding. Following that, a meta-analysis including 4,179 patients of endometrial diseases with cyto-histopathological results was performed to assess the value of the endometrial cytological method in endometrial cancer diagnosis. The pooled sensitivity and specificity of the cytological method in detecting endometrial atypical hyperplasia or cancer was 0.91[95% confidence interval (CI) 0.74–0.97] and 0.96 (95% CI 0.90–0.99), respectively. The pooled positive likelihood ratio and negative likelihood ratio was 25.4 (95% CI 8.1–80.1) and 0.10 (95% CI 0.00–0.30), respectively. The diagnostic odds ratio which was usually used to evaluate the diagnostic test performance reached 260 (95% CI 36–1905). So we recommend that D&C and Pipelle are still practical procedures to evaluate the endometrium, cytological examinations should be utilized as an additional endometrial assessment method.

#### Keywords: cytology, histology, endometrial cancer, diagnosis, atypical hyperplasia

#### INTRODUCTION

Endometrial cancer is becoming the primary reason of female deaths of genital track cancer in developed countries (1). Dilatation and curettage (D&C), as the traditional gold standard procedure for diagnosing endometrial cancer, is painful, expensive, requires general anesthesia and has a high rate of misdiagnosis (2). It has been reported that less than half of the uterine cavity is curetted in 60% of cases (3), and over 40% of women with complex atypical hyperplasia as a preoperative diagnosis have a final confirmation of endometrial cancer during hysterectomy (4, 5). Endometrial cytology is recently reported as a useful diagnostic method with high sensitivity and specificity in detecting endometrial malignancies (6–9), but no meta-analysis, which is considered more credible, has yet been performed to evaluate the diagnostic accuracy of endometrial cytology for endometrial carcinoma compared with histological diagnosis.

**6**

Here, we report a case of misdiagnosed endometrial cancer by D&C, but with a positive cytopathological finding. The patient has provided her written informed consent for the publication of this manuscript and any identifying images or data. After searching on PubMed, we believe it is the first case report of a misdiagnosis of endometrial cancer detected by cytopathology. Following this, a random-effects meta-analysis including 4,179 patients with both cytopathological and histopathological results was performed to assess the value of the endometrial cytology method in the diagnosis of endometrial atypical hyperplasia or cancer.

# CASE REPORT

A 60-year-old post-menopause female, from Baoji City of the Shaanxi province in China, went to a local hospital complaining of abnormal uterine bleeding for 2 months. No high risk factor for endometrial cancer was observed, such as genetic factors, obesity, diabetes, a history of tamoxifen use and so on. Curettage was performed with a histopathological diagnosis of complex hyperplasia endometrium. No medicine or therapeutic curettage was effective for her with a continued bleeding. Her type B ultrasound in Shaanxi Provincial People's hospital showed a 0.8 cm-thick endometrium. Then, she turned to the First Affiliated Hospital of Xi'an Jiaotong University for further treatment. After written informed consent, she volunteered to get cytological endometrial samplings by Li Brush (Xi'an Meijiajia Bio-Technologies Co. Ltd., China, 20152660054) for cytological examination before D&C. Her histopathological report revealed that papillary epithelial hyperplasia was found, and cancer was a concern according to the structure of tissue but could not be diagnosis due to insufficient tissue (**Figure 1A**). Meanwhile, the cytopathological report revealed that some malignant cells were found (**Figure 1B**). Her serum markers showed high serum carbohydrate antigen 19-9 (CA19-9, 42.08 U/ml) and squamous cell carcinoma antigen (SCC, 6.10 ng/ml). A diagnostic laparoscopic hystero-salpingo-oophorectomy was performed and the patient was converted to a laparotomy when intraoperative frozen section examination revealed an endometrial serous carcinoma with ovarian metastasis. Omentum resection, pelvic lymphadenectomy and para-aortic nodes dissection were performed. She was finally diagnosed with stage IIIc endometrial serous carcinoma.

# MATERIALS AND METHODS

#### Source of Material

We searched the PubMed and Embase databases with the heading terms and keywords as "cytology" and "endometrial" from Jan 1, 1995 to June 1, 2018. Then, the results were manually selected for studies to include and repeatedly checked by a second investigator. We searched the full-text articles about the comparison of cytological results and the histological results in endometrial samples.

## Standard of Inclusion and Exclusion

All candidate studies were evaluated and extracted by two independent investigators. Inclusion criteria: (1) patients were diagnosed by histopathological and cytological examination; (2) the histopathological results were paired with cytological results; (3) sufficient information was provided to conduct a statistical analysis; (4) endometrial cells were sampled by endometrial brushes; (5) studies were limited to human trials and published in English. Exclusion criteria include: (1) news, abstracts, case reports, letters, commentaries, and reviews studies; (2) other kinds of endometrial cells sampler like endometrial aspiration cytology; (3) different cytopathology report formats with others, that made it hard to re-group and analyze; and (4) studies with different positive result definition or duplicate data.

# Data Extraction

We set atypical hyperplasia and endometrial carcinoma as the positive results and the others as the negative results, including normal endometrium, non-atypical hyperplasia, endometrial polyp, simple endometrial hyperplasia, complex endometrial hyperplasia and so on.

Two investigators separately extracted the following information from each research: the name of the first author, year of publication, cytological sampling method, cytological specimen preparation, histological sampling method, number of patients enrolled, and true positive (TP), false negative (FN), false positive (FP), and true negative (TN) results. Any discrepancies between the two investigators were discussed by all the authors.

# Quality Assessment

A quality assessment of eligible studies was evaluated using the quality assessment of studies of diagnostic accuracy included in systematic reviews-2 (QUADAS-2). There were 13 questions (each of which was scored as yes, no, or unclear): (1) Was a consecutive or random sample of patients enrolled? (2) Was a case–control design avoided? (3) Did the study avoid inappropriate exclusions? (4) Were the cytopathological diagnoses interpreted without knowledge of the results of the gold standard (histopathological diagnosis)? (5) whether the blind method was used for pathologists? (6) Was the histopathological diagnosis likely to correctly classify the target condition? (7) Was there an appropriate interval between the cytological sampling and histological sampling? (8) Did all patients receive the histopathological diagnosis? (9) Were all patients included in the analysis? (10) Whether the diagnostic test steps were detailed? (11) Were there concerns that the included patients and setting do not match the review question? (12) Were there concerns that the target condition as defined by the gold standard does not match the question? (13) Were there concerns that the index test, its conduct, or its interpretation differ from the review question?

#### Statistical Analysis

The publication bias was checked by a Deeks funnel plot, and P < 0.05 was considered a significant publication bias. Statistical heterogeneity was detected by a Q test and an inconsistency index (I 2 ), with significant heterogeneity set at P ≤ 0.05 and I <sup>2</sup> > 50%.

If there was no significance in heterogeneity (P > 0.05), a fixed effects model was chosen. If it was the opposite (P < 0.05), a random effects model was chosen.

According to TP, FN, FP and TN results, we calculated sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), positive likelihood ratio (PLR) (>10 suggested strong concordance), negative likelihood ratio (NLR) (<0.1 suggested strong concordance), diagnostic odds ratio (DOR). PLR was calculated as: positive likelihood ratio = sensitivity/(1–specificity). NLR was calculated as: negative likelihood ratio = (1–sensitivity)/specificity. DOR was estimated by the Mantel-Haenszel formula. All statistical analyses, including 95% Confidence Interval (CI) were performed using STATA software (version 12.1, StataCorp LP) with the Midas module.

# RESULTS

# Search Results

After searching on PubMed and Embase, 9 of 4,182 studies were included in meta-analysis. **Figure 2** showed a flow diagram of the selection process. All data in researches were screened rigorously by our team.

#### Basic Characteristics of Studies

Our analysis included 9 eligible studies, which were shown in **Table 1**. In total, 2 studies were from Italy, 2 from the USA, 1 from China, 1 from Japan, 1 from England, 1 from Indonesia, and 1 from Greece. A total of 4,179 patients were included. Different endometrial brushes were used in these 9 studies, including the Tao brush (2), Endoflower (2), Endogyn (1), Cytobrush (1), and Uterobrush (1), and 1 study used six different devices. In all, 8 studies prepared the cytology specimens with a liquid-based cytology, and 1 study used the conventional way. In sum, 2 studies compared the cytological results to the D&C results, 3 studies compared the cytological results to the hysterectomy results, 2 studies compared the cytological results to the hysteroscopy and biopsy results, 1 study compared the cytological results to the biopsy or D&C results and 1 study compared the cytological results to the biopsy, D&C or hysterectomy results. Additionally, 5 studies research the pre/post-menopausal patients, 2 studies researched peri/post-menopausal patients, 1 study researched post-menopausal patients, and in 1 study, the menopause situation was unknown.

#### Study Quality

We assessed the quality of eligible studies by QUADAS-2 and found that the quality of all the studies was good (**Table 2**).

#### Diagnostic Accuracy

The pooled sensitivity and specificity of the cytological method in detecting endometrial atypical hyperplasia or cancer was 0.91 (95% CI 0.74–0.97) and 0.96 (95% CI 0.90–0.99), respectively (**Figure 3**). The pooled PLR and NLR were 25.4 (95% CI 8.1– 80.1) and 0.10 (95% CI 0.00–0.30), respectively. The DOR which used to evaluate the diagnostic test performance,reached 260 (95% CI 36–1905).

#### Heterogeneity and Sensitive Analysis

I 2 values of pooled sensitivity and specificity were 96.53 (95% CI, 95.23–97.83) and 98.29 (95% CI, 97.78–98.80), which indicated a statistically significant heterogeneity. A sensitivity analysis was performed to assess the influence of each study, in which each individual study was removed each time. No significant change or reversal of result was found (**Table 3**). The I 2 -value showed that none of the single study affected the heterogeneity of this meta-analysis.

#### Subgroup Analysis

Subgroup 1: the corresponding values of the subgroup with sample size <300 were 0.89 (95% CI: 0.55, 0.98) for sensitivity and 0.93 (95% CI: 0.81, 0.98) for specificity. While in subgroup sample size ≥300, the sensitivity was 0.93 (95% CI: 0.71, 0.99) and specificity was 0.98 (95% CI: 0.91, 1.00). Subgroup 2: the corresponding values of the studies of European countries showed the sensitivity of 0.96 (95% CI: 0.86, 0.99) and specificity of 0.99 (95% CI: 0.94, 1.00) and in other countries were 0.84(95% CI: 0.50, 0.96) and 0.92 (95% CI: 0.79, 0.97), respectively (**Table 3**).

#### Publication Bias Analysis

Deeks funnel plot asymmetry test was conducted to evaluate publication bias in this study (**Figure 4**), which showed statistically nonsignificant publication bias (P = 0.60).

#### Clinical Utility

Given the PLR and NLR, the cytological detection method of endometrial atypical hyperplasia or cancer was located in the left upper quadrant (**Figure 5A**), indicating that the cytological detection method could serve as a test to confirm and exclude endometrial atypical hyperplasia or cancer. Fagan's plot indicated a dramatic improvement in posttest probability. When the pretest probability of endometrial atypical hyperplasia or cancer was set to 20%, using the cytological method as a source to detect the above diseases could significantly raise the posttest probability of a positive result to 86% and lower the posttest probability of a negative result to 2% (**Figure 5B**).

#### DISCUSSION

It is currently estimated, that 60 million cervical cytology examinations are performed every year in the United States (19). Cytopathological screening, histopathological diagnosis and even the human papillomavirus vaccine are used to prevent and to make early diagnosis of cervical cancer, which helps the early detection and lowers the mortality of cervical cancer. In the absence of such effective screening programs and prevention methods, endometrial malignant diseases are becoming the most prevalent cancer of the female genital tract in developed countries, accounting for nearly 50% of all new diagnoses of gynecological cancer (20, 21). Nearly 75–80% of all endometrial cancer patients diagnosed at early stage (22, 23). But researchers are still paying attention to the early diagnosis of endometrial malignant diseases, especially its precancerous lesion.

Histological (D&C and Pipelle) and cytological diagnosis are two classes of endometrial sampling modalities. Both D&C and the most golden standard of evaluating the endometrium, hysteroscopic-guided uterine biopsy, are painful, expensive, and requires dilatation and anesthesia (24–26). Insufficient samples carry negative ramifications and increase the difficulty for the pathologist (27). An insufficient rate was reported as 6.4% (810/12,745) of curettage and 6.5% (310/4,777) of endometrial biopsy, and multiple factors contributed to such variation, including patient age, parity, endometrial thickness, sampling device, and provider technique. When stratified by age, the insufficient rate was 2.7% in the group of patients under 40 years old (3,454 cases), 5.8% in the group of 40 to 59 years old (11,838 cases), and 14.6% in the group of 60 years and older (2,230 cases) (28, 29). Sakhdari et al. (27) also showed that 15% (226/1,768) of the samples of women age 60 and older were reported as insufficient, and Barut et al. reported the insufficient rate was likely associated with menopause, with 6.5% (26/401) in premenopausal and 49.2% (120/244) in postmenopausal women (25). However, 75% of endometrial cancers occurred in women older than 55 years of age, with a median age of 62 (30). A meta-analysis evaluated the diagnostic rate of D&C and hysteroscopy in postmenopausal women. It pointed out that D&C had a high rate of non-diagnostic samples 31% (range 7–76%) and a high failure rate of 11% (range 1–53%), which lead to a missing diagnosis rate of 7% (range 0–18%) (31).

Pipelle, as another widely used endometrial biopsy apparatus, is safe, cost-effective, and easily preformed (24). A meta-analysis


TABLE 1 | Study characteristics of the nine included studies on the diagnostic accuracy of endometrial cytological sampling.

LBC, Liquid - based cytology; NA, not available; D&C, dilatation and curettage.

TABLE 2 | Risk of bias and concerns of applicability by study using a modified Quadas-2 tool.


of 39 studies, including 7,914 patients, revealed the concordance rate between Pipelle and D&C/hysteroscopy/hysterectomy in endometrial cancer detection of postmenopausal and premenopausal women was 99.6% and 91%, respectively (32). However, the Pipelle is random point sampling and said to sample 4.2% of the uterine cavity (33), and 25–36% women using Pipelle were found to have insufficient tissue for pathologic assessment (34).

Both D&C and Pipelle have their limitations in detecting endometrial cancer. Hysteroscopic guided biopsy showed a high diagnostic accuracy for endometrial cancer diagnosis (estimated sensitivity of 82.6% and specificity of 99.7%), data from a meta-analysis over 9,000 patients (35), but it could not be performed on asymptomatic women or used as a screening method. Are histological procedures (curettage or biopsy) enough to be the only methods in the diagnosis of endometrial diseases?

Endometrial cytology examination may be an inevitable method for endometrial cancer screening and a combined diagnostic procedure. It might have been hampered by the frequent presence of excess blood, mucus and overlapping cells and varied endometrium cell morphology with different sex

hormone levels. However, liquid-based preparation techniques improve the diagnostic accuracy of endometrial cytology (6, 36), and more and more scholars have made efforts on the endometrial cytology reporting system (37–39). With the establishment and maturation of universal standards for the reporting system, endometrial cytology will truly play an important role in the diagnosis of endometrial diseases and endometrial cancer screening. Kondo et al. tested some different methods in 114 consecutive symptomatic women, and they reported that the sensitivity of detecting malignancy increased


TABLE 3 | Sub-analysis and sensitivity analysis on the diagnostic accuracy of endometrial cytological sampling.

SEN, sensitivity; SPE, specificity; PLR, positive likelihood ratio; NLR, negative likelihood ratio; DOR, diagnostic odds ratio; CI, confidence interval.

from 92% to 98% when endometrial cytology was combined with suction curettage (40).

Our meta-analysis showed that endometrial cytology had a high diagnostic accuracy and could serve as a test to confirm or exclude endometrial atypical hyperplasia or cancer. The pooled sensitivity and specificity of the cytological method in detecting endometrial atypical hyperplasia or cancer was 0.91 (95% CI 0.74–0.97) and 0.96 (95% CI 0.90–0.99), respectively. Its diagnostic odds ratio reached 260 (95% CI 36–1905). The pooled positive likelihood ratio and negative likelihood ratio was 25.4 and 0.10, respectively. Therefore, we can conclude that the test results of endometrial cytology are very accurate in diagnosing endometrial atypical hyperplasia or cancer.

Therefore, we recommend that D&C and Pipelle are still practical procedures to evaluate the endometrium, cytological examinations should be utilized as an additional endometrial assessment method, especially for women at high-risk for endometrial cancer.

Additionally, endometrial cytology is inexpensive, tolerated well and can be performed without anesthesia in an outpatient clinic. It is now the most common test for an initial evaluation of endometrial cancer in Japan (7) and has been encouraged as the first level screening method for women at high risk for endometrial cancer (37). Japanese epidemiological data revealed that the overall death rate of endometrial cancer decreased from 20.0 per 100,000 in 1950 to 8.0 per 100,000 in 1999, and this was thought to be a consequence of cytological screening (41).

Many researchers reported a high risk of endometrial cancer with positive cervical cytology (42, 43). Abnormal cervical cytology was associated with high-grade endometrial cancer, worse 5-year median recurrence-free survival and worse diseasespecific survival (44). Positive cervical cytology should also be considered as a high risk of endometrial cancer, and endometrial cytology may benefit this kind of patients even with no clinical symptom.

An important strength of this meta-analysis is that we performed a thorough search for articles on the diagnostic accuracy in women with endometrial atypical hyperplasia or cancer using endometrial cytology. This article has several limitations. First, the risk of missing potentially relevant articles is a concern. Otherwise, the relatively small number of studies and variability in methods did not allow for more standard statistical analyses. Higher sensitivity and specificity could be found in subgroup of studies with sample size ≥ 300 and studies in European countries. However, after the subgroup analyses and sensitivity analysis, no factor showed associated with high heterogeneity. Patient age, menopause or not, different

#### REFERENCES


kinds of clinical symptoms, varies of cytological samplers and histological sampling methods might contribute to the high heterogeneity, and further study should approve it with enough data. What's more, the studies that are included in the metaanalysis are performed in symptomatic women. More data are needed before endometrial cytology being an effective screening tool for asymptomatic women with high-risks of endometrial cancer.

#### CONCLUSION

In conclusion, endometrial cytology is an efficient diagnostic method and could be applied in the diagnosis of endometrial disorders. The diagnostic accuracy of endometrial carcinoma will surely be improved by the combination of cyto-histopathological procedures and vaginal ultrasonography. Moreover, cytological examination, as a proper outpatient procedure, should be advised for endometrial screening, especial for those with high-risks of endometrial cancer.

#### AUTHOR CONTRIBUTIONS

QingW and QiW drafted the manuscript. QingW, LZ, and CS collected the case. QiW, LH, and SM performed the meta analysis. HH performed the pathological figure. QS helped to revise the manuscript. QL conceptualized the study.

#### FUNDING

This study was partly supported by the Shaanxi Provincial Collaborative Technology Innovation Plan (2017XT-026, 2018XT-002), Key Research and Development Project of Shaanxi Province Science and Technology Department (2017ZDXM-SF-068) and Medical Research Project of Xi'an Social Development Guidance Plan (2017117SF/YX011-3).


associated with curettage in the investigation of endometrial lesions from postmenopausal women. Cytopathology. (2005) 16:32–9. doi: 10.1111/j.1365-2303.2004.00224.x


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Wang, Wang, Zhao, Han, Sun, Ma, Hou, Song and Li. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# An Efficacious Endometrial Sampler for Screening Endometrial Cancer

Lu Han1,2, Jiang Du<sup>3</sup> , Lanbo Zhao<sup>4</sup> , Chao Sun1,2, Qi Wang1,2, Xiaoqian Tuo1,2 , Huilian Hou<sup>5</sup> , Yu Liu<sup>5</sup> , Qing Wang1,2, Qurat Ulain1,2, Shulan Lv <sup>2</sup> , Guanjun Zhang<sup>5</sup> , Qing Song1,6 and Qiling Li 1,2 \*

*<sup>1</sup> Center for Single-Cell Biology, First Affiliated Hospital, Xi'an Jiaotong University, Xi'an, China, <sup>2</sup> Department of Obstetrics and Gynecology, First Affiliated Hospital, Xi'an Jiaotong University, Xi'an, China, <sup>3</sup> Department of Gynecology and Obstetrics, Shaanxi Provincial People's Hospital, Xi'an, China, <sup>4</sup> Guipei 77, Health Science Center, Xi'an Jiaotong University, Xi'an, China, <sup>5</sup> Department of Pathology, First Affiliated Hospital, Xi'an Jiaotong University, Xi'an, China, <sup>6</sup> Cardiovascular Research Institute, Morehouse School of Medicine, Atlanta, GA, United States*

Recently, the research on early detection of precancerous change and endometrial carcinoma has been focusing on minimally invasive procedures for screening. On this basis, we aim to verify the feasibility of endometrial samplers for screening endometrial cancer using Li Brush. We recruited patients undergoing hysterectomy for different diseases from the Inpatient Department of the Department of Obstetrics and Gynecology. Before surgery, endometrial cells were collected by Li Brush. The cytopathologic diagnosis from Li Brush and the histopathologic diagnosis from hysterectomy in the same patient were compared to calculate sensitivity (Se), specificity (Sp), false-negative rate (FNR), false-positive rate (FPR), positive predictive value (PV+) %, and negative predictive value (PV-). The research enrolled 293 women into this self-controlled trial. According to the hypothesis test of paired four lattices, we obtained the following indicators: Se 92.73, Sp 98.15, FNR 7.27, FPR 1.85, PV+92.73, and PV−98.15%. The endometrial sampler Li Brush is an efficacious instrument for screening endometrial cancer.

#### Edited by:

*Massimo Nabissi, University of Camerino, Italy*

#### Reviewed by:

*Margaret Liang, University of Alabama at Birmingham, United States Saori Furuta, University of Toledo, United States*

\*Correspondence:

*Qiling Li Liqiling@mail.xjtu.edu.cn*

#### Specialty section:

*This article was submitted to Women's Cancer, a section of the journal Frontiers in Oncology*

Received: *09 September 2018* Accepted: *24 January 2019* Published: *19 February 2019*

#### Citation:

*Han L, Du J, Zhao L, Sun C, Wang Q, Tuo X, Hou H, Liu Y, Wang Q, Ulain Q, Lv S, Zhang G, Song Q and Li Q (2019) An Efficacious Endometrial Sampler for Screening Endometrial Cancer. Front. Oncol. 9:67. doi: 10.3389/fonc.2019.00067* Keywords: endometrial cancer, endometrial sampler, cytology, histopathology, screening

# INTRODUCTION

The morbidity and mortality of endometrial carcinoma is on the rise around the world in recent years. It has been the most common gynecologic malignancy in some developed countries such as Japan and US and ranked second in many developing countries (1, 2). The cancer-related costs are increasing significantly, constituting a challenge for social economics and female health. Efforts focusing on primary and secondary prevention remain central to the global charge to reduce the incidence of cancer and avoid one-third to one-half of cancer deaths (2, 3). With developing morbidity of endometrial cancer around the world, early detection and diagnosis would undoubtedly become the most important part. For endometrial carcinoma, the 5-year survival rate gradually decreases with the development of the stages. Eighty percent of the patients diagnosed with endometrial cancer are in stage I, with a 5-year survival rate of >95% (4). Endometrial atypical hyperplasia is considered to be the precancerous lesions of endometrial cancer. Thirty percent of atypical hyperplasia will develop into cancer a long time in the future; thus, we have the opportunity to screen for endometrial cancer within this long time period (5). The aim of screening is to detect endometrial atypical hyperplasia and the early stages of endometrial cancer. The ability to save lives would mean great social significance and economic benefits.

Endometrial carcinoma is a type of epitheliogenic malignant tumor that originates from the endometrium. As one of three major malignant tumors of the female reproductive system, the average onset age of endometrial carcinomas is 63 years, and > 90% occur in women above 50 years of age, and ∼4% occur in women younger than 40 years of age (4). Risk factors for endometrial cancer include early menarche (6), late menopause, nulliparity, Lynch syndrome (7), diabetes (8, 9), obesity (10), hypertension, estrogen, and tamoxifen treatment after menopause (11, 12), a family history of endometrial cancer or breast cancer (13), and polycystic ovary syndrome (14).

Because of the rising morbidity, the window period, and an explicit screening population, endometrial cancer screening is feasible. Until now, histopathology with dilatation and curettage (D&C) with or without hysteroscopy and surgery has been the gold standard for the diagnosis of endometrial carcinoma and precancerous lesions (15). However, the injury and discomfort caused by D&C have influenced its widespread use for screening. In recent years, more and more non-invasive endometrial devices have been invented and proposed for screening endometrial cancer, such as Pipelle, which was found to have an 86% sensitivity in one study (16); Tao Brush, which was found to have a 95.5% sensitivity (16); and SAP-1, which was found to have a 73% sensitivity, 95.8% specificity, 75% positive predictive value, and 95.3% negative predictive value (17). However, no available, specific, and effective screening method could be applied popularly for women until now. In this study, we compared the cytopathologic diagnosis of the Li Brush (**Figure 1A**, Xi'an Meijiajia Medical Co. 20152660054) with the histopathologic diagnosis of hysterectomy to evaluate the feasibility of the endometrial samplers for screening endometrial cancer.

# MATERIALS AND METHODS

#### Patients and Study Procedures

From January 2015 to July 2016, we recruited patients undergoing hysterectomy because of different diseases from the Inpatient Department (IPD) of the Department of Obstetrics and Gynecology. Patients in the IPD were excluded if they had already been diagnosed with pregnancy, acute inflammation of the genital system besides atrophic vaginitis, endogenous cervical carcinoma, dysfunction of blood coagulation, and other hematologic diseases that might influence coagulation function. Women with a body temperature >37.5◦C in two subsequent measurements in 1 day were also ruled out.

According to the following procedures, we collected endometrium specimens using the Li Brush and obtained the cytopathologic diagnosis from the Department of Pathology before surgery. First, the patients were placed in the lithotomy position, and the conventional perineal and vaginal disinfection were performed after emptying the bladder. Second, the uterine cervix was exposed by vaginal speculum and the uterine depth was detected with uterine probe. After the brush head was hidden in the drivepipe, the sampler was put into the fundus of uterus (**Figure 1B**). Then, the drivepipe was drawn out ∼5 mm to show the brush, and the handle was rotated 5–10 complete circles to gather cells of the uterine corpus (**Figure 1C**). Third, the drivepipe was advanced 3 mm and the handle was again rotated to gather cells from the uterine fundus (**Figure 1D**). Finally, the brush was removed from uterine cavity after protecting the brush head under the casing (**Figure 1E**). When sampling was complete, the brush head was placed into the preservation solution and shaken several times to release the cells into the solution.

The cell specimens were prepared for testing with a liquidbased cytologic test, and the tissues were embedded in paraffin and cut into cross-sections. Both cell and tissue specimens were stained using hematoxylin and eosin. After concealing identity information, all the samples were sent to the Department of Pathology of First Affiliated Hospital of Xi'an Jiaotong University and randomly diagnosed by two independent professors. The diagnoses of histology and cytology were independently conducted. According to the method of sample size calculation in the diagnosis experiment (18) and using the sensitivity of SAP-1 brush (4, 17) to predict the sensitivity of Li Brush, a minimum of 113 cases were required for the study. The self-control method was used in this study. The histopathologic diagnosis from the hysterectomy was defined as the standard. The outcomes were obtained by comparing the cytopathologic and histopathologic results of the same individuals.

The study was conducted in accordance with the Declaration of Helsinki, and the study protocol was approved by the Ethics Committee of the First Affiliated Hospital of Xi'an Jiaotong University(XJTU1AHCR2014-007). All the patients involved in the research were sufficiently informed of the content of the study and provided written informed consent.

#### Data Collection

For all eligible patients, the following information was collected: age, age at menarche, last menstrual period or menopausal age, childbearing history, endometrial thickness, tumor history, smoking history, with or without hormone replacement, and history of other diseases such as hypertension and diabetes. The data were used to determine sampling satisfaction, cytopathologic diagnosis, and histopathologic diagnosis.

#### Definition of Outcomes

According to the International Society of Gynecological Pathologists, the histopathologic diagnoses included the following: proliferative endometrium, secretory endometrium, atrophic endometrium, mixed endometrium, and simple hyperplasia including cystic glandular hyperplasia, complex hyperplasia defined as adenomatous hyperplasia without atypia, endometrial atypical hyperplasia, and endometrial carcinoma. The cytopathologic diagnoses were classified into seven categories, as follows: proliferative endometrial cells, secretory endometrial cells, atrophic endometrial cells, mixed endometrial cells, endometrial hyperplasia cells, endometrial atypical cells, and endometrial cancer cells. Positive results were defined as endometrial carcinoma, endometrial cancer cells, endometrial atypical hyperplasia, and endometrial atypical cells. Other categories were defined as negative results. When both cellular and histionic diagnostic results were positive, it was judged as true positive; if both were negative, it was

judged as true negative. If the cytopathologic result was positive and the histopathologic result was negative, it was judged as false positive; if the cytopathologic result was negative and the histopathologic result was positive, it was judged as false negative. Consistent outcome was when the cytopathologic and histopathologic diagnoses were both positive or both negative; otherwise, the outcomes were considered inconsistent. The sampling satisfaction was reflected in a sufficient number of cells and the correct location.

#### Statistical Analysis

Using the hypothesis test of paired four lattices, the following indicators were calculated: sensitivity (Se), false-negative rate (FNR), specificity (Sp), false-positive rate (FPR), positive predictive value (PV+), and negative predictive value (PV-). The differences of endometrial histopathology and cytopathology using Li Brush in the diagnosis of endometriosis were evaluated by the calculation of P-value. If P < 0.05, the difference was statistically significant. In contrast, if P > 0.05, the difference was not statistically significant.

#### RESULTS

#### Patients

We aimed to collect a total of 420 patients from the IPD of the Department of Obstetrics and Gynecology. We ruled out 112 patients because of insufficient or incomplete information. The remaining 308 women completed the study (**Table 1**). Satisfactory endometrial cells were not obtained in 37 patients (**Figure 2**).

#### Pathologic Images

Using pathological slide and microscopic camera technology, the following histrionic images were obtained: proliferative endometrium (**Figure 3A**, left), secretory endometrium (**Figure 3B**, left), atrophic endometrium (**Figure 3C**, left), mixed endometrium (**Figure 3D**, left), endometrial atypical hyperplasia (**Figure 3E**, left), and endometrial carcinoma (**Figure 3F**, left). The corresponding cytopathologic images were included: proliferative endometrial cells (**Figure 3A**, right), secretory endometrial cells (**Figure 3B**, right), atrophic endometrial cells (**Figure 3C**, right), mixed endometrial cells (**Figure 3D**, right), endometrial atypical cells (**Figure 3E**, right), and endometrial cancer cells (**Figure 3F**, right).

#### Data Calculation

According to the hypothesis test of paired four lattices, there were 51 true-positive, 212 true-negative, 4 false-positive, and 4 false-negative cases. The following indices were obtained: Se 92.73, Sp 98.15, FNR 7.27, FPR 1.85, PV+ 92.73, and PV- 98.15%. The data showed that there were no significant difference between cytopathologic results from the Li Brush and histopathologic results from hysterectomy (χ <sup>2</sup> = 0.125 < χ 2 0.05, α = 0.05, P > 0.05).

Furthermore, we compared the histopathologic results obtained from hysterectomy and the cytopathologic results of our samplers accurately to evaluate the feasibility of using Li Brush in the diagnosis of endometrial types. Through the statistics, there were 228 cases consistent and 34 cases inconsistent. In addition, 9 cases were diagnosed as endometrial simple hyperplasia with local polyps by hysterectomy and endometrial hyperplasia cells by our samplers. The overall degree of satisfaction and sensitivity of sampling were 87.02 and 87.63%, respectively (**Table 2**). The sensitivities of different types of endometrium were 85.88% for proliferative endometrium, 72.73% for secretory endometrium, 88.24% for atrophic endometrium, 83.33% for mixed endometrium, 94.64% for simple hyperplasia, 100% for complex hyperplasia, 80.00% for endometrial atypical hyperplasia, and 87.02% for endometrial carcinoma. Meanwhile,

#### TABLE 1 | Patient characteristics.


*Some information of the patients is missing.*

*a IPD, Inpatient Department.*

*<sup>b</sup>AUB, Abnormal uterus bleeding.*

*<sup>c</sup>Some patients were not examined by ultrasound, whose endometrial thickness is missing.*

the results of cells and tissues were separately analyzed and compiled into a bar chart by composition ratios (**Figure 4**).

#### DISCUSSION

Endometrial curetting or D&C have long been the standard diagnosis or treatment for evaluating suspicious endometrial lesions, especially in mainland China. Only patients who display symptoms that are geared to the indication of D&C, such as abnormal uterine bleeding, would refer themselves to medical help for diagnosis and treatment, which restricts the early diagnosis and treatment of endometrial carcinomas. Because only approximately 60% of curetting procedures can evaluate less than half of the uterine cavity, even when performed by the most experienced physician, the rate of false-negative results is high (19). Moreover, the pain and suffering caused by the procedure were not widely accepted patients. Despite its diagnostic value for patients who display symptoms, this method shows its deficiency as a screening procedure for endometrial lesions (20).

Recently, research on the early detection of endometrial carcinoma has been focusing on minimally invasive histopathologic and cytopathologic procedures (16, 21–24). Along with the improvement and widespread use of liquidbased preparation (LBP) of endometrial cell samples, direct cellular sampling has more commonly used as the primary screening procedure for endometrial lesions (25). Among the sampling techniques, endometrial brush cytology is minimally invasive, more economical, and more convenient compared with traditional diagnostic and curettage techniques. Therefore, it has already become a partly accepted method for the detection of endometrial lesions (26). The endometrial cytology by direct intrauterine sampling has a relatively high specificity and sensitivity for the diagnosis of endometrial cancer reported by some researchers (27). Related samplers have been studied, including histology samplers such as Pipelle (28), and cytology samplers, such as Tao brush (19) and SAP-1 sampler (4, 17), as well as Uterobrush (26). However, up to this point, we still required a more convenient, economical, and non-invasive tool to screen for endometrial cancer.

On that basis, we have invented a new endometrial sampler— Li Brush. Our sampler was awarded a utility model patent certificate from the State Patent Office (number: ZL.2014 2 0720056.8). The brush is made up of four parts: head,tube core, drivepipe, and hand shank (**Figure 1A**). The head is T shaped, which is close to the physiologic form of the uterine cavity. The fusiform brush allows easy access to the uterine cavity, fundus, and horn of the uterus. The elastic drivepipe works with the handle to protect the head from contamination of cervical cells. In addition, our samplers have other advantages, such as low cross-infection, good flexibility, less damage, low cost, and higher acceptability. In this study, we compared the diagnosis of cytology by the Li Brush with the diagnosis of histopathology by hysterectomy. A total of 271 cases were analyzed, with a sensitivity of 92.73% and a specificity of 98.15%. The proportions of the endometrial types were similar between histology and cytology. These findings showed that Li Brush will be able to play a role in the screening for endometrial cancer.

The two common causes of sampling failure using Li Brush were the inaccurate location of sampling and insufficient number of endometrial cells. The lower sampling satisfaction of atypical hyperplasia and complex hyperplasia are because of the limited sample size. The lower sampling satisfaction of atrophic endometrium is attributed to the atrophic cervix and the adhesion of cervix tube in post-menopausal women. To improve the sampling satisfaction, we considered whether using cervical



*Se, Sensitivity.*

*<sup>a</sup>There were 4 cases which the histopathological results were endometrial cancer but the cytopathologic results were proliferative endometrial or endometrial hyperplasia cells. One of this 5 cases which the histopathological result of was endometrial cancer but the cytopathologic result was endometrial atypical cells, so this case was considered to be true positive but inconsistent.*

*<sup>b</sup>There were 9 of all 308 cases diagnosed endometrial simple hyperplasia with local polyp.*

FIGURE 3 | Histopathologic and cytopathologic images. (A) proliferative endometrium (Left: HE × 400) and proliferative endometrial cells (Right: HE × 100); (B) secretory endometrium (Left: HE × 10) and secretory endometrial cells (Right: HE × 10); (C) atrophic endometrium (Left: HE × 10) and atrophic endometrial cells (Right: HE × 10); (D) mixed endometrium (Left: HE × 10) and mixed endometrial cells (Right: HE × 10); (E): endometrial atypical hyperplasia (Left: HE × 10) and endometrial atypical cells (Right: HE × 200); (F) endometrial carcinoma (Left: HE × 400) and endometrial cancer cells (Right: HE × 400).

clamps when using the brush could make it easier for the brush to smoothly enter the uterine cavity. We can also mechanically expand the cervical canal for patients with cervical adhesion or narrowing if necessary.

For endometrial polyps, Li Brush showed a high false-negative rate. According to histology, endometrial polyps are classified into four categories: non-functional glandular endometrial polyp, functional glandular endometrial polyp, adenomatoid polyp,

and polyp with malignant transformation (5). Polyps consist of proliferating glands, blood vessels, and stroma because of the hyperplasia due to the high sensitivity of endometrium to estrogen (17). When brushing, we only tend to sample the superficial cells or glands of polyps, which makes the sample look like endometrial hyperplasia, and ignores its real structure. Reagan and Ng et al. Study pointed out that when sampled cells were out of the endometrial cycle, only a quarter of them were from polyp (5). Thus, histology often shows more sensitive results than cytology for the diagnosis of polyps.

Our study found that Li Brush will be able to be a reliable approach for screening endometrial cancer and may provide great benefits for the social economy and women's health. However, there are still some shortcomings to the technique, such as the false-negative rate for the diagnosis of endometrial polyps is high and the sampling satisfaction rates are not low enough. In the future, after obtaining more data, we hope to use the brush to diagnose the detailed pathologic types of endometrium.

#### REFERENCES


#### AUTHOR CONTRIBUTIONS

LH, JD, and CS performed the clinic experiments. LZ, SL, and QL designed the endometrial sampler. HH and GZ performed the pathologic diagnosis. YL performed the pathological sections. HH also performed the pathological sections. QiW, QinW, and XT helped to perform the statistical analysis. LH and QL wrote the manuscript. QU revised the English grammar. QS helped to designed the experiments. QL conceived the study. All authors gave feedback and approved the final version of the manuscript.

#### FUNDING

This study was supported by the Key Research and Development Project of Shaanxi Provincial Science and Technology Department (2017ZDXM-SF-068), Shaanxi Provincial Collaborative Technology Innovation Plan (2017XT-026, 2018XT-002), and Medical Research Project of Xi'an Social Development Guidance Plan (2017117SF/YX011-3).

in uterine endometrioid carcinoma. Eur J Surg Oncol. (2016) 42:1367–71. doi: 10.1016/j.ejso.2016.03.009


diagnosis, treatment and follow-up. Radiother Oncol. (2015) 117:559–81. doi: 10.1016/j.radonc.2015.11.013


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Han, Du, Zhao, Sun, Wang, Tuo, Hou, Liu, Wang, Ulain, Lv, Zhang, Song and Li. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# A Novel Approach to Preoperative Risk Stratification in Endometrial Cancer: The Added Value of Immunohistochemical Markers

Vit Weinberger <sup>1</sup> , Marketa Bednarikova<sup>2</sup> , Jitka Hausnerova<sup>3</sup> , Petra Ovesna<sup>4</sup> , Petra Vinklerova<sup>1</sup> , Lubos Minar <sup>1</sup> , Michal Felsinger <sup>1</sup> , Eva Jandakova<sup>3</sup> , Marta Cihalova<sup>3</sup> and Michal Zikan<sup>5</sup> \*

*<sup>1</sup> Department of Gynecology and Obstetrics, University Hospital in Brno and Masaryk University, Brno, Czechia, <sup>2</sup> Department of Internal Medicine – Hematology and Oncology, University Hospital in Brno and Masaryk University, Brno, Czechia, <sup>3</sup> Department of Pathology, University Hospital in Brno and Masaryk University, Brno, Czechia, <sup>4</sup> Faculty of Medicine, Institute of Biostatistics and Analyses, Masaryk University, Brno, Czechia, <sup>5</sup> Institute of Biostatistics and Analyses, Faculty of Medicine, Masaryk University, Brno, Czechia*

#### Edited by:

*Massimo Nabissi, University of Camerino, Italy*

#### Reviewed by:

*Kruti P. Maniar, Northwestern University, United States Tonya J. Webb, University of Maryland, Baltimore, United States*

> \*Correspondence: *Michal Zikan michal.zikan@lf1.cuni.cz*

#### Specialty section:

*This article was submitted to Women's Cancer, a section of the journal Frontiers in Oncology*

Received: *07 December 2018* Accepted: *22 March 2019* Published: *12 April 2019*

#### Citation:

*Weinberger V, Bednarikova M, Hausnerova J, Ovesna P, Vinklerova P, Minar L, Felsinger M, Jandakova E, Cihalova M and Zikan M (2019) A Novel Approach to Preoperative Risk Stratification in Endometrial Cancer: The Added Value of Immunohistochemical Markers. Front. Oncol. 9:265. doi: 10.3389/fonc.2019.00265* Background: The current model used to preoperatively stratify endometrial cancer (EC) patients into low- and high-risk groups is based on histotype, grade, and imaging method and is not optimal. Our study aims to prove whether a new model incorporating immunohistochemical markers, L1CAM, ER, PR, p53, obtained from preoperative biopsy could help refine stratification and thus the choice of adequate surgical extent and appropriate adjuvant treatment.

Materials and Methods: The following data were prospectively collected from patients operated for EC from January 2016 through August 2018: age, pre- and post-operative histology, grade, lymphovascular space invasion, L1CAM, ER, PR, p53, imaging parameters obtained from ultrasound, CT chest/abdomen, final FIGO stage, and current decision model (based on histology, grade, imaging method).

Results: In total, 132 patients were enrolled. The current model revealed 48% sensitivity and 89% specificity for high-risk group determination. In myometrial invasion >50%, lower levels of ER (*p* = 0.024), PR (0.048), and higher levels of L1CAM (*p* = 0.001) were observed; in cervical involvement a higher expression of L1CAM (*p* = 0.001), lower PR (*p* = 0.014); in tumors with positive LVSI, higher L1CAM (*p* = 0.014); in cases with positive LN, lower expression of ER/PR (*p* < 0.001), higher L1CAM (*p* = 0.002) and frequent mutation of p53 (*p* = 0.008).

Cut-offs for determination of high-risk tumors were established: ER <78% (*p* = 0.001), PR <88% (*p* = 0.008), and L1CAM ≥4% (*p* < 0.001). The positive predictive values (PPV) for ER, PR, and L1CAM were 87% (60.8–96.5%), 63% (52.1–72.8%), 83% (70.5–90.8%); the negative predictive values (NPV) for each marker were as follows: 59% (54.5–63.4%), 65% (55.6–74.0%), and 77% (67.3–84.2%). Mutation of p53 revealed PPV 94% (67.4–99.1%) and NPV 61% (56.1–66.3%). When immunohistochemical markers were included into the current diagnostic model, sensitivity improved (48.4 vs. 75.8%, *p* < 0.001). PPV was similar for both methods, while NPV (i.e., the probability of extremely low risk in negative test cases) was improved (66 vs. 78.9%, *p* < 0.001). Conclusion: We proved superiority of new proposed model using immunohistochemical markers over standard clinical practice and that new proposed model increases accuracy of prognosis prediction. We propose wider implementation and validation of the proposed model.

Keywords: endometrial cancer, ER, imaging method, L1CAM, PR, preoperative biopsy, p53, risk stratification

#### INTRODUCTION

Endometrial carcinoma (EC) is one of the most common female cancers. It predominantly has a favorable prognosis, due to the early onset of signs and symptoms such as postmenopausal bleeding or spotting, which lead to early-stage diagnosis in most patients and five-year overall survival rates of up to 85% (1). However, 20% of those EC patients who are estimated to be at low risk of recurrence will nevertheless recur while up to 50% of those designated "high-risk" will not (2, 3). It is clear the prognostic markers currently used (FIGO stage, tumor subtype, and histological grade) are far from optimal in terms of preoperative stratification of patients into low- or high-risk groups regarding surgical planning and adjuvant treatment.

One of the currently used prognostic markers is FIGO stage. This is obligatory and determined by transvaginal ultrasound of the pelvis (US). Computed tomography (CT) of the chest and abdomen is an imaging method of choice and is routinely used to exclude retroperitoneal lymphadenopathy and metastases in parenchymal organs. The other prognostic markers, histotype, and grade of tumor differentiation, are assessed from a biopsy obtained either by dilatation and curettage of the uterus or by hysteroscopy. Based on the established FIGO stage, histotype, and tumor grade, patients are divided into two groups regarding the recurrence risk. Low-risk patients are treated with surgery alone, consisting of hysterectomy and bilateral salpingo-oophorectomy, while high-risk patients undergo more aggressive surgical treatment, including pelvic (PLN) and/or paraaortic lymphadenectomy (PALN) with or without adjuvant radiotherapy or chemotherapy. An aggressive therapeutic approach is associated with significantly higher side effects, such as increased blood loss, risk of thrombosis, infection, lymphoceles, lymphatic ascites, and lymphedema (4, 5).

The discovery of new histotype-specific and prognostic biomarkers for better stratification into high- or low-risk EC seems to be urgently needed in order to avoid over- or undertreatment of EC patients. The results of studies on potential new biomarkers assessed immunohistochemically (IHC), related to EC patient prognosis, were recently published. L1 cell adhesion molecule (L1CAM) overexpression and the loss of estrogen receptors (ER) and/or progesterone receptors (PR) are associated with poor prognosis and a high risk of relapse and death (6– 9). Mutations of the tumor protein p53 are associated with L1CAM expression, but not universally (10). However, to our best knowledge, neither the significance of L1CAM, ER, and PR expression nor knowledge of their relevant cut-offs together with determination of p53 mutation status in preoperative biopsies for pretreatment stratification into low- or high-risk have been established yet. No IHC biomarkers from preoperative biopsy are currently routinely used in the decision-making process for EC management.

The purpose of this study was to evaluate the clinical usefulness and added value of preoperatively assessed IHC biomarkers L1CAM, ER, PR, and p53 in differentiation between low- and high-risk EC patients through comparison of the current clinical practice model with a proposed model that includes immunohistochemical markers. The secondary objective of our study was to evaluate the correlation of IHC biomarkers with specific clinical (according to preoperative ultrasound and CT chest/abdomen) and pathological parameters.

#### PATIENTS AND METHODS

#### Patients

Patients undergoing surgical treatment for histologically proven or suspicious EC in the oncogynecological center of University Hospital Brno, Czech Republic, from January 2016 to August 2018 were consecutively included. The study was approved by the Institutional Ethical Board as was a version of written informed consent regarding tissue and clinical data use for scientific purposes obtained from each eligible patient.

#### Preoperative Imaging

All patients underwent a clinical examination, preoperative ultrasound staging examination, and CT of the chest/abdomen according to the local guidelines (11, 12). Each patient underwent both a transabdominal and a transvaginal US scan within 14 days before a board discussion led by one of the two oncogynecologists experienced in the field of US diagnostics in gynecologic oncology. Each US examination was immediately described in a written report; these reports were used for study analysis. Descriptions and examination reports were based on the standards applied by our center (13). During US staging examination of the uterine cavity, myometrium and cervix and pelvic lymph nodes were carefully assessed in every patient to describe the local extent of the tumor (14, 15).

Each patient underwent a CT scan of the chest, abdomen, and pelvis within 14 days before board discussion and admission to the operating theater. CT was performed with oral and intravenous contrast in order to exclude bowel wall implants, parenchymatous metastasis, and pathological lymphadenopathy. When lymph nodes measured >1 cm in the shorter axis or morphological changes as a rounded shape or necrosis were observed, tumor involvement was marked as suspicious.

# Risk Stratification and Clinical Management

The extent of the surgery was determined by the multidisciplinary board after dividing patients into the low- or high-risk group based on clinical staging and the preoperative histopathological examination and determination of the histotype and grading. The low-risk group was defined as endometrioid or mucinous carcinoma TNM stage cT1a or cT1b, grade 1 and/or endometrioid or mucinous carcinoma TNM stage cT1a, grade 2, all without clinical or imaging evidence of lymphadenopathy (cN0) or distant metastases (cM0). Patients were defined as high-risk unless these low-risk criteria were met. Type A radical hysterectomy with bilateral salpingoophorectomy was performed in all patients (16). Systematic pelvic and paraaortic lymphadenectomy was performed in the high-risk group only; in high-grade serous uterine cancer cases, total omentectomy and appendectomy were added to the staging procedure. The definitive histopathological examination was provided by one of three pathologists with experience in gynecological malignancies and contained data about stage, histotype, and grade, lymphovascular space involvement (LVSI), and measures of the IHC expression of markers L1CAM, ER, PR, and p53. Based on final histopathological findings, the patients were once again stratified into low- or high-risk groups based on the same preoperative criteria (i.e., irrespective of known IHC status of ER, PR, L1CAM, and p53) and, thereafter, decisions regarding adjuvant treatment and follow-up were made by the multidisciplinary board.

# Clinical Data

Age, results of US and CT scan with respect to depth of myometrial invasion, cervical involvement, lymphadenopathy, parenchymal organ involvement, and pathological data from biopsies (histotype, grading, IHC status of L1CAM, ER, PR, p53) were recorded.

# Tissue and Immunohistochemistry Analysis

All hematoxylin and eosin-stained slides were read by one of three experienced gynecological histopathologist to confirm histological subtype, grade, and (definitive excision specimen) stage and the presence or absence of LVSI. The evaluator was blinded to patient characteristics. All specimens were assessed according to the WHO Classification of Tumors of Female Reproductive Organs, 2014 (17). No additional later review of the slides was performed for the purpose of this study because it would not copy our real clinical practice. Immunohistochemical staining was performed on formalin-fixed and paraffin-embedded (FFPE) tissue sections. Immunohistochemistry for ER (clone SP1, product no. RBK 018-05, Zytomed, dilution 1:300), PR (clone 16, product no. NCL-L-PGR-312, Novocastra, dilution 1:80), L1CAM/CD171 (clone 14.10, product no. 826701, BioLegend, dilution 1:100), and p53 (clone DO-7, product no. M7001, DAKO, dilution 1:300) were performed using an automatic immunostainer (BenchMark Ultra, Ventana Medical Systems, Tucson, AZ, USA) according to the manufacturer's instructions. For ER, PR, and p53, only nuclear staining was scored as positive. Positivity of L1CAM was defined as distinct membrane staining. For ER, PR, and L1CAM, the percentage of positive tumor cells was assessed. p53 was classified into wild type or mutant (excessive = strong diffuse overexpression in more than 90% of tumor cells or completely negative) phenotypes. Representative microphotographs of the expression of estrogen receptor (ER), progesterone receptor (PR), L1CAM and p53 in serous (high risk) and grade 1 endometrioid (low risk) carcinoma are shown in **Figures 1A–H**.

#### Statistical Analysis

Categorical data were summarized using absolute and relative frequencies and compared by Fisher's exact test. Continuous variables were summarized as median with 10 and 90th percentile and tested by the Mann-Whitney U-test.

A model for the best classification of final risk was built using the CHAID growing method with crossover validation. Misclassification cost for wrongly determining high-risk patients as low-risk was set twice higher because of the preference for the correct high-risk group EC patient determination.

The success of risk-group classification was evaluated using four standard measures: (i) sensitivity is the ability of the test to correctly identify those with an occurrence of the assessed marker (true positive rate), whereas (ii) specificity is the ability of the test to correctly identify those without an occurrence of the assessed marker (true negative rate), (iii) positive predictive value (PPV) is the probability that the marker is present when the test is positive, whereas (iv) negative predictive value (NPV) is the probability that the marker is not present when the test is negative. All these statistics were accompanied by 95% confidence intervals (CI).

The comparison of sensitivities and specificities of the two binary diagnostic tests in a paired study design was performed using McNemar's test with continuity correction. Differences in (positive and negative) predictive values of two binary diagnostic tests were tested using a generalized score statistic proposed by Leisenring, Alonzo, and Pepe (18). All tests were performed as two-sided at the significance level 0.05. Analyses were done in IBM SPSS Statistics and R.

# RESULTS

# Clinical and Histopathological Characteristics

From January 2016 to August 2018, 132 patients underwent surgical treatment for EC in the oncogynecological center of University Hospital Brno, Czech Republic, and have been consecutively enrolled in the study. The median age was 66 years. According to ultrasound and CT staging, before operation 95 patients (72%) were evaluated as FIGO stage IA, while 25 (19%) were stage IB, 5 (4%) stage II, 3 (2%) stage III, and 4 (3%) were at an unknown stage. Preoperative biopsy was available for all 132 patients; 102 patients had endometrioid cancer, of whom 50 (49%) had endometrioid or mucinous carcinoma grade 1 (EG1), 45 (44%) grade 2 (EG2), 6 (6%) grade 3 (EG3), and one had a nondiagnostic grade. Seventeen (13%) patients were diagnosed with non-endometrioid carcinoma (NEC). Furthermore, there were

carcinoma with almost 100% cells positive; (F) nuclear positivity of PR expression in grade 1 endometrioid carcinoma with almost 100% cells positive; (G) complete negativity of L1CAM expression in grade 1 endometrioid carcinoma with 0% cells positive; (H) p53 wildtype immunohistochemical pattern in grade 1 endometrioid

eight cases (6%) of endometrial intraepithelial neoplasia (EIN) in preoperative biopsy (**Table 1**). Based on both histological and clinical findings, 94 (71%) patients were preoperatively classified as low-risk and 38 (29%) as high-risk by current model and, consequently, the recommendation for the extent of surgery was issued. In the high-risk group, PLN and PALN were performed for 26 (20%) patients, apart from hysterectomy and bilateral salpingo-oophorectomy. After the surgical procedure and definitive histopathological examination, 82 (62%) patients were at FIGO stage IA, 17 (13%) at FIGO stage IB, 19 (14%) FIGO stage II, 12 (9%) FIGO stage III and two (2%) with FIGO stage IV. Regarding endometrioid or mucinous carcinoma grade, 30 (28%) were at EG1, 69 (65 %) at EG2, 8 (7 %) EG3, 20 (15%) NEC, and 2 (2%) EIN. In contrast to the preoperative risk determination, the final post-operative stratification in risk groups was as follows: 70 (53%) low-risk patients and 62 (47%) high-risk patients (**Table 1**).

#### Immunohistochemical Characteristics

The expression of markers ER, PR, L1CAM, and p53 status were immunohistochemically evaluated from the specimen obtained both by diagnostic procedure and definitive surgery. The correlation of IHC markers between preoperative examination and definitive histopathological findings was statistically significant for all of the evaluated markers (p < 0.005). In the preoperative specimen, the expression was evaluable in 98 patients for ER and PR, 97 for L1CAM, and 98 for p53 mutational status; results are listed in **Table 2**.

carcinoma.



*values denote median (10-90th percentile) or n (%); p-values of chi-square or McNemar's test; G, grade; EC, endometrial cancer.*

#### Correlation of IHC Markers With Disease Extent (FIGO Staging)

The correlation was assessed between IHC markers in preoperative tissue samples and the final histopathological findings (e.g., myometrial invasion, cervical, and lymph node involvement). Moreover, the correlation with LVSI was evaluated because LVSI is one of the important markers for adjuvant treatment strategy decisions. There were statistically significant lower levels of ER (p = 0.024) and PR (0.048) and higher levels of L1CAM (p = 0.001) in tumors with myometrial invasion >50%. In tumors with cervical involvement, a significantly higher expression of L1CAM was observed (p = 0.001), while differences among levels of ER (p = 0.236) and PR (p = 0.108) did not reach statistical significance. PR were significantly lower (p = 0.014) and L1CAM higher (p = 0.014) in tumors with positive LVSI. In patients with positive LN, levels of ER and PR were lower (p = 0.001 and p < 0.001, respectively); on the other hand, levels of L1CAM were higher (p = 0.002) and the mutation of p53 more frequent (p = 0.008), see **Table 3**.

#### The Precision of EC Risk Stratification Based on Markers Currently Used in Clinical Praxis

Concerning depth of myometrial invasion, we classified all the patients in whom invasion reached ≥50% as high-risk (n = 36) and the patients with invasion <50% as low-risk (n = 96). Our approach to preoperative risk stratification of EC patients revealed a sensitivity of 48% and specificity of 89% in terms of high-risk group determination. Taking all current standard prognostic markers together, it can be concluded that, whereas low-risk EC patients are preoperatively classified with relatively high accuracy (62/70, 82%), the determination of high risk is far from optimal, since more than half of EC patients with actual high-risk disease were established as low-risk. See **Table 4**.

#### The Accuracy of EC Risk Stratification by Using IHC Markers

IHC markers were assessed in a preoperative tumor sample, and optimal cut-offs for continuous markers (obtained preoperatively) were designed using ROC analyses. At a cut-off for ER <78% (p = 0.001), PR <88% (p = 0.008), and L1CAM ≥4% (p < 0.001), high-risk tumors were determined with a sensitivity of 28% for ER (15.6–42.6%), 62% for PR (46.4–75.5%), and 72% for L1CAM (57.4–84.4%). Specificity was 96% (86.5–99.5%), 68% (52.1–79.2%), and 86% (73.3– 94.2%), respectively. The PPV was 87% for ER (60.8–96.5%), 63% for PR (52.1–72.8%), and 83% for L1CAM (70.5–90.8%); the NPV for each marker were as follows: 59% (54.5–63.4%), 65% (55.6–74.0%), and 77% (67.3–84.2%), respectively. The sensitivity of p53 mutated status (p < 0.001) for high-risk detection was low (34%), but the specificity was high (98.0%, CI 88.7–99.9%), which represents PPV 94% (67.4–99.1%) and NPV 61% (56.1–66.3%), respectively. If p53 mutated, there was a high probability the patient fit into the high-risk group (15 out of 16 patients in our series). As far as accuracy of high-risk determination in the largest number of patients was concerned, the marker L1CAM seemed to be the most robust: 34 from 41 patients who had L1CAM values ≥4% were classified as high-risk (**Table 5**, **Figure 2**).

#### The Added Value of IHC Markers for Improvement of EC Risk Stratification

To evaluate whether IHC markers would contribute to more accurate stratification of EC patients into risk groups, a model consisting of parameters obtained by imaging methods (myometrial invasion, cervical involvement, lymph node involvement), histology (endometrioid or mucinous vs. nonendometrioid), grade and IHC markers (ER, PR, L1CAM, p53) was introduced. According to risk stratification, the following parameters have been shown as statistically significant and crucial for the model: L1CAM, PR, and myometrial invasion.


*Values denote mean (SD) and median (10–90th percentile) or n (%); means and SD are shown only for exploratory purpose since data are not normally distributed; IHC, immunohistochemical markers; ER, estrogen receptors; PR, progesterone receptors; L1CAM, L1cell adhesion molecule; mut, mutated; wt, wild type; EG1, endometrioid or mucinous cancer, grade 1; EG2, endometrioid or mucinous cancer, grade 2; EG3, endometrioid or mucinous cancer, grade 3; EIN, endometrioid intraepithelial neoplasia; NEC, non-endometrioid cancer.*

TABLE 3 | Correlation of IHC markers from preoperative biopsy with staging after surgery.


*Values denote mean (SD) and median (10-90*th *percentile) or n (%), p-value of Mann-Whitney U-test or Fisher's exact test; means and SD are shown only for exploratory purpose since data are not normally distributed; LVSI, lymphovascular space involvemen; mut, mutated; wt, wild type.*

The procedure of classification is shown in **Figure 3**. The overall EC risk stratification success was 78% for this model. Successful group inclusion was observed for 80% of the low-risk patients (56/70), and for 76% of high-risk patients (47/62 patients).

Positive predictive values were similar for both methods, while negative predictive value (i.e., the probability of extremely low risk in negative test cases) was significantly improved (66 vs. 78.9%, p < 0.001) (**Table 6**, **Figure 4**).

#### New Model in Comparison to Current Practice

Immunohistochemical markers included in the current diagnostic practice would significantly improve sensitivity (48.4 vs. 75.8%, p < 0.001) associated with a slightly, statistically non-significant decrease in specificity (to 80%, p = 0.238).

#### DISCUSSION

The determination of an appropriate surgery and its adequate extent is a crucial part of treatment in newly diagnosed EC patients and significantly differs between the high- and low-risk groups. Existing models determining patient risk are based on the

FIGURE 3 | Patients with L1CAM positivity >4% were identified as high risk; in fact 86% of them were high-risk indeed. Patients with L1CAM value from 1 to 4% were further divided according to myometrial invasion. In case of myometrial invasion <50% they were stratified as low-risk, on the contrary in case of myometrial invasion >50% or unknown they were classified as high-risk. Patients with L1CAM <1% were further disaggregated by PR value. In case of PR >85% they were identified as low-risk (with success rate of 100%), on the other hand in case of PR ≤85% they were stratified as high-risk according to clinical preference for more precise high-risk group determination even with the expectation of higher false positivity.

#### TABLE 4 | Accuracy of low-/high-risk group classification according to current practice.


*CI, confidence interval; PPV, positive predictive value; NPV, negative predictive value. Bold values are significant.*

#### TABLE 5 | Correlation of IHC with final EC risk stratification.


*p-value of Fisher's exact test; CI, confidence interval; PPV, positive predictive value; NPV, negative predictive value; mut, mutated; wt, wild type. Bold values are significant.*

synthesis of information obtained from the result of preoperative biopsy (histotype, grading) and imaging methods. Based on the results, patients are included in a risk group before the surgery and so only hysterectomy and salpingo-oophorectomy are indicated, or the procedure is extended by pelvic and paraaortic lymphadenectomy. Proper preoperative inclusion of a



*IMG, imaging method; IHC, immunohistochemical markers; current model, histology, grading, imaging method. Bold values are significant.*

patient in the risk group is crucial for her treatment and overall survival and is a clinically crucial question.

Other approaches to assess the biological behavior of endometrial cancers are under development. Several research groups have defined immunohistochemical and/or mutation profiles to allow distinguishing endometrial cancer subtypes. The Cancer Genome Atlas (TCGA) project provided the most comprehensive molecular study on endometrial cancer so far. They identified four group with distinct molecular changes that correlate with progression free survival – POLE (Polymerase Epsilon subunit) ultramutated, MSI (microsatellite instability) hypermutated, copy-number low, and copy-number high (19). This approach allows objective categorization of endometrial cancers, however, methodologically remains costly, complex and unsuitable for wider clinical application.

Others introduced a concept of sentinel lymph node detection in endometrial cancer patients. The large prospective study led by Rossi (20) showed very high sensitivity (97.2%) and low false negativity rate (3%) for sentinel lymph node (SLN) detection. SLN detection concept is based on low risk of paraaortic lymph nodes involvement in patients with negative pelvic lymph nodes (21). However, controversy regarding sentinel lymph node detection in high-risk disease and management of low volume nodal disease on ultrastaging still remains.

In many centers, frozen section of uterus is still standard-ofcare in terms to confirm or to more specify type and grade of the tumor. Accuracy of frozen section histopathological evaluation is, however, comparable to imaging methods and interobserver agreement regarding both the categories, type and grade, is poor (22–26).

The current development of risk prediction model is mainly focused on combination of imaging and molecular predictors, as our study does. In 2014, Van Holsbeke et al. published a study that externally validated two mathematical models of preoperative risk group prediction in a particular patient (27). The models were based on histology, grading, and the preoperative sonographic evaluation of tumor invasion into the myometrium and cervix. Both models achieved sensitivity of 78–83% and specificity of 68–72% in the detection of high-risk EC patients. In our study, the current clinical model reliably determined low-risk patients (correctly in 83% of cases), while only 8 (11%) patients in the study were false positives included in the high-risk group. The current model preoperatively stratified patients to high-risk with sensitivity of only 48% (35.5–61.4%) and specificity 89% (78.7–94.9%), NPV 66% (60.0–71.4%), and PPV 79% (65.0–88.3%) (**Table 4**).

A number of ultrasound studies have been published for the assessment of individual staging parameters to determine the depth of tumor invasion into the myometrium, with ultrasound sensitivity from 61 to 93% and specificity from 71 to 92%, when performed by an expert sonography specialist (28–32). The sensitivity reported in the evaluation of tumor invasion in cervical stroma was lower, from 25 to 93%, and specificity from 85 to 99% (30, 33, 34). In a study utilizing expert sonography in a specialized center, Fruhauf et al. reported a PPV of 67.6% and NPV of 83.3% for the detection of deep myometrial invasion and PPV of 60.0% and NPV of 88.1% in the detection of tumor affection of the uterine cervix. According to a recent metaanalysis of 18 studies, CT sensitivity is 47% and specificity 93%; as for ultrasound, sensitivity is 55% and specificity up to 85% for the detection of malignant lymphadenopathy (35). In our group of 132 female patients, the invasion of the tumor to half the thickness of the myometrium was determined correctly in 90%; a false-negative result in the high-risk group of patients was reported in 39 cases. CT did not detect pathological lymphadenopathy in six cases out of nine. A total of nine patients were classified as false positives on the basis of US and CT as validated by the definitive histology.

Studies showing the discrepancy between histology obtained from preoperative curettage or hysteroscopy and definitive histological findings have been published (36–38). On the other hand, there are studies showing good concordance between histology, grade, and immunohistochemical staining in curettage and hysterectomy samples (39, 40). We confirmed that the preoperatively determined histological type, grade, and immunohistochemical biomarkers L1CAM, ER, PR, p53 correlated with the final preparation.

In our study, 70 (70/132) patients were classified as low-risk and 62 (62/132) as high-risk. According to the pre-operative staging, PLN and PALN were performed in 26 patients (20%) in our cohort. However, if the definitive risk were known, staging lymphadenectomy would be performed in all 62 patients in the high-risk group (47%). Due to an inappropriate staging surgery, patients underwent repeated surgery or adjuvant radiotherapy, which may have been avoided if a complete surgical staging with negative histological findings of the presence of the tumor in the lymph nodes had been performed. On the contrary, there are onco-gynecological centers which report extensive PLN + PALN in EC patients, thereby increasing post-operative morbidity without an oncology safety increase (41). We focused on currently promising prognostic IHC markers ER, PR, L1CAM, and p53 mutation to determine whether these markers can help to refine the preoperative stratification of patients into high- and low-risk categories to assist the gynecological oncology surgeon selecting the adequate surgical extent.

A study published by van der Putten et al. revealed that L1CAM expression in curettage specimens is associated with features of aggressive endometrial cancer disease and poor survival of EC patients (42). van der Putten et al. (42) stated that L1CAM, ER, PR were associated with advanced stage, high-grade, non-endometrioid histology, lymphovascular space invasion (LVSI), and reduced disease-free survival (42). Trovik et al. (43) reported that combined ER/PR loss is a significant predictor of nodal affection and overall poor prognosis of patients. ER and PR are prospectively investigated in an ongoing study where the decision of whether to perform or not to perform lymphadenectomy is based on the preoperational condition of hormone receptors (44). Prospective studies PIPENDO and PORTEC 4 are currently underway. The first study examines the use of molecular risk markers to identify high-risk patients requiring extensive surgery and/or adjuvant therapy (8). PORTEC 4 uses molecular risk factors for the stratification and indication of adjuvant radiotherapy (45).

In our study, ER, PR, L1CAM and p53 values from preoperative histology were related to definitive histology and grading. In endometroid carcinoma, there was a greater percentage of ER, PR receptors and no or ultimately low percentages of L1CAM mutations. The opposite ratio was seen in the occurrence of markers in non-endometroid ECs; p53 was mutated dominantly in endometrial grade 3 and nonendometroid carcinoma. The correlation of IHC markers with the extent of disease shows a decrease in ER and PR expression in higher stages of the disease. Furthermore, an increase in L1CAM expression can be observed when compared with early stages. Similarly, the p53 mutation was more common. Our results are in line with the published data in larger patient cohorts (43, 46). The correlation of markers with the presence of distant metastases could not be assessed as there were only two patients with distant metastases at the time of diagnosis in our study group.

To our best and honest knowledge, this is the first study to evaluate the added value of L1CAM, ER, PR and p53 markers in low- and high-risk EC preoperative diagnostics. This is the first study attempting to determine the cut-off of the individual markers for this classification.

We focused on the correlation of IHC markers with the determination of high-risk EC and the assessment of optimal cut-offs for continuous markers using ROC analyses. At the cut-off for ER <78% (p = 0.001), PR <88% (p = 0.008), and L1CAM ≥4% (p < 0.001), high-risk tumors were determined with a sensitivity of 28, 62, and 72%, respectively, and with respective specificity of 96, 72, and 86%. The PPV for ER, PR, and L1CAM were 87, 63, 83; the NPV for each marker were as follows: 59%, 65%, and 77%. The sensitivity of p53 mutated status (p < 0.001) for high-risk detection was low (34%), but the specificity was high (98%), which represents PPV 94% and NPV 61%, respectively.

A cut-off of 10% for positive L1CAM staining has been reported (6, 40, 47). van Gool et al. reported that when using a cut-off of 10% for positive staining, tumors in the study were classified as L1CAM-positive, with no significant association between L1CAM positivity and the rate of distant metastasis (p = 0.195). However, increasing the threshold for L1CAM positivity to 50% resulted in a reduction of the frequency of L1CAM-positive tumors and a significant association with the rate of distant metastasis (p = 0.018) (10). Estrogen receptors and PR are considered lost when expression is seen in <10% of the tumor cells. This cut-off used with breast cancer management in the prediction of hormone resistance was also evaluated in EC for prognosis prediction and published (42, 43, 48). In our cohort, we determined optimal cut-offs to distinguish low- and high-risk EC for ER <78% (p = 0.001), PR <88% (p = 0.008), and L1CAM ≥4% (p < 0.001). These cut-offs were established in a prospectively assessed cohort of consecutively included patients. In contrast to the cut-offs we defined, the published values are determined in retrospective cohorts of women with recurrence during follow-up, or with an adverse course of their disease with metastatic spread in parenchymatous organs and retroperitoneal lymph nodes. The incidence of L1CAM positivity and loss of ER, PR in these patients in retrospective cohorts may be significantly higher as it is an already pre-selected group of patients. The explanation may be based on the fact that patients who do not exceed the published cut-off values (10% for L1CAM, ER, and PR or 50% for L1CAM) but exceed the cut-offs set for the high-risk group in our study cannot be traced back in the retrospective studies, as they had not been radically treated with combined surgical and ± adjuvant therapy (radiotherapy, chemotherapy) and there was no recurrence during the long follow-up. Our results were confronted with 10 and 50% cut-offs for L1CAM; this setting led to good differentiation (high specificity) but at a very low sensitivity and good specificity.

To evaluate whether IHC markers would contribute to more accurate stratification of EC risk stratification, a new model was established. Parameters were obtained by imaging methods (myometrial invasion, cervical involvement, lymph node involvement), histology (endometrioid or mucinous vs. non-endometrioid), grade and IHC markers (ER, PR, L1CAM, p53). As deciding for risk stratification, the following parameters have been shown as crucial: L1CAM, PR, and myometrial invasion. The procedure of classification is shown in **Figure 3**. EC risk stratification's overall success was 78% for this model. Successful inclusion into a low-risk group was observed in 80% (56/70 patients), while for high-risk it was 76% (47/62 patients). We provided the comparison of current procedure represented by a model based on histotype, grading, and imaging methods with a new model consisting of imaging examination along with markers (IMG+IHC) (**Table 6**). IHC markers included in the current diagnostic would significantly improve sensitivity (48.4 vs. 75.8%, p < 0.001) associated with a slightly, statistically non-significant decrease in specificity to 80% (p = 0.238). Positive predictive value was similar for both methods, while negative predictive value (i.e., the probability of being true negative if the test is negative) was significantly improved (66 vs. 78.9%, p < 0.001), (**Figure 4**). Using our model, a significantly higher proportion of patients would be properly determined as high-risk.

When comparing the accuracy of the parameters used in the old model with the definite histology, we find discrepancies, especially in the grade and cervical invasion category (**Table 1**). At the present level of knowledge, no significant improvement in preoperative diagnostic accuracy can be expected by, for example, using imaging methods. Therefore, we are introducing a new model using molecular markers that are not dependent on imaging or other methods of clinical examination.

The strength of our study is that it is a cohort from a real clinical practice with prospective data collection and complete knowledge of preoperative and postoperative data. This is the first study to deal with the real implementation of new IHC markers in the pre-operational decision model. Our study design represents daily routine practice.

We acknowledge the study also has weaknesses. This is a relatively small cohort of EC patients, where all stages are not adequately represented; only two female patients in the FIGO IV stage were present. Considering the excellent correlation between preoperative and postoperative histology and grading, the weakness of the model is in its imaging method, which in the case of ultrasound is dependent on the expert skills of a particular sonographer or imaging specialist.

#### CONCLUSION

We have demonstrated in our cohort that incorporating IHC markers into preoperative practice in endometrial cancer patients increases prognosis prediction accuracy and allows for the development of a new model for more accurate patient clinical management. Should we prefer a higher specificity model, then the most accurate classification is based on L1CAM

#### REFERENCES


values, myometrial invasion, and the condition of PR receptors. However, for wider implementation and the validation of proposed model, additional study is needed. Ideally, a prospective randomized trial would evaluate the role of IHC markers L1CAM, ER, PR, and p53 in a preoperative setting together with imaging method and histology/grade. To further improve the new model, it would be interesting to focus on general weaknesses in the accuracy of preoperative imaging methods and in the quality of preoperatively obtained samples with a full IHC examination on a routine daily basis. Incorporating IHC markers seems to be the best way to treat EC patients more accurately.

#### ETHICS STATEMENT

The text of informed consent, information for patients and the study protocol was approved by Institutional Ethical Board of University Hospital Brno.

#### AUTHOR CONTRIBUTIONS

VW and MZ: idea of research project, overseeing, coordination of analyses; MB, PV, LM, and MF: collection and preparation of clinical data; JH, EJ, and MC: immunohistochemistry and interpretation of pathological data; PO: statistics; VW, MB, JH, PO, PV, and MZ: writing manuscript. All authors: final reading and approval of manuscript.

#### FUNDING

This work was supported by the Ministry of Health, Czech Republic—conceptual development of a research organization (FNBr 65269705) and grant project No. 17-32030A.

collaboration study. Br J Cancer. (2016) 115:716–24. doi: 10.1038/bjc. 2016.235


endometrial cancer patients. Br J Cancer. (2017) 117:840–7. doi: 10.1038/bjc. 2017.235


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Weinberger, Bednarikova, Hausnerova, Ovesna, Vinklerova, Minar, Felsinger, Jandakova, Cihalova and Zikan. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Integrated Bioinformatic Analysis of a Competing Endogenous RNA Network Reveals a Prognostic Signature in Endometrial Cancer

Leilei Xia1†, Ye Wang2,3†, Qi Meng1†, Xiaoling Su<sup>4</sup> , Jizi Shen<sup>1</sup> , Jing Wang<sup>1</sup> , Haiwei He<sup>1</sup> , Biwei Wen<sup>1</sup> , Caihong Zhang<sup>1</sup> and Mingjuan Xu<sup>1</sup> \*

 *Department of Obstetrics and Gynecology, Changhai Hospital, Second Military Medical University, Shanghai, China, Department of Cell Biology, Center for Stem Cell and Medicine, Second Military Medical University, Shanghai, China, Department of Urology, Changhai Hospital, Second Military Medical University, Shanghai, China, <sup>4</sup> Department of Obstetrics and Gynecology, No. 455 Hospital, Second Military Medical University, Shanghai, China*

#### Edited by:

*Massimo Nabissi, University of Camerino, Italy*

Reviewed by: *Changqing Su, Eastern Hepatobiliary Surgery Hospital, China Xiaofan Lu, China Pharmaceutical University, China Qin Yang, Shanghai Jiaotong University, China*

> \*Correspondence: *Mingjuan Xu 13636373419@163.com*

*†These authors have contributed equally to this work*

#### Specialty section:

*This article was submitted to Women's Cancer, a section of the journal Frontiers in Oncology*

Received: *21 September 2018* Accepted: *13 May 2019* Published: *29 May 2019*

#### Citation:

*Xia L, Wang Y, Meng Q, Su X, Shen J, Wang J, He H, Wen B, Zhang C and Xu M (2019) Integrated Bioinformatic Analysis of a Competing Endogenous RNA Network Reveals a Prognostic Signature in Endometrial Cancer. Front. Oncol. 9:448. doi: 10.3389/fonc.2019.00448* In endometrial carcinoma, the clinical outcome directly correlates with the TNM stage, but the lack of sufficient information prevents accurate prediction. The molecular mechanism underlying the competing endogenous RNA (ceRNA) hypothesis has not been investigated in endometrial cancer. Multi-bioinformatic analyses, including differentially expressed gene analysis, ceRNA network construction, Cox regression analysis, function enrichment analysis, and protein-protein network analysis, were performed on the sequence data acquired from The Cancer Genome Atlas (TCGA) data bank. A ceRNA network comprising 366 mRNAs, 27 microRNAs (miRNAs), and 66 long non-coding RNAs (lncRNAs) was established. Survival analysis performed with the univariate Cox regression analysis revealed nine lncRNAs with prognostic power in endometrial carcinoma. In multivariate Cox regression analysis, a signature comprising LINC00491, LINC00483, ADARB2-AS1, and C8orf49 showed remarkable prognostic power. Risk score and neoplasm status, but not TNM stage, were independent prognostic factors of endometrial carcinoma. A ceRNA network comprising differentially expressed mRNAs, miRNAs, and lncRNAs may reveal the molecular events involved in the progression of endometrial carcinoma. In addition, the signature with prognostic value may discriminate patients with increased risk for poor outcome, which may allow physicians to take accurate decisions.

Keywords: endometrial cancer, ceRNA network, prognostic factor, lncRNA, TCGA

# INTRODUCTION

Endometrial cancer was shown to cause ∼11,350 deaths in the United States this year, and its incidence has increased mainly owing to the rise in obesity, a known risk factor. Clinical outcomes directly correlate with the clinical stages, and the 5-year overall survival rate has sharply decreased from 95% in patients with stage I cancer to 16% in those with stage IV cancer (1, 2). In clinical follow-up, aside from imageological examination such as B-mode ultrasound and magnetic resonance imaging (MRI), serum tumor markers, including CA125, CA199, and CEA, may serve as the indicators for the predictive outcome in patients (3, 4). Recently, L Salmena et al. proposed that mRNAs, long non-coding RNAs (lncRNAs), and pseudogenes may regulate the expression of each other by targeting microRNAs (miRNAs) (5). lncRNAs were the biggest and most diverse class of non-coding RNAs in the human genome (6). Plenty lncRNAs play a part in pathogenesis of cancer such as unmanageable proliferation, or metastasis (7, 8), and can serve as oncogenes or antioncogenes, or by interreacting with famous oncogenes or antioncogenes such as MYC or p53, on both a transcriptional or post-transcriptional level (9, 10). Although the molecular events involved in the progression of endometrial carcinoma have been well studied, the complicated interaction between mRNAs, miRNAs, and lncRNAs that exerts crucial influence on the progression and prognosis of endometrial cancer is yet unclear.

The Cancer Genome Atlas (TCGA), a public integrated database, provides multiplatform genomic data along with the clinical information of matched patients. This database has driven the development of genomics to characterize the molecular landscape of cancers (11). Using TCGA, we analyzed differentially expressed genes, including mRNA, miRNA, and lncRNA, and constructed a lncRNA-miRNA-mRNA competing endogenous (ceRNA) network in endometrial cancer. Furthermore, we used Cox regression analysis to identify a signature based on LINC00491, LINC00483, ADARB2-AS1, and C8orf49. Of note, this signature may serve as an independent prognostic factor in endometrial cancer. This study demonstrates that these lncRNAs would allow identification of patients with endometrial cancer that are at higher risk for poor clinical outcome.

#### METHODS AND MATERIALS

#### Data Source

All the foundation data of TCGA-UCEC project, including genetic data, transcriptome profiling, and clinical information, were acquired from the Genomic Data Commons of the National Cancer Institute (http://portal.gdc.cancer.gov). Among 587 endometrial cancer profiles, 35 were obtained from paracarcinoma tissues, while others were endometrial cancer tissues. These data were available with no restrictions for research, and this study was performed under the guidelines of TCGA. GENCODE v.27 was used to annotate RNAs in the original transcriptome profiling, and total of 19676 mRNAs, 14447 lncRNAs, 1881 miRNAs were annotated. In clinical information, overall survival data were calculated from the date of diagnosis to the date of death or last follow-up.

#### Differentially Expressed mRNAs, miRNAs, and lncRNAs

The genomic data and transcriptome data from TCGA were downloaded and subjected to normalization with the calcNormFactors function with method of trimmed mean of M-values (TMM) in edgeR package. In addition, to avoid low abundance impact on the next procedure, RNAs with an average value of <1 were excluded. The differentially expressed mRNAs, miRNAs, and lncRNAs were analyzed with the exactTest function using the edgeR package. RNAs with a cutoff false discovery rate (FDR) adjusted p < 0.01 and |logFC| ≥ 2 were considered statistically different between cancer and normal groups (12). Heatmap was plotted using pheatmap R package.

## Construction of the ceRNA Network

According to the hypothesis of ceRNA, it is vital to match the differentially expressed mRNAs, miRNAs, and lncRNAs; thus, the network could highlight a new molecular mechanism involved in the development of endometrial cancer. Pairs of miRNAlncRNA were establish using the miRcode database (13). Pairs of miRNA-mRNA were built using the basic data supplied by TargetScan (14), and the mRNA predicted by the database was characterized as the target mRNA and used in the subsequent step. Pearson correlation was calculated between lncRNAs and mRNAs mediated by miRNAs (15), only the pairs with coefficient >0.4 were considered may involved in ceRNA network (16). Then, to quantify the regulatory effect of lncRNA/mRNA over mRNA/lncRNA via a specific miRNA, Sumazin et al. proposed the use of conditional mutual information (17), which was calculated in JAMI software implemented in Java (18). In this analysis, pairs with a value of p < 0.05 was considered statistically significant.

The clusterProfiler R package created by Guang et al. (19) was used to perform functional enrichment analyses, including Gene Ontology and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses. Terms with a value of p < 0.05 was considered statistically significant.

#### Identification of a Prognostic Signature Based on the ceRNA Network

Prognostic data were created on the matrix of lncRNAs involved in the ceRNA network and matched follow-up data. Patients were classified according to the median expression of lncRNAs into high or low expression groups. Univariate Cox regression analysis was used to identify the lncRNA with prognostic value. In addition, lncRNA with a p < 0.05 was used in the multivariate Cox regression analysis. As the number of lncRNA was high, it is important to create a signature comprising a limited number of variables and the best Akaike information criterion (AIC). These steps in the multivariate Cox regression analysis used the function of coxph in survival R package. After the identification of the best signature that predicted the outcome of the patients with endometrial cancer, the risk score was calculated as the summation of the product of each gene and its coefficient. In addition, patients were classified into high and low risk groups with the cutoff of the median risk score. Log-rank test was used to compare the survival distribution of these two groups, as estimated by the Kaplan-Meier analysis. In addition, a receiver operating characteristic (ROC) analysis was used to estimate the predictive power of this signature using 3 years as the predicted time.

The relativity between risk score and clinical factors, including age at diagnosis, TNM stage, and neoplasm tumor status, was analyzed using the chi-square test. Both univariate and multivariate Cox regression analyses were employed to discriminate between prognostic factors in endometrial carcinoma.

## Protein-Protein Interaction Network Construction

mRNAs involved in ceRNA network based on lncRNA from the signature were subjected to protein-protein interaction network analysis using the STRING website (20).

#### Statistical Analysis

Kaplan-Meier curve was conducted by SPSS 21.0 using log-rank test (21), while other statistical tests were executed by R 3.5.1 using the corresponding R package mentioned above, hazard ratios was used in Cox model (22).

# RESULTS

### Differentially Expressed mRNAs, miRNAs, and lncRNAs

By employing differential gene expression analysis between cancer tissues and normal adjunct tissues, as per the cutoff FDR adjusted p < 0.01and |logFC| ≥ 2, 2,609 differentially expressed mRNAs (1,648 overexpressed and 961 down-regulated), 189 differentially expressed miRNAs (140 overexpressed and 49 down-regulated), and 1,121 differentially expressed lncRNAs (798 overexpressed and 323 down-regulated) were identified (**Supplement Figure 1**).

#### Construction of the ceRNA Network

We constructed the ceRNA network comprising mRNAs, miRNAs, and lncRNAs. The pairs of lncRNA-miRNA were matched using miRcode, 95 lncRNAs and 27 miRNAs formed 530 potential lncRNA-miRNA pairs. The miRNA-mRNA pairs were matched based on TargetScan. As a result, 1126 pairs of lncRNAsmRNAs have the Pearson correlation >0.4. Then, in conditional mutual information calculated in JAMI software, 1605 pairs of lncRNA-miRNA-mRNA have the p < 0.05 which means that in these pairs, miRNAs are mediating that interaction. Thus, the ceRNA network was completely constructed and constituted of mRNAs, miRNAs, and lncRNAs (**Supplement Table 1**).

It is well known that hub nodes play critical roles in biological networks. Therefore, we calculated all node degrees of the lncRNA involved in ceRNA network. According to the previously study by Han et al., in which they defined a hub as a node degree exceeding 5, we found that 15 lncRNAs could be chosen as hub nodes, and the results are shown in **Supplement Figure 2**.

To discover the biological terms associated with these dysregulated genes, GO and KEGG function enrichment analyses were separately performed on dysregulated mRNAs. Terms with a value of p < 0.05 were considered as statistically significant.

In GO analysis, the overexpressed mRNAs were mainly enriched in epidermis development, intermediate filament, and transcription factor activity, RNA polymerase II proximal

FIGURE 1 | Functional enrichment analysis of mRNAs in the ceRNA network. (A) Gene Ontology enrichment analysis of upregulated mRNAs. (B) Gene Ontology enrichment analysis of downregulated mRNAs. (C) KEGG pathway analysis of upregulated mRNAs. (D) KEGG pathway analysis of downregulated mRNAs. Horizontal axis represents gene count. Vertical axis represents enrichment analysis terms. Color of each plot represents the *p* value while the size represents the gene number in this term.

promoter sequence-specific DNA biding. The down-regulated mRNAs were significantly enriched in functions such as muscle system process, extracellular matrix, and transcription factor activity, RNA polymerase II proximal promoter sequencespecific DNA binding. In KEGG analysis, the overexpressed mRNAs were mainly enriched in Maturity onset diabetes of the young, and Alcoholism. The down-regulated mRNAs were mainly enriched in cGMP-PKG signaling pathway and Vascular smooth muscle contraction (**Figure 1**).

# Construction of Prognostic Signature Based on the ceRNA Network

Patients with incomplete clinical information (age, race, TNM stage, tumor grade, histological type, type of neoplasm, and follow-up information) were excluded from the following procedure. Univariate Cox regression analysis was applied to lncRNAs involved in the ceRNA network. Survival status and overall survival time analyses revealed nine lncRNAs with prognostic values in endometrial carcinoma (**Figure 2**). These

nine lncRNAs were subjected to multivariate Cox regression analysis. In this step, a function of step was applied to identify the best signature that predict the outcome of patients with endometrial carcinoma. A risk score formula based on LINC00491, LINC00483, ADARB2-AS1, and C8orf49 had the lowest AIC and was selected as the best signature. The risk assessment score for the prediction of overall survival was calculated as follows: Risk score = expLINC00491 × 0.13335 + expLINC00483 × 0.32495 + expADARB2−AS1 × 0.25997 + expC8orf49 × 0.22279. Patients were classified into two clusters using the cutoff value of the median risk score. Kaplan-Meier survival analysis indicated that the 5-year survival rates for low- and high-risk groups were more than 0.9 and 0.6, respectively (p < 0.001). The area under the curve in ROC analysis was 0.753, suggesting that this signature has a promising power in predicting the clinical outcome of patients with endometrial carcinoma (p < 0.001, sensitivity: 0.847, specificity: 0.393) (**Figure 3**).

each individual. Color of each plot represents the survival status of each patient. (E) Risk score of each individual.

Clinical information of patients with endometrial carcinoma is shown in **Table 1**. We used the chi-square test to estimate the correlation between risk level and other clinical factors, and found that the risk level was significantly correlated with TNM stage (p = 0.005), tumor grade (p = 0.005), histological type (p < 0.001), neoplasm type (p = 0.017), and vital status (p < 0.001). This finding indicates that the risk score signature was closely correlated with the above-mentioned clinical parameters (**Table 2**).

To evaluate the predictive power of this risk signature, clinical parameters such as race, age at diagnosis, tumor grade, TNM stage, pathological type, and neoplasm type were included in the survival analysis. As shown in **Figure 4**, aside from tumor grade (p = 0.004), TNM stage (p < 0.001), and pathological type (p = 0.006), type of neoplasm (p < 0.001) and risk score (p < 0.001) were directly related to prognosis of patients. Multivariate Cox regression analysis showed that only type of neoplasm (p < 0.001) and risk score (p = 0.001), but not TNM stage (p = 0.206), tumor

Xia et al. Prognostic Signature in Endometrial Cancer



grade (p = 0.558), and tumor pathological type (p = 0.576), were statistically independent predictive factors of poorer prognosis for endometrial cancer (**Figure 5**).

#### Protein-Protein Network Analyses

To better understand the mechanisms underlying the function of the four lncRNAs, protein-protein interactions of mRNAs involved in the ceRNA network of these four lncRNAs were constructed using the STRING website. In this PPI network, MEF2C has the closet connection with other proteins (**Figure 6**).

#### DISCUSSION

Endometrial cancer is one of the three leading gynecologic tumors. The Cancer statistics of 2019 revealed 61,880 new cases and 12,160 deaths in United States. (1). Several tumor markers such as CA125, HE4, CA199, and CEA are clinically used for the diagnosis of endometrial cancer. However, the pathological process of the occurrence and development of endometrial cancer is still unclear. More precise preoperative staging and preoperative diagnosis demand better pathophysiological development and new tumor markers of endometrial cancer. Epigenetics of genes, especially lncRNAs, have been recently used for the study of endometrial cancer. Although most lncRNAs lack the capacity of coding protein, many other functions of lncRNAs


\**p* < *0.05.*

have been found in endometrial cancer, including epithelial-tomesenchymal transition (4). But due to the technical limitations, functional studies of lncRNAs are not easy in comparison with those of coding RNAs. ceRNA hypothesis provided a new solution for achieving better functional studies of lncRNAs. It proposed that lncRNA can regulate miRNA abundance by binding and sequestering them. As such, lncRNAs can regulate the expression of target mRNAs. Thus, it has been shown that an efficient way to infer the potential function of lncRNAs is by studying their relationship with miRNAs and mRNAs, whose functions have been annotated. Taken advantage of that, we mapped the ceRNA network in endometrial cancer which could provide new insights to explore the mechanism of it.

In this research, we analyzed the differentially expressed genes to develop ceRNA network and investigated the molecular events that facilitate the development of endometrial carcinoma. Using univariate and multivariate Cox regression analyses, a signature based on four lncRNAs was developed that showed promising outcomes with respect to the prediction of the patient's overall survival. This signature was closely correlated with the TNM stage and tumor grade clinical parameters and served as an independent factor, like neoplasm cancer status and unlike TNM stage. Some clinical studies have suggested the association between diabetes as well as hypertension with the outcome of patients with endometrial cancer (23, 24); however, we did not perform statistical analysis of diabetes and hypertension in the present study, as more than 50% values were missing.


FIGURE 4 | Forest map of clinical characters in univariate analysis. The coordinate of diamond represents the odds ratio. Univariate Cox regression analysis was performed. Subgroup with a value of *p* < 0.05 was considered statistically significant.

FIGURE 5 | Forest map of clinical characters in multivariate analysis. The coordinate of the blue diamond represents the odds ratio. Multivariate Cox regression analysis was performed. Subgroup with a value of *p* < 0.05 was considered statistically significant.

In general, the survival outcome for patients with endometrial cancer is mainly predicted by two elements, namely, TNM stage and type of neoplasm, but a quantifiable index is lacking.

Several studies have shown that the expression of hormone receptors such as estrogen and progesterone receptors is a favorable independent prognostic factor (25, 26). In addition,

hormonal therapy was considered as a supplemental therapy in clinical setting. Mutation of the well-known gene P53 is associated with poor clinical outcome, and the overall survival of patients with endometrial cancer with alterations in p53 gene expression was much lower than that of patients with the wildtype p53 (27). Moreover, the overexpression of human epidermal growth factor receptor 2 (HER2) gene is an independent prognostic factor associated with poor overall survival (28).

An efficient and sensitive marker that predicts the outcome of endometrial cancer is still lacking. In this study, we identified a signature based on LINC00491, LINC00483, ADARB2-AS1, and C8orf49 to discriminate patients with endometrial cancer that are at increased risk for poor outcome in combination with the information related to TNM stage and type of neoplasm.

The results of this analysis showed that the expressions of four lncRNAs (LINC00491, LINC00483, ADARB2-AS1, and C8orf49) were markedly different between endometrial cancer tissues and normal endometrial tissues. Many of these genes are incompletely studied. C8orf49 is also called as GATA4 downstream membrane gene (G4DM). GATA4 (GATA transcription factor 4) is a zinc-finger transcription factor involved in the development of heart and adult cardiomyocytes. Mutations in GATA4 gene may cause congenital heart diseases (29) such as the tetralogy of Fallot, atrial septal defect, ventricular septal defect, atrioventricular septal defect, and dilated cardiomyopathy. A key feature of GATA4 is its twozinc finger domain, which binds to the specific region of the target gene. In serval cancer types, GATA4 serves as a potential tumor suppressor, and hypermethylation and hypomethylation of GATA4 are closely related to the malignant behavior of cancers (30, 31). GATA4 may also be used as a biomarker for ovarian cancer (32). The expression of GATA4 may change during cardiocyte differentiation through the effect of the transcription of the target gene. GATA4 gene is also expressed in the uterus, and C8orf49 is one of the target genes of GATA4. Therefore, C8orf49 may play an important role in the differentiation of endometrial cancer cells. In addition, studies have indicated that ADARB2-AS1, as an open-reading frame, may contribute to the risk of pancreatic ductal adenocarcinoma (PDAC). Along with other seven lncRNAs, ADARB2-AS1 showed better accuracy than the standard clinical and radiologic features in distinguishing aggressive/malignant IPMNs (33). However, the underlying mechanism is unclear.

It is well known that hub genes play critical roles in biological networks. Therefore, node degrees of lncRNA involved in this ceRNA network were calculated. A lncRNA with a node degree >5 was considered as hub lncRNA. In this study, total of 15 lncRNAs were identified with high degree in the ceRNA network. C8orf49 which has prognostic value also act as hub lncRNA in endometrial cancer. This suggests that C8orf49 may play critical roles in the origin and development of endometrial cancer. Here, we demonstrate for the first time the construction of a ceRNA network in endometrial cancer to reveal the molecular mechanism that facilitates the development of endometrial cancer. A signature based on LINC00491, LINC00483, ADARB2- AS1, and C8orf49 was identified as a biomarker to discriminate between patients with high and poor risk outcome. The lncRNAs involved in this signature may serve as therapeutic targets for precision medicine in endometrial cancer. Further studies are warranted to explore the biological function and reveal the molecular mechanism underlying the role of LINC00491, LINC00483, ADARB2-AS1, and C8orf49 in endometrial cancer.

#### CONCLUSION

This study focused on a ceRNA network to provide a novel perspective and insight into endometrial cancer and suggested that the signature based on LINC00491, LINC00483, ADARB2-AS1, and C8orf49 could serve as an independent prognostic biomarker in endometrial cancer.

#### ETHICS STATEMENT

High-throughput sequencing-counts (HTSeq-counts) and miRNA sequencing profiles were obtained from the TCGA data portal, which is a publicly available dataset. Therefore, no ethics approval is needed.

## AUTHOR CONTRIBUTIONS

LX constructed ceRNA network using R software. QM analyzed the data using SPSS. YW download data from TCGA. XS and BW used photoshop and illustration software. JS and CZ organized

#### REFERENCES


data. JW perform GO and KEGG analysis. HH perform PPI network. MX designed this study.

#### FUNDING

This work was supported by National Key Research and Development Program of China (NO:2016YFC1303100).

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fonc. 2019.00448/full#supplementary-material

Supplement Figure 1 | Heatmap of differentially expressed RNAs. The one on the left is the heatmap of 50 randomly selected differentially expressed mRNAs. The center one is the heatmap of 50 differentially expressed lncRNAs, while the one on the right is the heatmap of 50 randomly selected differentially expressed miRNAs. Orange indicates high-level RNA expression, whereas blue indicates low expression. The first row of each map is the type of each sample. Pink represents normal samples, while blue represents cancerous samples.

Supplement Figure 2 | All lncRNA node degree analysis reveals specific properties of the ceRNA network.

Supplement Table 1 | ceRNA network of endometrial cancer.


prognosis of early stage endometrial carcinoma. Am J Obstet Gynecol. (1985) 151:922–32. doi: 10.1016/0002-9378(85)90671-4


Biomed Pap Med Fac Univ Palacky Olomouc Czech Repub. (2013) 157:294– 7. doi: 10.5507/bp.2013.079


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Xia, Wang, Meng, Su, Shen, Wang, He, Wen, Zhang and Xu. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# A Micro-Costing Study of Screening for Lynch Syndrome-Associated Pathogenic Variants in an Unselected Endometrial Cancer Population: Cheap as NGS Chips?

Neil A. J. Ryan1,2, Niall J. Davison<sup>3</sup> , Katherine Payne<sup>3</sup> , Anne Cole<sup>4</sup> , D. Gareth Evans 2,4,5 and Emma J. Crosbie1,5,6 \*

*<sup>1</sup> Gynaecological Oncology Research Group, Division of Cancer Sciences, School of Medical Sciences, Faculty of Biology, Medicine and Health, The University of Manchester, Manchester, United Kingdom, <sup>2</sup> Division of Evolution and Genomic Medicine, School of Medical Sciences, Faculty of Biology, Medicine and Health, The University of Manchester, St Mary's Hospital, Manchester, United Kingdom, <sup>3</sup> Manchester Centre for Health Economics, Division of Population Health, Health Services Research and Primary Care, The University of Manchester, Manchester, United Kingdom, <sup>4</sup> Manchester Centre for Genomic Medicine, St Mary's Hospital, Manchester University NHS Foundation Trust, Manchester Academic Health Science Centre, Manchester, United Kingdom, <sup>5</sup> Division of Cancer Sciences, Faculty of Biology, Medicine and Health, NIHR Manchester Biomedical Research Centre, School of Medical Sciences, The University of Manchester, Manchester, United Kingdom, <sup>6</sup> Department of Obstetrics and Gynaecology, St Mary's Hospital, Manchester University NHS Foundation Trust, Manchester Academic Health Science Centre, Manchester, United Kingdom*

#### Edited by:

*Massimo Nabissi, University of Camerino, Italy*

#### Reviewed by:

*Margaret Liang, University of Alabama at Birmingham, United States Suzana Arenhart Pessini, Universidade Federal do Rio Grande do Sul (UFRGS), Brazil*

\*Correspondence:

*Emma J. Crosbie emma.crosbie@manchester.ac.uk*

#### Specialty section:

*This article was submitted to Women's Cancer, a section of the journal Frontiers in Oncology*

Received: *14 November 2018* Accepted: *21 January 2019* Published: *26 February 2019*

#### Citation:

*Ryan NAJ, Davison NJ, Payne K, Cole A, Evans DG and Crosbie EJ (2019) A Micro-Costing Study of Screening for Lynch Syndrome-Associated Pathogenic Variants in an Unselected Endometrial Cancer Population: Cheap as NGS Chips? Front. Oncol. 9:61. doi: 10.3389/fonc.2019.00061* Background: Lynch syndrome is the most common inherited cause of endometrial cancer. Identifying individuals affected by Lynch syndrome enables risk-reducing interventions including colorectal surveillance, and cascade testing of relatives.

Methods: We conducted a micro-costing study of screening all women with endometrial cancer for Lynch syndrome using one of four diagnostic strategies combining tumor microsatellite instability testing (MSI), immunohistochemistry (IHC), and/or *MLH1* methylation testing, and germline next generation sequencing (NGS). Resource use (consumables, capital equipment, and staff) was identified through direct observation and laboratory protocols. Published sources were used to identify unit costs to calculate a per-patient cost (£; 2017) of each testing strategy, assuming a National Health Service (NHS) perspective.

Results: Tumor triage with MSI and reflex *MLH1* methylation testing followed by germline NGS of women with likely Lynch syndrome was the cheapest strategy at £42.01 per case. Tumor triage with IHC and reflex *MLH1* methylation testing of MLH1 protein-deficient cancers followed by NGS of women with likely Lynch syndrome cost £45.68. Tumor triage with MSI followed by NGS of all women found to have tumor microsatellite instability cost £78.95. Immediate germline NGS of all women with endometrial cancer cost £176.24. The cost of NGS was affected by the skills and time needed to interpret results (£44.55/patient).

**45**

Conclusion: This study identified the cost of reflex screening all women with endometrial cancer for Lynch syndrome, which can be used in a model-based cost-effectiveness analysis to understand the added value of introducing reflex screening into clinical practice.

Keywords: micro-costing, Lynch syndrome, endometrial cancer, genetic testing, screening

# INTRODUCTION

Lynch syndrome is an inherited predisposition to a constellation of different cancers, of which colorectal and endometrial cancer are the most common (1). Estimates of the prevalence of Lynch syndrome among the general population are as high as 300 per 100,000 (2). Lynch syndrome confers a lifetime risk of endometrial cancer of 30 to 40% (3). Endometrial cancer may be the sentinel event in women with Lynch syndrome, providing an early diagnostic opportunity (4). Those found to have Lynch syndrome are offered risk-reducing interventions including colorectal surveillance to reduce cancer-specific mortality (5). A diagnosis of Lynch syndrome enables cascade testing within families and its identification in those who are yet to develop cancer (6). Identified women may be offered prophylactic surgery to reduce their risk of gynecological cancer (7).

Lynch syndrome arises from germline pathogenic variants within the highly conserved mismatch repair (MMR) system. The molecular characteristics of Lynch syndrome-associated cancers enable tumor-based triage and targeted germline sequencing of the MMR genes, commonly performed by next generation sequencing (NGS). A Lynch syndrome-associated tumor classically shows aberrant expression of associated MMR proteins, MLH1, MSH2, MSH6 and/or PMS2, and microsatellite instability (MSI) (8). Loss of MLH1 expression through somatic methylation of the MLH1 promotor region is a common sporadic event in endometrial cancer; MLH1 methylation testing is therefore an effective way of reducing the number of women with MLH1 loss by IHC or whose tumors are MSI-H from expensive germline testing (9).

The National Institute for Health and Care Excellence (NICE) recommends universal screening for Lynch syndrome in people with colorectal cancer (10). Universal screening was identified to be a cost-effective use of healthcare resources because of the number of colorectal cancers prevented in family members who are also found to carry Lynch syndrome (11). Screening women with endometrial cancer provides a further opportunity to save lives from Lynch syndrome-associated cancer (12), however, the costs associated with this screening strategy are not known. The aim of this study was to identify and quantify the resource use and costs associated with different diagnostic strategies relevant to screening for Lynch syndrome in an unselected endometrial cancer population.

# METHODS

A micro-costing study was performed to identify the resource use and cost per patient of four diagnostic testing strategies for Lynch syndrome in an unselected endometrial cancer population. The study assumed the perspective of the National Health Service (NHS) in England. The direct costs associated with providing each diagnostic testing strategy were identified. The time horizon for identifying the relevant resources use in this study started with the process of gaining informed consent for any Lynch syndrome testing and ended with generating a report of the final diagnostic test result.

#### Diagnostic Testing Strategies

Four Lynch syndrome testing technologies that reflect current and emerging national clinical practice (10) were included in this study: microsatellite instability (MSI) testing; immunohistochemistry (IHC); MLH1 promoter hypermethylation pyrosequencing (methylation) testing; and next generation sequencing (NGS) (13). The technologies used to produce each diagnostic test were conceptualized into representative clinical pathways using a decision tree to provide a structured approach to represent how each technology would form part of a diagnostic testing strategy for a defined population of women with suspicion of Lynch syndrome. The relevant outcome of testing was defined as either Lynch syndrome diagnosed or not diagnosed. The four diagnostic testing strategies were:

**Strategy 1:** Initial tumor triage with MSI followed by germline NGS testing for pathogenic variants of the MMR genes for all those found to have microsatellite instability (MSI-H).

**Strategy 2:** Initial tumor triage with MSI followed by reflex MLH1 methylation testing for MSI-H tumors, and germline NGS testing for women where the tumor MLH1 methylation test shows no hypermethylation.

**Strategy 3:** Initial tumor triage with IHC followed by reflex MLH1 methylation testing for tumors with MLH1 loss. Germline NGS testing for pathogenic variants of the MMR genes for all women whose tumors show MSH2, MSH6 or PMS2 loss, or MLH1 loss where the MLH1 methylation test shows no hypermethylation.

**Strategy 4:** No initial tumor triage. All women with endometrial cancer undergo direct germline NGS testing for pathogenic variants of the MMR genes.

The decision tree (**Figure 1**) was conceptualized through discussion with a panel of 10 local and national experts. These experts included two consultant histopathologists, three senior

**Abbreviations:** NGS, Next generation sequencing; MSI, Microsatellite instability; MSI-H, Microsatellite instability high; IHC, Immunohistochemistry; MMR, Mismatch repair; NICE, National Institute for Health and Care Excellence; NHS, National Health Service, England.

(>band 8a Agenda For Change pay scale) clinical laboratory scientists, two consultant gynecological oncology surgeons, two consultant clinical geneticists, and one consultant genetic pathologist. The decision tree represented the proportion of cases testing positive with each technology and subsequent impact on the need to conduct further testing. The input values for the decision tree were informed by a pragmatic review of the literature (see **Supplementary Appendix 1**), which identified the relevant studies to inform the probability of a positive or negative test result in each scenario.

#### Identifying Resource Use

The use of resources was identified for each diagnostic testing strategy, assuming the NHS perspective. Resource use included clinical and laboratory staff time, capital equipment, and laboratory consumables.

Clinical and laboratory staff time was collected using a prospective study at a large tertiary genetics and gynecology oncology surgical referral center in the North West of England between 2015 and 2017. The study was approved by the North West Research Ethics Committee (15/NW/0733) and all patients gave written, informed consent to participate. Non-participant direct observation was used to identify the time members of staff dedicated to each element of the diagnostic test process. A single observer recorded timings with a stopwatch over multiple rounds of observation between 2015 and 2017, using a structured data collection tool and recording information on sample batch size and staff pay grade. Participant direct observation was used to record the process of obtaining informed consent for Lynch syndrome testing. Because this was done as part of a research study, all participants were consented prior to any testing. The consent process included discussion about the various tumor tests (MSI, IHC, methylation testing) and indicative germline Lynch syndrome testing.

Hospital information systems were used to derive resource use on capital equipment including diagnostic platforms. Total capital throughput for a given piece of equipment was identified as the total annual number of samples it processed. Total Lynch syndrome-associated capital throughput for a given piece of equipment was identified as the total annual number of Lynch syndrome-associated samples it processed, assuming all newly diagnosed endometrial cancer patients were offered testing.

Consumables were identified from laboratory standard operating procedures and through completion of a structured data collection form by seven laboratory staff who routinely test patients for Lynch syndrome, including biomedical technicians, a consultant clinical scientist, a principal clinical scientist, and clinical genetics technicians.

# Collating Unit Costs

The unit costs (UK sterling; £) for consumables and equipment were extracted from published list prices, or hospital invoices where list prices were unavailable. The unit costs for staff labor were defined as cost per minute using the midpoints of salary grades as per NHS Agenda for Change or the British Medical Association's Hospital Doctor pay scales (see **Table S4**). The price year for unit costs was standardized at 2017. **Table S2** shows the complete list of unit costs and sources.

# Data Analysis

All analysis was carried out using Microsoft Corporation software program Microsoft Office Excel 2011. The base case analysis calculated the direct medical costs of each of the four diagnostic testing strategies shown in **Figure 1**. The costs were calculated for testing a single sample by multiplying the relevant unit cost (see **Table S2**) with the relevant items and quantities of resource use.

One-way sensitivity analysis was used to identify the impact of using different estimates for the probability of a positive or negative test result for each diagnostic testing strategy. The ranges of the probability input values were informed by a pragmatic literature search (**Supplementary Appendix 1, Table S1**) and used to assess the effect different test outcomes would have on the overall cost of each strategy. Two separate scenario analyses explored the impact of the timing of consent on the cost of each diagnostic testing strategy. Two scenarios were explored to understand the cost if consent was taken before any testing, compared with a scenario in which consent was only taken for those needing germline NGS testing.

# RESULTS

**Table 1** shows the individual quantities of resource use for MSI, IHC, MLH1 methylation testing, and NGS, which were used to calculate the total costs incurred per patient for each of the four diagnostic testing strategies.

Tumor triage with MSI and reflex MLH1 methylation testing followed by germline NGS of women with likely Lynch syndrome (Strategy 2) was the cheapest at £42.01 per patient. Tumor triage with IHC and reflex MLH1 methylation testing of MLH1 protein-deficient cancers followed by NGS of women with likely Lynch syndrome (Strategy 3) cost £45.68. Tumor triage with MSI followed by NGS of all women found to have MSI-H tumors (Strategy 1) cost £78.95. Immediate germline NGS of all women with endometrial cancer (Strategy 4) cost £176.24.

The cost of consenting a woman for Lynch syndrome testing was calculated from 269 directly observed episodes. Two women declined Lynch syndrome testing. On average, the process of gaining consent took 7 min and 39 s (SD: 5 min, 16 s) of consultant time. This cost an average of £5.32.

Immediate, unselected germline NGS testing for pathogenic variants of the MMR genes in all women with endometrial cancer was the most expensive testing strategy. Consumables were an expensive component, costing £115.14 overall, including DNA extraction. As germline DNA is required, it was assumed that all samples for NGS would require de-novo DNA extraction from blood. Equipment costs were expensive for NGS, at £5.43 per sample tested. Labor costs were also relatively high due to the complexity of data interpretation, costing £50.35 per sample.

Strategies one to three involved the use of tumor-based triage with IHC and MSI. Tumor based triage by IHC was cheaper than MSI (£21.17 vs. £27.67). The most expensive resource was consumables (£12.23 vs. £19.19 for IHC and MSI, respectively). Labor costs were similar (£8.51 vs. £8.30 for IHC and MSI, respectively); this was despite the need for a consultant grade doctor to interpret the IHC results, because the per-sample time was relatively short. Equipment costs were more expensive for IHC at £0.43 per sample due to use of a dedicated staining platform and associated maintenance costs. These costs were cheaper for MSI testing at £0.18 per sample, using a commercially available kit.

MLH1 methylation testing was a component of diagnostic strategies two and three, and cost £20.60 and £28.41, respectively, when needed. Methylation testing was cheaper in the context of Strategy two because DNA extraction had already been done for the initial MSI testing. Methylation testing included labor costs at £3.94 (including DNA extraction) or £2.07 (excluding DNA extraction), and equipment costs at £0.23 per sample. Therefore, assuming 30% of endometrial tumors are MSI-H, methylation testing saves £36.95 per patient tested by this strategy because it removes the need for expensive germline NGS by the majority. Incorporating methylation testing in Strategy three (assuming 35% of samples show MMR loss, of which 27% is due to loss of MLH1) reduces the cost of this strategy by £35.21 per patient.

#### Sensitivity Analysis

One-way sensitivity analysis explored the potential variation in the cost of each diagnostic testing strategy using pessimistic and optimistic values for the probability of a positive test result and need for subsequent testing (see **Supplementary Appendix 2, Table S3**). Depending on the source of data, between 22 to 30% of cases in Strategy one require subsequent germline NGS testing (14, 15). The incorporation of MLH1 methylation testing in Strategy two reduces this proportion to 5–10% of cases (14–16). For Strategy three, MLH1 loss is observed in 16–27% of cases (15, 17), but ∼93% of these are due to MLH1 hypermethylation, meaning that just 7% of those women with MLH1 loss by IHC require germline NGS testing for Lynch syndrome (15). Non-MLH1 protein loss by IHC is seen in 6–8% endometrial tumors and all of these require germline NGS testing (**Table S3**).

The range of women who were found to carry a pathogenic variant associated with Lynch syndrome in Strategy 2 is between 1 and 3% of cases (14, 15). For Strategy three, 11% of those with MLH1 loss and no MLH1 hypermethylation tested by NGS will have a pathogenic variant in MLH1, according to current literature (15). Twenty percent of those with MSH2, MSH6, or PMS2 loss by IHC will be found to carry pathogenic variants of MSH2, MSH6, or PMS2 (15). Varying the proportions of women requiring subsequent tests in the decision tree, based on pessimistic and optimistic values for the probability of a positive test result in, showed no significant impact on the expected costs for each diagnostic testing strategy.

A scenario analysis explored the impact of the timing of taking consent for Lynch syndrome testing. If consent was only taken at the point of germline NGS testing, the overall cost for Strategy one is cheaper at £75.22 per patient. The overall cost of Strategies two, three, and four would be £36.95, £40.89, and


Frontiers in Oncology | www.frontiersin.org

TABLE 1 | Resource use and unit costs.


Frontiers in Oncology | www.frontiersin.org

**50**

∧*Equipment*

#*DNA extraction would only needed to be done once per somatic or genomic sample.*

◦*Some activities include periods of automation but only hands-on time is recorded.*

 *costs per sample consider estimated equipment life span and percentage use for endometrial*

 *cancer lynch syndrome diagnostics*

£176.24, respectively. Therefore, taking consent only at the point of germline NGS testing saved the Strategy cost by around £4.52 per person, on average (range £3.73–£5.06).

## DISCUSSION

This study presents the first comprehensive micro-costing analysis of diagnostic strategies for Lynch syndrome testing in endometrial cancer. In total, four pathways were quantified, reflecting the diagnostic strategies relevant to current clinical practice in the UK. The expected costs were £42.01 or £45.68 per case, respectively if MSI or IHC were used for tumor triage and reflex MLH1 methylation testing was followed by germline NGS of women with likely Lynch syndrome. Immediate germline NGS for all women with endometrial cancer cost £176.24.

Goverde et al. (18) cites estimates close to our calculated costs for three of the diagnostic strategies (MSI e89, IHC e135, MLH1 hypermethylation e99). However, the cost of NGS was considerably more expensive than our calculation, at e2152 per test (18). Our findings indicate the cost of Lynch syndrome testing in endometrial cancer is sometimes considerably cheaper than previously described. Published cost effectiveness studies to date may have overestimated the cost of Lynch syndrome testing in clinical practice (18–20). In a published model-based costeffectiveness analysis of screening for Lynch Syndrome in people with colorectal cancer, the estimated unit cost for IHC was £210 and £202 for MSI and £136 per test for MLH1 hypermethylation. Using NGS for four MMR gene NGS was estimated to cost between £650 and £860 (21). These unit costs were derived from expert estimates from the UK Genetic Testing Network (21). Three model-based cost-effectiveness analysis of unselected endometrial cancer screening for Lynch syndrome base their analysis on these estimates or on insurance charges (18–20). Two of these published studies concluded that using tumor triage with IHC and reflex MLH1 methylation testing of MLH1 proteindeficient cancers followed by NGS of women with likely Lynch syndrome (our strategy 3) was cost effective for Lynch syndrome screening, despite using higher costs in their modeling(18, 20). One study that used the highest estimated cost for the diagnostic tests indicated that Lynch syndrome screening was not costeffective (22). These findings are consistent with the observation of Grosse (23) who indicated that the assumed cost of the diagnostic strategy is a key driver of the relative cost-effectiveness of testing for Lynch syndrome (23).

Ours is the first micro-costing study of Lynch syndrome testing in endometrial cancer. Previous micro-costing studies of genetic-based tests have only measured the costs of unselected sequencing of samples (akin to our strategy 4), without quantifying the impact of using tumor based triage (24, 25). Griffith et al (26) micro-costed two gene (MLH1/MSH2) mutational screening at £1212.17, but this study pre-dates NGS technology and is therefore no longer relevant to clinical practice (26).

Our study was set within a gynecological oncology center comparable with other centers in the UK. It benefited from prospective recruitment and therefore direct observation of 269 patient consent episodes. All but two of 269 patients agreed to Lynch syndrome testing. This is higher than expected from the literature and may reflect testing within a publicallyfunded healthcare system rather than one based on health insurance, where a positive test result is likely to impact future insurance premiums (27). Taking informed consent is a variable process as each patient encounter is unique; understanding the uncertainty of the process was possible using the multiple observations. We also used direct observations for all non-automated step in the laboratory testing process. The decision tree was conceptualized by a panel of experts with representation from across the diagnostic pathway, therefore reflecting current clinical practice. The sensitivity analysis drew on multiple high-quality studies sourced through an extensive literature search.

To the best of our knowledge, ours may be the first study to micro-cost NGS testing for any indication. Our microcosting study indicated that analysis of raw NGS output is time consuming and requires considerable expertise. Indeed, 80% of labor costs, and 25% of the overall costs for NGS, were spent on data analysis. It is departmental policy for all NGS results to be analyzed twice, first by an agenda for change (AFC) band 5 and then a band 7 member of staff. Reports are then authorized by a senior member of staff (band 8a). This is for quality assurance purposes and is in keeping with international recommendations (28).

Another key finding was the impact of the timing of patient consent on the overall costs of testing. Consent is fundamental to germline genetic testing since patients have the absolute right to refuse to be tested (29). However, the point at which consent is sought has a considerable impact on the overall cost of Lynch syndrome testing. Consent taken prior to any testing would add this cost uniformly to all strategies. However, if consent were taken only at the point of germline NGS analysis, only a small proportion of women (5–30%) would need to be consented. Somatic tumor analyses are commonplace in histopathology, for example p53 IHC. Consent is not taken for such tests because they are integral to accurate diagnosis, as well as informing prognosis and treatment planning. Crucially such tests make inferences about cancer biology and not the individual's genome. Such an argument could easily be applied for tumor-based Lynch syndrome testing as such tests merely stratify an individual's risk of having Lynch syndrome and do not diagnose a germline condition. Moving the consent to the point of germline testing would require an additional face-to-face meeting with the patient. This could take place in the context of routine cancer follow-up, therefore mitigating the need for an additional appointment and its associated costs. However, any impact of moving consent to the point of germline testing on uptake and health state utility cannot be ascertained from our data, since all patient consents were taken at recruitment into the study, before any testing was carried out.

Our study is limited by the fact that all observations originate from a single site and therefore may not be generalizable. Their application outside England is difficult to assess given geographically distinct populations, diverse health systems and variable costs. We have not accounted for failed tests that need to be repeated at additional cost. It was also not possible to analyse capital costs like heating, lighting, and rent. Capital costs are not insignificant, however the confidentiality implicit in private finance initiatives prevented their incorporation in our analysis. Therefore, the true cost of Lynch syndrome testing is likely to lie somewhere between the cost identified by this micro-costing study, which was as thorough as possible but not exhaustive in its pursuit of all applied costs, and the well-sourced expert estimates, which overestimate costs to ensure the service does not operate at a net loss (30).

A further limitation of our work is that the proportion of positive test results used in our decision tree originated from populations studied outside the UK and therefore may not be fully representative of the local situation. This is of particular concern given that such patients were tested within insurancebased healthcare systems, potentiating a selection bias in which high risk individuals decline testing to avoid would-be increased premiums. This limitation is unavoidable given the complete lack of UK data relating to the prevalence of Lynch syndrome in endometrial cancer patients.

The extrapolation of our data to inform Lynch syndrome testing in colorectal cancer patients is problematic given their use of BRAF V600E as a proxy of MLH1 promoter hypermethylation in tumor-based triage (31). Furthermore, it is not clear if the proportions used in our decision tree are transferable to a colorectal cancer population. Nonetheless, the costs of IHC, MSI, and NGS testing (as opposed to the cost of testing strategies) are directly transferable and could be used to improve the robustness of model-based cost effectiveness analysis of Lynch syndrome testing in colorectal cancer.

Regarding tumor-based triage, a strategy using MSI analysis is marginally cheaper than one using IHC as the primary test (£42.01 vs. £45.68). A £3.67 saving per tumor would save the NHS over £33,000 annually, were all 9,000 new diagnoses of endometrial cancer tested for Lynch syndrome. However, the choice of tumor triage is complex and beyond a simple cost comparison; this is especially the case when the cost difference is marginal. Both methods have equitable sensitivity and specificity according to the literature (11). IHC has the advantage of identifying the likely mutated gene, which can aid the subsequent interpretation of sequencing data (32). IHC can be performed in most histopathology departments, whereas MSI requires specialist laboratories. IHC is thought to be more sensitive in the case of those carrying a MSH6 pathogenic variant, where tumors may not be MSI-H (33). However, interpretation of MMR expression patterns requires consultant pathology expertise, and only MSI can identify missense pathogenic variants in MMR genes whereby the protein is expressed but is not functional (34). The choice of tumor triage depends on the availability of local services, expertise, and infrastructure; these data do not infer that any one strategy is clinically superior to another for the diagnosis of Lynch syndrome.

The importance of our work is its ability to inform healthcare policy. There is a growing call for screening all endometrial cancer patients for Lynch syndrome (12, 35). A potential barrier to the transition from expert opinion to clinical application is cost. To date there have been no micro-costing data available to inform policy makers as to the actual costs of Lynch syndrome testing in the UK. As outlined above, the current estimated costs might be prohibitively high and thus impede the implementation of Lynch syndrome testing in endometrial cancer. Our data should prompt healthcare providers to look at this again.

# CONCLUSION

We present a micro-costing study for Lynch syndrome testing in unselected endometrial cancer patients from a large tertiary referral center in the North West of England. The use of tumor triage with MSI and reflex MLH1 methylation testing is the cheapest strategy at £42.01 per case. Substituting MSI for IHC as the initial tumor-based triage increases costs marginally to £45.68. Moving the point of consent for Lynch syndrome testing to just before germline testing reduces the costs of those strategies that incorporate tumor triage. NGS panel testing is considerably cheaper than current estimates at £176.24 per test. The next phase is to use these estimates in model-based cost-effectiveness analysis to understand the relative value of a national Lynch syndrome-testing programme for endometrial cancer.

# AUTHOR CONTRIBUTIONS

NR recruited to the study and collected data. NR and ND analyzed data and wrote the first draft of the manuscript. AC collected data. EC and KP designed the study, supervised study execution and contributed to data interpretation. DE provided study oversight. EC was Chief Investigator and study guarantor. All authors provided critical comment, edited the manuscript, and approved its final version.

# FUNDING

NR is a Doctoral Medical Research Council (MRC) Research Fellow (MR/M018431/1), DE an National Institute for Health Research (NIHR) Senior Investigator (NF-SI-0513-10076), EC an (NIHR) Clinician Scientist (NIHR-CS-012-009), and their work is supported through the NIHR Manchester Biomedical Research Centre (IS-BRC-1215-20007). This article presents independent research funded by the NIHR and MRC. The views expressed are those of the authors and not necessarily those of the MRC, NHS, NIHR, or the Department of Health.

# ACKNOWLEDGMENTS

We are very grateful to all the patients and clinical staff who kindly agreed to take part in this study. Special thanks goes to Dr. Andrew Wallace, Ms. Shona Esquibel, Mrs. Catherine Higgins, Ms. Sancha Bunstone, Dr. Simon Tobi, Ms. Emma Hobson, and Ms. Megan Adaway who provided access to their laboratories and allowed us to time their activities. We would like to thank Dr. James Bolton, Dr. Rhona McVey, Dr. Ian Frayling, Prof Richard Edmondson, and Dr. Fiona Lalloo for their expert advice.

#### REFERENCES


#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fonc. 2019.00061/full#supplementary-material

endometrial tumors and clinical features in young women compared with older women. Int J Gynecol Cancer. (2010) 20:1549–56.


classification. J Med Genet. (2012) 49:151–7. doi: 10.1136/jmedgenet-2011- 100714


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Ryan, Davison, Payne, Cole, Evans and Crosbie. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Assessing the Role of Selenium in Endometrial Cancer Risk: A Mendelian Randomization Study

Pik Fang Kho<sup>1</sup> , Dylan M. Glubb<sup>1</sup> , Deborah J. Thompson<sup>2</sup> , Amanda B. Spurdle<sup>1</sup> and Tracy A. O'Mara<sup>1</sup> \* on Behalf of the Endometrial Cancer Association Consortium†

<sup>1</sup> Molecular Cancer Epidemiology Group, Genetics and Computational Biology Department, QIMR Berghofer Medical Research Institute, Brisbane, QLD, Australia, <sup>2</sup> Department of Public Health and Primary Care, Centre for Cancer Genetic Epidemiology, University of Cambridge, Cambridge, United Kingdom

Endometrial cancer is the most commonly diagnosed gynecological cancer in developed countries. Based on evidence from observational studies which suggest selenium inhibits the development of several cancers (including lung and prostate cancer), selenium supplementation has been touted as a potential cancer preventative agent. However, randomized controlled trials have not reported benefit for selenium supplementation in reducing cancer risk. For endometrial cancer, limited observational studies have been conducted assessing whether selenium intake, or blood selenium levels, associated with reduced risk, and no randomized controlled trials have been conducted. We performed a two-sample Mendelian randomization analysis to examine the relationship between selenium levels (using a composite measure of blood and toenail selenium) and endometrial cancer risk, using summary statistics for four genetic variants associated with selenium levels at genome-wide significance levels (P < 5 × 10−<sup>8</sup> ), from a study of 12,906 endometrial cancer cases and 108,979 controls, all of European ancestry. Inverse variance weighted (IVW) analysis indicated no evidence of a causal role for selenium levels in endometrial cancer development (OR per unit increase in selenium levels Z-score = 0.99, 95% CI = 0.87–1.14). Similar results were observed for sensitivity analyses robust to the presence of unknown pleiotropy (OR per unit increase in selenium levels Z-score = 0.98, 95% CI 0.89–1.08 for weighted median; OR per unit increase in selenium levels Z-score = 0.90, 95% CI = 0.53–1.50 for MR-Egger). In conclusion, these results do not support the use of selenium supplementation to prevent endometrial cancer.

Keywords: Mendelian randomization, endometrial cancer, toenail selenium, circulating selenium, genome-wide association study

#### INTRODUCTION

Endometrial cancer is the most commonly diagnosed cancer of the female reproductive system in developed countries (1). Unlike breast and cervical cancers where a screening program is available to the general population, there is currently no available screening test for endometrial cancer and diagnosis relies on biopsy in symptomatic patients (2). Furthermore, the incidence of endometrial cancer is rising (3), highlighting the need for preventative measures. Selenium has received considerable attention as a possible cancer preventive agent [reviewed in (4)]. While randomized controlled trials have shown no benefit for selenium supplementation in reducing cancer risk over a period of up to 8 years (5), some observational longitudinal studies assessing

#### Edited by:

Massimo Nabissi, University of Camerino, Italy

#### Reviewed by:

James Yarmolinsky, University of Bristol, United Kingdom Carolina Bonilla, University of São Paulo, Brazil

\*Correspondence: Tracy A. O'Mara tracy.omara@qimrberghofer.edu.au

†Full authorship details for Endometrial Cancer Association Consortium are provided in the Supplementary Note

#### Specialty section:

This article was submitted to Women's Cancer, a section of the journal Frontiers in Oncology

Received: 28 November 2018 Accepted: 04 March 2019 Published: 27 March 2019

#### Citation:

Kho PF, Glubb DM, Thompson DJ, Spurdle AB and O'Mara TA (2019) Assessing the Role of Selenium in Endometrial Cancer Risk: A Mendelian Randomization Study. Front. Oncol. 9:182. doi: 10.3389/fonc.2019.00182 selenium intake or selenium levels, over a period up to 25 years, have shown an inverse association between selenium and cancer risk [reviewed in (4)]. Thus, although findings from the longitudinal studies have been inconsistent (4), they may provide insight into the longer term effects of selenium exposure. A recent meta-analysis examining the association between selenium intake (dietary and supplemental) and overall cancer risk, has suggested that there was a reduction in cancer incidence among people consuming more than the recommended daily allowance of selenium (55 µg/day; RR = 0.96, 95% CI = 0.92– 0.99) (6).

Very few studies have assessed the effects of selenium on endometrial cancer. In terms of cellular studies, it has been shown that a selenium metabolite can inhibit endometrial cancer cell proliferation, potentially through disruption of estrogen signaling (7). Findings from human studies, however, have been more equivocal. A population-based, case-control observational study of 417 endometrial cancer cases and 395 controls specifically assessed the role of dietary and supplemental selenium intake (as measured by questionnaire in the 6 months prior to diagnosis or enrolment as a control) in endometrial cancer development (8). In a comparison of the highest (≥103.2 µg) and lowest (<72.4 µg) selenium quartiles, this study did not support an association between selenium intake and endometrial cancer risk (OR = 0.74, 95% CI = 0.47–1.17) (8). Two small casecontrol studies (n < 100) have assessed serum selenium levels in endometrial cancer cases and controls. Sundstrom et al. (9) reported lower blood selenium levels in 64 cases as compared to 61 non-cancer controls, with an average of 1.01 ± 0.05 v 1.40 ± 0.08 µmol/L blood selenium in cases and controls, respectively (P < 0.001). A subsequent study of 35 endometrial cancer cases and 32 non-cancer controls reported a similar finding (average of 1.14 ± 0.04 vs. 1.26 ± 0.03 µmol/L blood selenium in cases and controls, respectively, P < 0.01) (10). Inconsistent results from these observational studies may be due to small sample sizes (8–10), reverse causation bias (9, 10), recall bias and measurement error in the dietary assessment (8). No prospective studies have examined the association of pre-diagnostic selenium levels with endometrial cancer risk. Thus, the role of selenium in endometrial cancer development remains inconclusive.

As no intervention study has yet been performed to explore the role of selenium in endometrial cancer risk, we employed a two-sample Mendelian randomization approach which uses germline genetic variants associated with selenium levels to proxy for selenium exposure (11). These germline genetic variants are largely independent from environment or lifestyle factors, and are established prior to disease onset, thus analyses using these genetic variants as instrumental variables are less susceptible to biases from confounding and reverse causation. Further, genetic effects on exposure of interest are lifelong, and hence it is comparable to a lifelong randomized controlled trial.

#### MATERIALS AND METHODS

Summary statistics for 12 genetic variants associated with selenium levels at genome-wide significance (P < 5 × 10−<sup>8</sup> ) were extracted from a genome-wide association study (GWAS) meta-analysis of circulating selenium levels [n = 5,477; (12)] and toenail selenium levels [n = 4,162; (13)] in European-ancestry individuals. These variants were at two separate genetic loci; 5q14 (9 variants) and 21q22 (3 variants). To analyze the effect of selenium exposure on endometrial cancer risk, we used summary statistics from the Endometrial Cancer Association Consortium (ECAC) GWAS of 12,906 endometrial cancer cases and 108,979 controls of European descent (14). One of the 5q14 seleniumassociated genetic variants, rs558133, was excluded because it was not assessed by the ECAC GWAS (it does not appear on the 1,000 Genomes v3 reference panel) and no proxy with r <sup>2</sup> > 0.8 could be found. These potential instrumental variables were pruned for linkage disequilibrium (LD; r <sup>2</sup> < 0.05) and four seleniumassociated genetic variants (two independent variants per locus) remained as instrumental variables. We used PhenoScanner v2 (15) to explore the possibility of horizontal pleiotropy among the instrumental variables and their highly correlated variants (r <sup>2</sup> > 0.8). Specifically, we examined traits associated with known risk factors of endometrial cancer (i.e., body mass index, age at menarche, age at menopause, postmenopausal serum estradiol levels, nulliparity, infertility, and insulin levels) in the published literature at P < 7.14 × 10−<sup>3</sup> (i.e., 0.05/number of known risk factors explored, n = 7); none of these instrumental variables were associated with these traits.

The reported effect for circulating and toenail selenium instrumental variables was expressed in Z-score units per effect allele. For the purpose of Mendelian randomization analysis, Zscores were converted to beta and standard error values using the following equations, as per Taylor et al. (16), where N is the sample size, eaf is the effect allele frequency, and SE is the standard error of converted beta:

$$\begin{aligned} Beta &= \frac{Z - score}{\sqrt{N}} \times \frac{1}{\sqrt{eqf\left(1 - eaf\right)}}\\ SE &= \frac{Beta}{Z - score} \end{aligned}$$

Converted selenium level summary statistics for these instrumental variables and their association with endometrial cancer risk are shown in **Table 1**. Because summary statistics were expressed in Z-scores, neither the converted beta values for associations of genetic variants with selenium levels nor the effect sizes from the Mendelian randomization analysis have interpretable units, however they do provide the direction and statistical strength of associations.

Individual Wald-type ratios for each of the instrumental variables were determined as a ratio of instrumental variableendometrial cancer regression over the instrumental variableselenium levels regression (17). Individual Wald-type ratios were meta-analyzed using the inverse variance weighted (IVW) approach. A random effect model was used to account for heterogeneity. The IVW approach assumes that instrumental variables do not exhibit horizontal pleiotropy (where a single genetic variant has simultaneous effects on other phenotypes that affect the outcome independently of the exposure of interest) or, if this is violated, that the horizontal pleiotropy is "balanced" across all instrumental variables. Thus, we implemented sensitivity analyses that are more robust to pleiotropy when it is "unbalanced" (i.e., exhibiting directional pleiotropy): (i) weighted

TABLE 1 | Genetic associations with selenium levels and endometrial cancer risk.


\*from hg19; † pairwise LD in Europeans (1000 Genomes) provided for instrumental variables at the same locus; Se, Selenium; EC, Endometrial cancer; EA, Effect allele; OA, Other allele; EAF, Effect allele frequency from each GWAS; Beta, effect size; SE, Standard error; P, P-value. BetaEC and SEEC are the natural log odds ratio of endometrial cancer risk and associated standard error, respectively. Estimates for Selenium levels have been taken from (13) and estimates for EC from (14).



Se, Selenium; EC, Endometrial cancer; Beta, effect size in standard deviation unit; SE, Standard error; P, P value.

median analysis, which provides valid causal estimate even when up to 50% of the weight comes from instrumental variables with horizontal pleiotropic effects (18); and (ii) random effect MR-Egger analysis, which provides valid pleiotropy-corrected causal estimates even if all instrumental variables are invalid (19). MR-Egger analysis corrects for the directional pleiotropy by introducing an intercept which captures the average pleiotropic effects of all included variants on the outcome. An exponentiated MR-Egger intercept that deviates from 1 is an indicator of directional pleiotropy. It should also be noted that the validity of IVW and MR-Egger regression estimates rely on satisfaction of the InSIDE (instrument strength independent of direct effect) assumption where the instrument strength does not correlate with the horizontal pleiotropic effects on the outcome (19).

To assess the strength of the instruments, F statistics and the proportion of variance (R<sup>2</sup> ) in circulating and toenail selenium explained by instrumental variables were calculated as per Rees et al. (20) and Yarmolinsky et al. (21). We used the I 2 GX (22) statistic to assess weak instrument bias for MR-Egger analysis using the "MendelianRandomization" package in R (23). This statistic quantifies the regression dilution bias due to violation of the NO Measurement Error (NOME; genetic associations with exposure of interest are measured without error) assumption. An I 2 GX statistic approaching 1 indicates that violation of the NOME assumption does not substantially dilute the effect estimates of MR-Egger analysis toward a null association. Unless otherwise stated, Mendelian randomization analyses were performed using the "TwoSampleMR" package in R (24).

#### RESULTS

The combined multi-allelic instrument explained 2.9% of the variation in circulating and toenail selenium levels. Individual Wald-type ratios and F statistics for instrumental variables are presented in **Table 2**. F statistics for these instrumental variables were all >10 (range 19.24–44.55) indicating instruments were unlikely to suffer from weak instrument bias. Mendelian randomization analysis did not support an association between selenium levels and endometrial cancer risk using the IVW method (OR per unit increase in selenium levels Z-score = 0.99, 95% CI = 0.87–1.14, P = 0.93). We found limited evidence for heterogeneity amongst the individual causal estimates for the included variants by Cochran's Q statistic (25) (Cochrain's Q statistics = 7.22, P = 0.07). The exponentiated intercept of MR-Egger regression was 1.03 (95% CI = 0.91–1.16, P = 0.72) and therefore provided no evidence of directional pleiotropy across the multi-allelic instrument. Further, the I 2 GX statistic, quantifying weak instrument bias in the context of MR-Egger, was minimal (I 2 GX= 92%). This suggests that any potential bias toward a null association as a result of NOME violation is ≤8%. Association estimates from sensitivity analyses (MR-Egger regression and weighted median analysis) were consistent with that reported by IVW analysis (OR per unit increase in selenium levels Z-score = 0.90, 95% CI = 0.53– 1.50, P = 0.72 for MR-Egger; OR per unit increase in selenium levels Z-score = 0.98, 95% CI = 0.89–1.08, P = 0.70 for weighted median).

#### DISCUSSION

To our knowledge, this is the first Mendelian randomization study evaluating the effect of selenium on endometrial cancer. This analysis does not support a causal relationship between selenium levels and endometrial cancer risk. However, given the fact that the combined multi-allelic instrument explains a small amount of the variance in circulating and toenail selenium levels (<3%), the power to detect a causal association in Mendelian randomization analysis may be limited and thus, we cannot rule out the possibility that genetically predicted selenium levels have some effect on endometrial cancer risk. This analysis should be revisited when more genomewide significant selenium variants are identified from future, larger GWAS studies. Further, statistical power for Mendelian randomization analyses may also be increased through the use of more precise effect estimates from larger GWAS of endometrial cancer.

The validity of Mendelian randomization analysis holds under the condition that three important assumptions are fulfilled. These assumptions require that genetic variants chosen as instrumental variables are:


Our instrumental variables have high F-statistics (>10), thus fulfilling assumption 1. Assumptions 2 and 3 are difficult to validate. We have attempted to minimize violation of assumption 2 by scanning associations of instrumental variables from the literature, finding none of the instrumental variables to be associated with known endometrial cancer risk factors. However, we are limited in exploring this assumption by the GWAS that have been conducted for these risk factors, and we cannot discount the possibility that associations between these variants and unknown endometrial cancer risk factors may exist. Sensitivity testing (by MR-Egger regression and weighted median analysis) has been used to address assumption 3 and we have not found evidence that this assumption has been violated. However, given the limitations of these tests (e.g., the low statistical power of the MR-Egger intercept test, discussed below), we cannot rule out this possibility.

The strengths of our study include incorporation of multiple selenium level-associated genetic variants as a multi-allelic instrument to maximize the variation in selenium levels explained; and use of the largest available GWAS datasets to provide the greatest statistical power possible. Limitations of this study include use of instrumental variables from mixed gender GWAS which were assessed in female-only endometrial cancer GWAS. Although both selenium GWASs controlled for the effect of sex, we cannot not exclude the possibility that there is a residual effect of this covariate which may violate the assumption that instrumental variables are strongly associated with the exposure. A potential limitation of two-sample Mendelian randomization is that by using two different GWAS sample sets to obtain the instrumental variable-exposure and -outcome effect, population stratification may have confounded the observed associations despite all populations being of European descent. Weaknesses of the MR-Egger regression sensitivity analysis performed in our study include its relatively lower statistical power as compared to the IVW and weighted median analysis methods, and its vulnerability to weak instrument bias which may bias MR-Egger regression toward the null (19). However, we assessed the extent to which weak instrument bias may have affected our MR-Egger results using the I 2 GX statistic, and found it to be negligible.

The identification of preventative agents for cancer is an attractive avenue of research because unlike other approaches for disease prevention, such as lifestyle changes, taking a dietary supplement (e.g., selenium) should be considerably easier to implement. Candidate dietary supplements can be identified by observational studies; however, moving these candidates through to human use requires the establishment of expensive randomized controlled trials. For example, a recent prostate cancer prevention trial, examining the benefit of selenium and/or vitamin E supplement on cancer risk, failed because of adverse effects and lack of efficacy, at a cost of >US\$110 million (26, 27); whereas, a subsequent Mendelian randomization study was able to recapitulate the results of this trial using publicly available GWAS data (21).

In conclusion, Mendelian randomization analysis provided no support for selenium supplementation in the prevention of endometrial cancer. More generally, these findings further highlight the value of Mendelian randomization for rapidly excluding proposed interventions that are unlikely to be successful, prior to the initiation of expensive and lengthy trials. This approach could allow resources to be targeted toward trials of alternative interventions with more promising genetic evidence.

# DATA AVAILABILITY

The datasets for this manuscript are not publicly available because Data are available from the authors of the original papers on request. Requests to access the datasets should be directed to Tracy O'Mara, tracy.omara@qimrberghofer.edu.au.

# AUTHOR CONTRIBUTIONS

PK, DG, DT, AS, and TO conception or design of the work; the acquisition, analysis and interpretation of data for the work, drafting the manuscript, and agreement to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. All authors were involved in revision of the manuscript and provide final approval of the version to be published.

# FUNDING

This work was supported by a National Health and Medical Research Council (NHMRC) Project Grant (APP1109286). PK is supported by an Australian Government Research Training Program Ph.D. Scholarship and QIMR Berghofer Postgraduate Top-Up Scholarship, TO is supported by an NHMRC Early Career Fellowship (APP1111246), AS is supported by an NHMRC Senior Research Fellowship (APP1061779).

# ACKNOWLEDGMENTS

Full funding and acknowledgments for ECAC can be found in the **Supplementary Note**.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fonc. 2019.00182/full#supplementary-material

# REFERENCES


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Kho, Glubb, Thompson, Spurdle and O'Mara. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# miRNAs as Candidate Biomarker for the Accurate Detection of Atypical Endometrial Hyperplasia/Endometrial Intraepithelial Neoplasia

Simona Giglio<sup>1</sup> \*, Viviana Annibali 2†, Roberto Cirombella<sup>1</sup> , Omar Faruq<sup>1</sup> , Stefano Volinia<sup>3</sup> , Claudia De Vitis <sup>1</sup> , Margherita Pesce<sup>1</sup> , Donatella Caserta<sup>4</sup> , Angela Pettinato<sup>5</sup> , Filippo Fraggetta<sup>5</sup> and Andrea Vecchione<sup>1</sup> \*

Edited by: Jian-Jun Wei, Northwestern University, United States

#### Reviewed by:

Tamara Louise Kalir, Icahn School of Medicine at Mount Sinai, United States Massimo Nabissi, University of Camerino, Italy

\*Correspondence:

Simona Giglio simona.giglio@uniroma1.it Andrea Vecchione andrea.vecchione@uniroma1.it

> †Present Address: Viviana Annibali, Merck, Rome, Italy

Specialty section:

This article was submitted to Women's Cancer, a section of the journal Frontiers in Oncology

Received: 03 December 2018 Accepted: 30 May 2019 Published: 21 June 2019

#### Citation:

Giglio S, Annibali V, Cirombella R, Faruq O, Volinia S, De Vitis C, Pesce M, Caserta D, Pettinato A, Fraggetta F and Vecchione A (2019) miRNAs as Candidate Biomarker for the Accurate Detection of Atypical Endometrial Hyperplasia/Endometrial Intraepithelial Neoplasia. Front. Oncol. 9:526. doi: 10.3389/fonc.2019.00526 <sup>1</sup> Department of Clinical and Molecular Medicine, "La Sapienza" University, Sant'Andrea Hospital, Rome, Italy, <sup>2</sup> Department of Neurosciences, Mental Health and Sensory Organs, Centre for Experimental Neurological Therapies (CENTERS), "La Sapienza" University, Sant'Andrea Hospital, Rome, Italy, <sup>3</sup> Department of Internal Medicine, Biosystems Analysis, LTTA, Department of Morphology, Surgery and Experimental Medicine, Università Degli Studi, Ferrara, Italy, <sup>4</sup> Department of Medical-Surgical Sciences and Translational Medicine, "La Sapienza" University, Sant'Andrea Hospital, Rome, Italy, <sup>5</sup> Pathology Unit, Cannizzaro Hospital, Catania, Italy

Endometrial cancer is the most common gynecologic malignancy in developed countries. Estrogen-dependent tumors (type I, endometrioid) account for 80% of cases and non-estrogen-dependent (type II, non-endometrioid) account for the rest. Endometrial cancer type I is generally thought to develop via precursor lesions along with the increasing accumulation of molecular genetic alterations. Endometrial hyperplasia with atypia/Endometrial Intraepithelial Neoplasia is the least common type of hyperplasia but it is the type most likely to progress to type I cancer, whereas endometrial hyperplasia without atypia rarely progresses to carcinoma. MicroRNAs are a class of small, non-coding, single-stranded RNAs that negatively regulate gene expression mainly binding to 3′ -untranslated region of target mRNAs. In the current study, we identified a microRNAs signature (miR-205, miR-146a, miR-1260b) able to discriminate between atypical and typical endometrial hyperplasia in two independent cohorts of patients. The identification of molecular markers that can distinguish between these two distinct pathological conditions is considered to be highly useful for the clinical management of patients because hyperplasia with an atypical change is associated with a higher risk of developing cancer. We show that the combination of miR-205, −146a, and −1260b has the best predictive power in discriminating these two conditions (>90%). With the aim to find a biological role for these three microRNAs, we focused our attention on a common putative target involved in endometrial carcinogenesis: the oncosuppressor gene SMAD4. We showed that miRs-146a,−205, and−1260b directly target SMAD4 and their enforced expression induced proliferation and migration of Endometrioid Cancer derived cell lines, Hec1a cells. These data suggest that microRNAs-mediated impairment of the TGF-β pathway, due to inhibition of its effector molecule SMAD4, is a relevant molecular alteration in endometrial carcinoma development. Our findings show a potential diagnostic role of this microRNAs signature for the accurate diagnosis of Endometrial hyperplasia with atypia/Endometrial Intraepithelial Neoplasia and improve the understanding of their pivotal role in SMAD4 regulation.

Keywords: microRNAs, endometrial hyperplasia, endometrial cancer, biomarkers, SMAD4, TGF-β pathway

#### INTRODUCTION

Worldwide, endometrial cancer (EC) represents 4% of all cancers in women and is the most common malignant tumor of the female genital tract in industrialized countries (1). The etiology of EC is not yet fully understood, although there is some evidence that molecular modifications and hormonal influences contribute to its initiation and progression (2).

Endometrial cancer is divided into two major classes: estrogen-dependent tumors (type I, endometrioid endometrial carcinomas) that represent 80% of cases and non-estrogendependent (type II, non-endometrioid endometrial carcinomas) that account for the rest. EC type I is thought to develop via precursor lesions along with the increasing accumulation of molecular genetic aberration (3, 4).

Recently, the World Health Organization classification of tumors (WHO) (5) divided endometrial hyperplasias into two categories: hyperplasia without atypia (Benign Hyperplasia, BH) and atypical hyperplasia/endometrioid intraepithelial neoplasia: (AH/EIN) (5).

Indeed, AH/EIN is most likely to progress to type I endometrial carcinoma (∼30%), and has been reported to be associated with invasive EC in 62% of endometrial biopsy (6), whereas BH rarely progresses to EC (<5%) (7). Therefore, discerning between these two entities has significant clinical implications (8).

Unfortunately, recognition of atypia in endometrial hyperplasia is subjective among pathologists with a low inter observer reproducibility (<50% in almost all studies) (9, 10).

Although the recent two-tier classification of these entities by WHO (5, 11) improved reproducibility, management of endometrial pre-cancers is compromised by a longstanding debate.

MicroRNAs (miRs) are a class of small, non-coding, singlestranded RNAs that negatively regulate gene expression mainly binding to 3′ -untranslated region (UTR) of target mRNAs at the post-transcriptional level (12). Several studies showed that they are important in many biological processes, thus their aberrant expressions are closely associated with the development, invasion, metastasis, and prognosis of various cancers, including EC (13–17).

Up until now, several miR signatures have been documented in either normal or neoplastic endometrium, but the role of miRs in endometrial hyperplasia with or without atypia remains poorly understood (18–22). In the present study, we investigate the hypothesis that changes in miRs may represent useful biomarkers for the diagnosis of AH/EIN.

# MATERIALS AND METHODS

#### Endometrial Tissue Samples and Patients

Eighty-five archived formalin-fixed, paraffin embedded (FFPE) tissue blocks of BH (41 cases), and AH/EIN (44 cases), were obtained from the Pathology Department of Sant'Andrea Hospital and Ospedale Cannizzaro, Catania from 2004 to 2013. Patient's age ranged from 37 to 84 years, with a median of 56 years. This study was authorized by the institutional ethics committee board at S. Andrea Hospital Rome, Italy (Aut. #168/03). Written informed consent was obtained from all patients enrolled.

The selected cases were randomly divided into a training set (23 BH, 19 AH/EIN) and into a validation set (21 BH, 22 AH/EIN).

Hyperplasia was macro or laser-microdissected, were appropriate, for this study.

#### RNA Extraction

Total RNA, including miRs fraction, was extracted from FFPE tissues using the High Pure miRNA isolation kit (Roche) according to the manufacturer's instructions. RNAs concentration were assessed using Nanodrop (ThermoScientific).

#### Affymetrix Gene Chip miRNA Array

RNA quality and purity were assessed with the use of the RNA 6000 Nano assay on Agilent 2100 Bioanalyzer (Agilent). Briefly, 500 ng of total RNA was labeled using FlashTag Biotin HSR (Genisphere LLC) and hybridized to GeneChip <sup>R</sup> miRNA 2.0 Arrays. The arrays were stained in the Fluidics Station 450 and then scanned on the GeneChip <sup>R</sup> Scanner 3000 (Affymetrix, USA).

#### Microarray Data Analysis

The statistical analysis was performed by Transcriptome Analysis Console (TAC) software (Thermo Fisher Scientific).

To survey outliers that could disturb the dataset, a Principal Component Analysis (implemented by means of R statistical software) was performed and its visualization, which led to the knowledge of which subjects needed to be excluded from the dataset. MicroRNA probe outliers were defined from the manufacturer's instructions (Affymetrix, USA), and further analysis included data summarization, normalization, and quality control using the web-based miRNA QC Tool software (Affymetrix).

The microarray data has been submitted and assigned a GEO omnibus accession number GSE85105.

#### Reverse Transcription and Quantitative Real-Time PCRs

Each sample was reverse-transcribed using miRNA miRCURY LNA Universal RT kit (Exiqon) according to the manufacturer's protocol.

Reverse transcription and quantitative real-time PCRs were performed for miRNAs using miRCURY LNA Universal RT microRNA PCR LNA primers set with miRCURY LNA cDNA Synthesis Kit II and ExiLENT Syber Green master mix, in triplicate (Exiqon). RNU48 (U48) was used to normalize input total small RNA.

Expression of each miR was presented as the ratio between miR and RNU48 (RQ). The relative miRs expression was calculated using the 11Ct method. At least three separate experiments were performed, and each sample was assayed in triplicate.

#### Cells Transfections

Pre-designed Pre-miR (miR Precursors) for each miR was obtained from Ambion (ThermoFisher Scientific). A negative-control miRNA mimic [Pre-miR miRNA negative #1 (Ambion, ThermoFisher Scientific)] was used to address the specificity of the observed effect to the specific miR sequence. Cells were transfected using Lipofectamine RNAimax (ThermoFisher Scientific).

#### Protein Extraction, Western Blotting, and Antibodies

Hec1a cell lines were obtained from ATCC (Atcc, HTB-112) and cultured according to the manufacturer's protocol. Total cell extracts with RIPA buffer (Sigma Aldrich) were collected at 24 h and analyzed by western blot to assess proteins expression levels. Briefly, Hec1a cells were rinsed in ice-cold PBS and subsequently lysed in ice-cold RIPA lysis buffer (Sigma Aldrich) and Complete inhibitor (Roche). Proteins were analyzed on precast polyacrylamide gel (Bio-Rad), transferred onto nitrocellulose membranes (Bio-Rad) and blocked with 5% BSA (Sigma Aldrich), and incubated with specific primary antibodies.

Polyclonal antibody against SMAD4 (sc-7966, Santa Cruz Biotechnology) diluted 1:200. Monoclonal antibody against PAX2 (aJ1589a, Abgent) diluted 1:1,000. Monoclonal antibody against Pten (560002, BD Biosciences) diluted 1:1,000.

To normalize protein loading, membranes were probed for 1 h at room temperature with an anti-vinculin antibody (sc- 25336, Santa Cruz Biotechnology).

Secondary antibodies (labeled HRP anti-rabbit or anti-mouse, Bio-Rad) were incubated for 45 min at room temperature and revealed with chemiluminescent ECL method (Bio-Rad).

Digital images of autoradiography were acquired with ChemiDOC XRS (Bio-Rad).

#### Plasmids and Constructs

The 3′UTR of the SMAD4 gene was obtained from GeneArt Gene Synthesis (Invitrogen, Thermo Fisher Scientific) by cloning 900bp of SMAD4 3′UTR into pMir-vector (Promega) giving rise to the pMir-3′UTRSMAD4 construct.

Site-direct mutagenesis into the miR-205, miR146a, and miR-1260b binding sites of the SMAD4 gene 3′UTR were introduced using GeneArt Site-Directed Mutagenesis PLUS System kit (Thermo Fisher Scientific) according to manufacturer's instructions. The primers used were:

FP 5′CTTCACCTGTTATGTAcctgccAATCATTCCAGT GC3′

RP 5′GCACTGGAATGATTggcaggTACATAACAGGTG AAG3′

FP 5′GCTGATTTTAAAGGCAGAGAAccgtcgAAAGTTA ATTCACC3′

RP 5′GGTGAATTAACTTTcgacggTTCTCTGCCTTTA AAATCAGC3′

FP 5′GTTATTCCTAGTGacccgtTGTTGATGAAGTAT ACTTTTCCCC3′

RP 5′GGGGAAAAGTATACTTCATCAACAacgggtCACTA GGAATAAC3

#### Luciferase Activity Assays

Hec1a cells were cultured in 12-well-plates and transfected with 500 ng of pMir-3′UTRSMAD4 wt or mutated plasmid or pMir control vector together with 50 ng of β-GAL vector and 50 pmoles of pre-miR-205, pre-miR-146a, premiR-1260b, or pre-miR-negative control#1 (Thermo Fisher Scientific). Transfections were carried out using Lipofectamine 2000 and OPTI-MEM as recommended by the manufacturer (Thermo Fisher Scientific). At 48 h after transfection, luciferase activity was measured using the Luciferase Reporter Assay (Promega). Each transfection was repeated twice in triplicate. Transfection efficiency was corrected to β-GAL expression in all cases.

#### Cell Proliferation Assay

Cell proliferation was measured using Muse <sup>R</sup> Count & Viability Assay Kit and Muse <sup>R</sup> Cell Analyzer as recommended by the manufacturer (Merck Millipore). Cells were transfected with each pre-miR and the pre-miR-negative control#1 into a 35 mm dish as described below and incubated for 72 h. Three independent experiments were performed in duplicate.

#### Transwell Migration Assay

The migration ability of Hec1A cells was determined in a Boyden Chamber. Twenty-four hours after transfection, the cells were seeded into 8µm Transwells (6.5 mm diameter, Corning) at 5 × 104 cells well with serum-free culture medium. Medium containing 10% FBS was added into the lower chamber and served as the chemoattractant. After incubation for 24 h, the cells remaining on the upper surface of the filter were removed by gently wiping with a cotton swab. The cells migrated through the filter were fixed with methanol, stained with MGG quick staining (Bio Optica), and visualized by an inverted fluorescence microscope.

#### Statistical Analysis

Data were expressed as the mean ± SD from at least three independent experiments. Statistical analysis between two samples was performed using Student's t-test. Statistical comparisons of more than two groups were performed using one-way analysis of variance (ANOVA).

The diagnostic ability of miRs-205, −146a, and −1260b in diagnosing atypical hyperplasia was examined via the area under the corresponding receiver operating characteristic curve (AUC). All statistical analyses were performed using Graph Pad Prism software (GraphPad software). p < 0.05 was considered statistically significant.

#### RESULTS

#### Dysregulated miRs in BH vs. AH/EIN

To investigate whether miRs could discriminate BH from AH/EIN, we analyzed the expression of 1,105 human miRs (miRbase version 15) in the training set (23 BH, 19 AH/EIN). Examples of BH and AH are presented in **Figures 1A,B**.

As shown in **Figure 1C**, we could identify 14 differentially expressed miRs capable of discriminating BH from AH/EIN (FC ≥ 1.5, p ≤ 0.05). In particular, 13 miRs were upregulated (miRs-205,-146a, −200b\_star, −1274a, −1260b, −200b, −200a, −192, −183, −10, −194, and −200a\_star) and 1 (miR-379) was downregulated in AH/EIN compared to BH samples.

Using multiple logistic regression, the statistical significant variables (Age, BMI, Parity, miR values) were assessed in univariate analysis and investigated comparing BH to AH/EIN. No significant correlation was observed (data not shown).

Differentially expressed miRs were then validated in an independent validation set (21 BH, 22 AH/EIN). Out of the 14 miRs initially identified, we could confirm three miRs all up-regulated (miR-205, −146a, and −1260b) (**Figure 2A**), suggesting that these miRs could discriminate between the two groups.

To assess the ability of each miR to differentiate between AH/EIN and BH, receiver-operating characteristic curves (ROC) were constructed and the area under the curve (AUC) was calculated. Univariate analysis for each individual miR showed an AUC of 0.8 [95% confidence interval (CI) = 0.66–0.93 p = 0.0009] for miR-205, an AUC of 0.8 (95% CI = 0.68–0.94 p =

0.0008) for miR-146a, and an AUC of 0.9 (95% CI = 0.88–1.01 p < 0.0001) for miR-1260b, respectively (**Figure 2B**). Performing a multivariate analysis for the combination of the three miRs, we observed an AUC of 0.95 (95% CI 0.88–1.01 p < 0.0001) showing that these three miRs have a high predictive power in discriminating AH/EIN from BH (**Figure 2C**).

#### SMAD4 Is a Target of miRs-205, 146a, and 1260b

To investigate a biological role for these miRs in AH/EIN, we searched different prediction algorithms. We found that highly conserved binding sites for each of these miRs were present in the mRNA of the oncosuppressor gene SMAD4, which has been shown to be down-modulated in EC (23).

We used endometrial cancer-derived cell lines Hec1a and tested the endogenous expression of these miRs.

QRT-PCR analysis showed that Hec1a cells expressed detectable amounts of each miRs (**Figure 3A**). To investigate the effects of these miRs on SMAD4 expression, we transiently transfected pre-miRs-146a, −205, and −1260b or control into Hec1a cells.

In miRs transfected cells, we observed a significant suppression of SMAD4 compared to control (**Figure 3B**). To further confirm this observation, we transiently transfected

efficiency of Hec1a cells transfected in B. Each bar shows miR expression normalized to RNU48 ±SD of three independent experiments.

Hec1a cells with antagomirs. This resulted in an increase of SMAD4 levels, compared to control (**Figures 3B,C**), as expected.

Co-loss of both PTEN and PAX2 has been reported in AH/EIN and it has been regarded as the reference markers for its diagnosis (24).

Therefore, we analyzed Pax2 levels by Western blot and we observed a strong protein reduction in miR-1260b transfected cells. There is no evidence of interaction between these two players, but miR binding sites algorithm prediction showed that other Pax family members are the putative target of miR-1260b. We also found that pTEN was downregulated in miR-205 transfected cells confirming previous evidence showing the potential role of miR-205 in regulating PTEN in endometrial tissue (25).

Although by different means, Smad4 could be a target of these miRs, one can argue that miR-146a, miR-205, and miR-1260b interact with other unknown targets that downregulate Smad4 protein levels (18–22) (**Figure 4A**). To address this concern, we performed a luciferase reporter assay, cloning a 900 bp 3′UTR of human SMAD4 into a pMIR vector (p3′UTRSmad4pMir). Therefore, Hec1a cells were transfected either with miR-146a, miR-205, mir-1260b, or the pre-miRcontrol, and p3′UTRSmad4pMir vector. As shown in **Figure 4B**, all miRs decrease luciferase activity of the p3′UTRSmad4pMir compared to control, showing that each miR has a direct effect on their target in this cell lines (**Figure 4B**). To determine this direct miR-target interaction, we constructed a plasmid with mutagenesis of the three seed sequences (**Figure 4C**). As expected, we observed only a slight effect on luciferase activity when we compared the wild-type vector with the p3′UTRSmad4pMir mutants in the presence of each miRs overexpressed, showing that the modification of the seed sequence is enough to block the function of each miR (**Figure 4C**).

#### SMAD4 Repression by miRs-205, 146a, and 1260b Induces Proliferation and Migration in Hec1a Cell Lines

Smad4 is involved in the signal transduction pathway of the transforming growth factor ß (TGF-ß) that acts as a tumor suppressor gene in several cancers (26, 27). To gain further insights into how dysregulation of these miRs may play a role in endometrial cancer cells, we performed different assays to study the biological effects of the interaction between these miRs and their target Smad4 into endometrial adenocarcinoma-derived cell lines Hec1a.

First, we tested cells proliferation. Cells transfected with pre-miRs-146a, −205, or −1260b showed a higher rate of proliferation compared with non-treated or control transfected cells (**Figure 5A**). Cells transfected with anti-miRs-146a, −205, or −1260b showed a reduction of proliferation compared to control (**Figure 5B**).

To better understand how the dysregulation of these miRs may change the behavior of endometrial cancer-derived cell line, we examined the influence of Smad4 knockdown on Hec1a cell migration. We found a significant increase in the migration capability in pre-miR-205, −146a, or −1260b transfected cells compared to control, showing that these miRs positively regulate the migration of cultured endometrial cancer cells (**Figures 5C,D**).

Since miRs may have multiple targets, to ensure that Smad4 mediated the observed effects, we repeated the assay using a specific Smad4 siRNA.

We confirm significant down-regulation of Smad4 through qRT-PCR (**Figure 5E**). Next, we observed that Hec1a cells transfected with Smad4 siRNA displayed a higher proliferation and migration rate respect to controls (**Figures 5C,F**).

#### DISCUSSION

Identification of molecular markers that can differentiate between AH/EIN and BH are considered to be highly useful for clinical management of patients because hyperplasia with atypical change and/or Endometrial Intraepithelial Neoplasia are associated with a higher risk to progress to cancer (4).

Since there aren't reference markers, the diagnosis is based only on histological features, such as the presence of nucleoli and other atypical characteristics, which are not consistently associated with that diagnosis (9).

Although the new WHO classification is more likely to successfully identify premalignant lesions, the low interobserver reproducibility among gynecological pathologist in diagnosing atypical hyperplasia/EIN should be improved (6, 10, 28–30).

Atypical endometrial hyperplasia/EIN and EC shares several molecular alterations with each other, including microsatellite instability, PAX2 inactivation, mutation of PTEN, KRAS, and CTNNB1 (β-catenin), but there is not a linear accumulation of mutational events leading to cancer (31). Identifying the diseaserelated miRs will improve the diagnosis and understanding of pathogenesis of these lesions.

Since over 50% of miRs reside in cancer-associated genomic regions, they have been indicated to play an important role as diagnostic biomarkers (13, 32, 33).

Most of the miRs studies on endometrium have been focused on the identification of their implications in EC development, almost neglecting their possible diagnostic role in precursor lesions (34). In fact, several authors showed an altered expression of miRs that may discriminate EC from non-atypical or atypical hyperplasia (9, 14, 21, 35–40).

In particular, expression of five miRs (miRs-182, 183, 200a, 200c, and 205) was significantly higher in EC when compared with complex atypical hyperplasia, simple hyperplasia (SH) and normal endometrial tissue (P < 0.05, respectively) (41).

To our knowledge, our study is the first to identify a miRs signature able to discriminate between atypical hyperplasia/EIN and benign endometrial hyperplasia with the capability to better distinguish between low- and high- risk lesions. Identification of miR-target genes and pathways to understand the molecular basis of endometrial cancer pathogenesis is a major challenge, as there are numerous pathways that drive cancer. Accordingly, in this study, we proposed a novel miR-based classification method to categorize the high risk pre-cancerous endometrial lesions.

In fact, we showed a high predictive power, above 90%, using a three miRs-signature (miRs-146a, −205, and −1260b) in distinguishing between non-atypical and atypical

hyperplasia/EIN provides a supplementary diagnostic tool when required.

Interestingly, a previous study conducted by Snowdon and colleagues examined a miRs profile in atypical hyperplasia compared to normal proliferative controls. The microarray expression profile shares some important similarities with our data. MiRs-146a, miR-200a, miR-200b, miR-200b-star, and miR-205 resulted up-regulated and miR-542-5p down-regulated in

FIGURE 5 | SMAD4 regulation by miRs-205, −146a, and −1260b induces proliferation and migration in Hec1a cells. (A,B) Hec1a cells proliferation was measured at 72 h. Cells were transiently transfected with negative control #1 (C) and miRs-205, −146a, −1260b in A or anti-miRs-205, −146a, and −1260b (a-miRs) in B. (C) Transwell migration assay shows that overexpression of miRs-205, −146a, −1260b, and siSMAD4 enhance cell migration ability of Hec1a cells. (D) Representative photographs of the Transwell migration assay is shown. All data are presented as mean ± SD, \*p < 0.05. (E,F) Effects of siSMAD4 expression in Hec1a cells. (E) RT-PCR (upper panel) and western blot analysis (lower panel) are shown. LC (loading control, β-actin). PCR bars depict SMAD4 expression in control (ctrl) and si transfected cells normalized to β-actin. (F) Measure of Hec1a cells proliferation in siSMAD4 and SiCTRL transfected cells is reported. Data represent the mean (from three independent experiments) ±SD. \*P < 0.05.

atypical hyperplasia vs. normal endometrium, adding further emphasis to our results (22). The up-regulation of miR-200 family members and miR-205 in EC is observed among different studies, indicating that these miRs may play a role in driving oncogenesis in the endometrium (21, 35, 37, 42, 43).

Some authors showed that miR-205 is a negative prognostic marker for EC and its levels were significantly increased in endometrial cancer cell lines and endometrial tumors compared to normal tissues (25, 26). Mir-205 is directly involved in PTEN regulation that represents one of the most commonly investigated markers implicated in endometrial tumorigenesis (44). Our results have enforced this effect indicating a remarkable influence of miR-205 on regulating essential target genes involved in different signal pathways in endometrial cells. On the other hand, Lacey et al. have shown that a loss of expression of PTEN status was not associated with progression risk of endometrial hyperplasia (45).

Even if EC seems to be characterized by elevated expression of miR-205, a recent study conducted by Wilczynski showed that higher levels of miR-205 may be a marker of an early stage disease and is associated with a more favorable prognosis, whereas patients with lower levels of miR-205 had worse survival (42).

No evidence has revealed the dysregulation of miRs-1260b and −146a in EC, suggesting that up-regulation of these miRs may be specific of AH/EIN.

Interestingly, authors found that a single nucleotide polymorphism (SNP) rs2910164 G>C within miR-146a is associated with the increased risk of gastric cancer and papillary thyroid carcinoma (46, 47). A recent publication showed that overexpression of miR-146a inhibited cell proliferation, enhanced apoptosis, and increased sensitivity to chemotherapy drugs in epithelial ovarian cancers cells showing, therefore, that the role of miR-146a is still to be elucidated (48).

MiR-1260b has been found to be highly expressed in the prostate, renal cell, and in colorectal carcinomas (49, 52). Recently, it was demonstrated that in Hepatocellular carcinoma, MiR-1260b promotes cell migration and invasion through the G-protein signaling 22 (50).

All of this evidence confirmed that our identified miRs regulate genes involved in different signal pathways that may trigger the endometrial cellular transformation.

Thus, we investigated a possible common pathway that could be regulated by these miRs and could be implicated in cellular transformation, and we found that they target SMAD4. Also noteworthy are several other reports, which demonstrated a direct interaction of these miRs and this transcript (51–53).

Smad4 is a gene implicated in several cancers, including EC, albeit its role in endometrial carcinogenesis is yet not clear (54–56).

Impairment of the Smad pathway results in escape from growth inhibition and leads to the promotion of cell proliferation, contributing to carcinogenesis (57).

The disturbances in Smad proteins expression and/or differences in their intracellular distribution, that trigger a TGF-β signaling pathway deregulation, it was reported in endometrial carcinomas, but it is still not well-understood (58). The region within 18q21 where Smad4 is located is frequently deleted in endometrial carcinomas, showing its involvement in EC, however, an immunohistochemical study showed that inactivation of this gene occurs infrequently in this tumor.

Changes in the expression of the TGF-β signaling cascade in type I ECs seem to be associated mainly with deregulation of TGF-β receptors and SMAD expression at the protein level, indicating SMAD4 as a central molecule of this pathway (59, 60).

Thus, the potential pathogenic role of SMAD4 in endometrial hyperplasia is supported by our finding, albeit further studies are required to understand its biological and diagnostic role in this environment.

Our results clearly demonstrated that overexpression of miRs-146a, −205, and −1260b induced Hec1a proliferation and migration through SMAD4 inhibition, providing an insight into the possible mechanisms underlying the function of these miRs in endometrial hyperplasia.

Thus, our work highlights the relevance of miRs in regulating cellular processes that may ultimately lead to tumorigenesis.

Taken together these results strongly show that miRs-205, −146a, and −1260b contribute to enhancing proliferation and migration properties of endometrial cancer cells through Smad4 inhibition.

In conclusion, distinguishing between hyperplasia and true pre-cancerous lesions has significant clinical implications because distinct endometrial pre-cancerous conditions require intervention. Thus, we proposed a three miR-signature (146a −205, −1260b) as a potential biomarker for diagnosis of atypical endometrial hyperplasia/EIN that could have a significant impact on treatment decisions. Furthermore, the regulatory capability of these three miRs on cell proliferation and migration, possibly through impairment of TGF-β signaling Smad4-mediated, highlights their crucial role in endometrial hyperplasia outcome.

Although we believe that this study represents a step forward in investigating the molecular relationship between miR deregulation and EIN lesions, we analyzed a relatively small group of patients, and therefore, a prospective analysis is needed to strengthen the accuracy of our results.

# ETHICS STATEMENT

This study was carried out in accordance with the recommendations of name of guidelines, name of committee with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the IRB committee.

# AUTHOR CONTRIBUTIONS

SG and AV: study conception and design and drafting the manuscript. SG, VA, RC, OF, CD, and MP: methodology. SG, SV, and AV: analysis and interpretation of data. DC, AP, FF, and AV: access to clinical data. SG, DC, AP, FF, and AV: critical revision.

# FUNDING

This work was supported by grants from Associazione Italiana per la Ricerca sul Cancro (AIRC IG 16862 to AV) and La Sapienza University Fondi Ateneo 2017 to AV.

#### Giglio et al. miRNAs as Biomarker of AH/EIN

#### REFERENCES


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Giglio, Annibali, Cirombella, Faruq, Volinia, De Vitis, Pesce, Caserta, Pettinato, Fraggetta and Vecchione. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Long Non-coding RNA CDKN2B Antisense RNA 1 Gene Contributes to Paclitaxel Resistance in Endometrial Carcinoma

#### Chao Shang<sup>1</sup> , Cheng N. Ao<sup>2</sup> , Chi C. Cheong<sup>2</sup> and Lirong Meng<sup>2</sup> \*

*<sup>1</sup> Department of Neurobiology, School of Life Science, China Medical University, Shenyang, China, <sup>2</sup> School of Health Sciences, Macao Polytechnic Institute, Macau, China*

Endometrial cancer (EC) is the most common malignancy of the female reproductive tract. In this study, we clarified the clinical significance of CDKN2B antisense RNA 1 (CDKN2B-AS) gene, and its effects on paclitaxel sensitivity in EC. Firstly, CDKN2B-AS gene was highly expressed in EC tissues and cell lines. The high-expression of CDKN2B-AS gene was associated with high pathological grade and low paclitaxel sensitivity of EC tissues. Knockdown of CDKN2B-AS gene sensitized Ishikawa/PA and HEC1A/PA cells to paclitaxel, and promoted paclitaxel-induced cytotoxicity. Secondly, the low-expression of miR-125a-5p was closely associated with low paclitaxel sensitivity of EC cells, and up-regulation of miR-125a-5p could increase paclitaxel sensitivity of Ishikawa/PA and HEC1A/PA cells. MiR-125a-5p also mediated the suppressive effects of knockdown of CDKN2B-AS on paclitaxel resistance in EC cells. Thirdly, B-cell lymphoma-2 (Bcl2) and Multidrug Resistance-Associated Protein 4 (MRP4) genes were target genes of miR-125a-5p, which modulated paclitaxel resistance of Ishikawa/PA and HEC1A/PA cells through targeted silencing Bcl2 and MRP4. In conclusion, high-expression of CDKN2B-AS is associated with a poor response to paclitaxel of EC patients, and knockdown of CDKN2B-AS inhibits paclitaxel resistance through miR-125a-5p-Bcl2/MRP4 pathway in EC patients. Our findings help elucidate the molecular mechanisms of chemoresistance in EC patients.

#### Edited by:

*Massimo Nabissi, University of Camerino, Italy*

#### Reviewed by:

*Amanda Lee Hudson, Bill Walsh Translational Cancer Research Laboratory, Australia Luigi Formisano, Vanderbilt University Medical Center, United States*

> \*Correspondence: *Lirong Meng lrmeng@ipm.edu.mo*

#### Specialty section:

*This article was submitted to Women's Cancer, a section of the journal Frontiers in Oncology*

Received: *11 August 2018* Accepted: *10 January 2019* Published: *29 January 2019*

#### Citation:

*Shang C, Ao CN, Cheong CC and Meng L (2019) Long Non-coding RNA CDKN2B Antisense RNA 1 Gene Contributes to Paclitaxel Resistance in Endometrial Carcinoma. Front. Oncol. 9:27. doi: 10.3389/fonc.2019.00027* Keywords: non-coding RNA, endometrial carcinoma, CDKN2B antisense RNA 1, miR-125a-5p, chemotherapy, paclitaxel

# INTRODUCTION

Endometrial cancer (EC) is one of the most common malignancy of the female reproductive tract and is increasing in incidence. The mortality of EC is next to ovarian and cervical cancer, and that is currently increasing year-by-year (1, 2). Chemotherapy is extensively used for treatment of EC, and it can significantly improve the prognosis and inhibit the recurrence and metastasis (3). Drug resistance reduces the sensitivity to chemotherapeutic drugs, and contributes a barrier, leading to treatment failure of EC. Accordingly, that is pivotal to identify the therapeutic target, re-sensitizing EC to chemotherapeutic drugs and its underlying mechanism. Several scholars have proposed diverse hypotheses, accounting for chemotherapy failure of EC patients, including DNA repair deregulation, aberrant function of efflux pumps, imbalance of signaling pathway, and so on Brasseur et al. (4), Shang et al. (5), and Liu et al. (6).

**72**

In recent years, non-coding RNAs (ncRNAs) have become a hotspot in the research of life science, especially in oncology, involving long non-coding RNAs (lncRNAs) and short noncoding RNA (e.g., microRNAs). Recent studies reported that some ncRNAs could be prognostic and diagnostic biomarkers, as well as being therapeutic targets for tumors (7–9). Increasing evidences have indicated that ncRNAs are involved in formation and progress of chemotherapy resistance, including EC (10–12). Our previous study revealed that Tumor Suppressor Candidate 7 (TUSC7) gene could specifically combine and silence miR-23b, and then inhibit the resistance of cisplatin and taxol in EC (12). Our preliminary experiments showed that CDKN2B antisense RNA 1 (CDKN2B-AS) gene was associated with paclitaxel resistance of EC, thus, we attempted to study the molecular mechanism of CDKN2B-AS triggered regulation for chemotherapy resistance of EC.

CDKN2B antisense RNA 1 (CDKN2B-AS) gene is a lncRNA gene identified by Pasmant et al. during the genetic study of a melanoma-neural system tumor family in 2007, that also named Antisense Non-coding RNA In The INK4 Locus (ANRIL) (13). Recent studies revealed that the CDKN2B-AS gene was up-regulated and acted as an oncogene in several malignant tumors, such as breast cancer, prostate cancer, and cervical cancer (14–17). At present, there is no report on CDKN2B-AS gene associated with EC. It is well-known that lncRNA could be associated with microRNA to regulate its expression and function. This study demonstrated that CDKN2B-AS could specifically silence miR-125a-5p expression of in EC cells.

MiR-125a-5p originates from 5′ end of pre-miR-125a and belongs to miR-125 family. Accumulating evidences found that miR-125a-5p was down-regulated in a variety of tumors and participated in the tumorigenesis and malignant progression of tumors by regulating its target genes to play the role of a tumor suppressor gene (18–20). The expression and potential function of miR-125a-5p in EC patients have still remained unclear. Our findings confirmed that B-cell lymphoma-2 (Bcl2) and Multidrug Resistance-Associated Protein 4 (MRP4) were the target genes of miR-125a-5p.

It is widely accepted that B-cell lymphoma-2 (Bcl2) gene is an important member of Bcl2 family and can inhibit apoptosis and promote cell survival, while Bcl2 has abnormal expression or function in almost all tumors. Defect to the Bcl-2 gene has been identified as a cause of resistance to cancer treatments. Bcl2 was negatively regulated by lncRNA GAS5 and contributed to doxorubicin resistance of bladder transitional cell carcinoma (21). Notch3-specific inhibition reduced the expression of Bcl2 and reversed paclitaxel resistance of ovarian cancer (22). Chon et al. reported that knockdown of Bcl2 antagonist of cell death (BAD) pathway increased the cisplatin resistance of Ishikawa and HEC1-A cells (23). However, it is not clear enough whether Bcl2 is involved in paclitaxel resistance of EC.

MRP4 gene belongs to ATP-binding cassette (ABC) transporters superfamily, that also named ATP Binding Cassette Subfamily C Member 4 (ABCC4). MRP4 can expel the chemotherapeutic drugs from the cells before they work, so as to reduce the damage of the chemotherapeutic drugs to the cells. (24). Overexpression of MRP4 mediated the acquired docetaxel resistance, targeting MRP4 treatment which resensitized docetaxel-resistant prostate cancer cells to docetaxel chemotherapy (25). The roles of MRP4 on chemotherapy resistance of EC need to be further studied.

Therefore, this study explored the clinical significance of CDKN2B-AS expression, and clarified the mechanism of CDKN2B-AS, contributing to paclitaxel resistance through miR-125a-5p/Bcl2&MRP4 pathway in EC.

# MATERIALS AND METHODS

#### Clinical Specimens

In this study, 87 cases of EC patients were diagnosed and treated at Shengjing Hospital of China Medical University (Shenyang, China) from October 2015 to November 2016. The paracancerous normal endometrium tissue (PNET) and EC tissue specimens were obtained through hysteroscopy accompanied with biopsy, and diagnosed by two pathologists, in which the complete clinical data were collected as well. All the patients were not treated with radiotherapy or chemotherapy before diagnosis. The EC tissues specimens without treatment were used to detect the expression level of CDKN2B-AS and miR-125a-5p in EC patients. After diagnosis, all EC patients were treated with paclitaxel. After two or three cycles of chemotherapy, the curative effects were verified according to hysteroscopy with biopsy and imaging detection, and all EC patients were divided into two groups, including sensitive group (n = 36) and insensitive group (n = 51).

This study was conducted in accordance with the Declaration of Helsinki, and was approved by the Ethics Committee of Shengjing Hospital of China Medical University, and written informed consent was obtained from all participants as well.

### Cell Lines and Culture

Human endometrial cell lines (HEC-251), human EC cell lines (Ishikawa, HEC-1A), and human embryonic kidney cell lines (HEK293T) were obtained from the Cell Resource Center of Chinese Academy of Medical Sciences (Beijing, China). Paclitaxel-resistant EC cell lines (Ishikawa/PA and HEC1A/PA cell lines) were set up previously from parental cell lines (Ishikawa, HEC-1A), and stored in our laboratory (12). Those cells were cultured in Dulbecco's modified Eagle's medium (DMEM), containing 10% fetal bovine serum (FBS; Shanghai ExCell Biology, Inc., Shanghai, China) in a 95% air/5% CO<sup>2</sup> incubator at 37◦C.

#### Quantitative Reverse Transcription Polymerase Chain Reaction (RT-qPCR)

Total RNA was extracted using TRNzol reagent (TIANGEN, Beijing, China) and reversely transcribed into cDNA using lnRcute lncRNA First-Strand cDNA Synthesis Kit (TIANGEN, Beijing, China). The expression level of CDKN2B-AS was examined using an lnRcute lncRNA qPCR Detection Kit (TIANGEN, Beijing, China) in accordance with manufacturer's instructions. The sense primer of CDKN2B-AS was 5′ -TGCTCT ATCCGCCAATCAGG-3′ and its antisense primer was 5′ -GGG CCTCAGTGGCACATACC-3′ (26), in which the specificity was checked, that could not be used to amplify CDKN2B gene. The expression level of miR-125a-5p was examined with Taqman Universal Master Mix II (Life Technologies, Carlsbad, CA, USA). The relative expression levels of CDKN2B-AS and miR-125a-5p were calculated using 2−11CT method after normalization with reference genes (β-actin and U6).

#### Cells Transfection

The inhibitor of CDKN2B-AS (smart silencer-CDKN2B-AS, ss-CDKN2B-AS) and its negative control (ss-NC) were designed and synthesized by Ribobio Co. (Guangzhou, China), and transfected into EC cells via HiPerFect reagent (QIAGEN, Hilden, Nordrhein-Westfalen, Germany) in a 6-well-culture plate in accordance with the manufacturer's instructions. The stable transfected cells were selected using Geneticin (Sigma-Aldrich, St Louis, MO, USA).

The agonist and antagonist of miR-125a-5p (agomiR-125a-5p and antagomiR-125a-5p), as well as their negative controls (agomiR-NC and antagomiR-NC) were synthesized by GenePharma Co. Ltd. (Shanghai, China). The expression plasmid of Bcl2 and MRP4 (pUC-Bcl2 and pUC-MRP4) and their negative control (pUC-NC) were synthesized by Cyagen Inc. (Santa Clara, CA, USA). The microRNAs and plasmids were transiently transfected into EC cells using HiPerFect reagent.

HEC-1A/PA cells under treating with 10 mg/L paclitaxel. \*\**P* < 0.01 vs. ss-NC group.

cell viability of Ishikawa/PA and HEC-1A/PA cells under treating with 10 mg/L paclitaxel. \*\**P* < 0.01 vs. ss-NC group. (H) The cell apoptosis of Ishikawa/PA and

#### Cell Proliferation Assay

Enhanced Cell Counting Kit-8 (Beyotime Institute of Biotechnology, Beijing, China) was applied to examine cell proliferation. The cells in logarithmic growth phase were digested with trypsin, washed by phosphate-buffered saline (PBS), and suspended in the culture medium. Then, 2,000 cells in 100 µl medium were added into one pore of 96-well plates, 10 µl enhanced CCK-8 solution was added, and incubated for 1 h. The value of optical density was detected with the help of an MK3 microplate reader (Thermo Fisher Scientific, Waltham, MA, USA) at the wavelength of 450 nm.

#### Cell Apoptosis Detection

Annexin V-FITC/PI Apoptosis Detection Kit (Jiancheng, Nanjing, Jiangsu, China) was used to detect cell apoptosis rate according to the manufacturer's instructions. In addition, 2 × 10<sup>5</sup> cells were re-suspended in 500 µl binding buffer, 5 µl Annexin V-FITC and 5 µl Propidium iodide (PI) were added, and incubated at 25◦C for 10 min. The apoptosis rate was detected and analyzed by FACScan flow cytometry with Diva 8.0 software (Becton Dickinson, Franklin Lakes, NJ, USA). The apoptosis rate was presented as the percentage of cells with FITC-Annexin V positive/PI negative in the right lower quadrant.

#### Drug Sensitivity Assay

The Ishikawa/PA and HEC1A/PA cells were treated with paclitaxel (10, 20, 50, 100, and 150 mg/L) (12). The cell viability was examined after 24 h. Then, the half maximal inhibitory concentration (IC50) of paclitaxel was calculated according to their dose-response curve.

#### Western Blotting

Protein of cells was extracted using a Protein Extraction Kit (Beyotime Institute of Biotechnology, Beijing, China), and quantified by using a Bradford Protein Assay Kit (Beyotime Institute of Biotechnology, Beijing, China). Protein (30 µg) was separated by polyacrylamide gel electrophoresis and transferred to a polyvinylidene fluoride (PVDF) membrane. PVDF membrane was blocked with Tween-Tris-buffered saline (TTBS), containing 5% non-fat milk at 25◦C for 2 h, and hybridized with Bcl2 and MRP4 antibodies (#2872 and #12705; Cell Signaling Technology, Danvers, MA, USA) at 4◦C overnight. After that, PVDF membrane was incubated with the second antibody at 25◦C for 2 h. The PVDF membrane was treated with Chemiluminescent reagent (TIANGEN, Beijing, China) to visualize the bands. Then, the bands were analyzed by ImageJ software (NIH, Bethesda, MD, USA).

#### RNA Pull-Down Assay

The biotinylated probes for CDKN2B-AS and miR-125a-5p (bio-CDKN2B-AS-W and bio-miR-125a-5p-W, containing wild-type binding site), as well as their negative controls (bio-CDKN2B-AS-M and bio-miR-125a-5p-M, containing mutant binding site) were synthetized by GenePharma Co. Ltd. (Shanghai, China). Probes were dissolved in the buffer and incubated with Dynabeads M-280 Streptavidin (Thermo Fisher Scientific, Waltham, MA, USA) for 10 min at 25◦C to form probe-coated beads. Those probe-coated beads were incubated with the lysates from Ishikawa/PA and HEC1A/PA cells, and eluted with the washing buffer. The pulled down RNAs were detected by RT-qPCR.

#### RNA Immunoprecipitation (RIP) Assay

RIP was assayed using a Magna RIPTM RNA-Binding Protein Immunoprecipitation Kit (Millipore Sigma, Burlington, MA, USA) according to the manufacturer's instructions and a previous study (27). Whole-cell lysate from Ishikawa/PA and HEC1A/PA cells was incubated with RIP buffer, containing magnetic beads conjugated with human anti-Ago2 antibody or negative control normal mouse IgG. Samples were incubated with Proteinase K, and immunoprecipitated RNA was isolated. The RNA concentration was measured by a spectrophotometer (NanoDrop, Thermo Fisher Scientific, Waltham, MA, USA) and the quality of RNA was assessed by using a bio-analyzer (Agilent, Santa Clara, CA, USA). Furthermore, purified RNAs were extracted and analyzed by RT-qPCR to demonstrate the presence of the binding targets.

#### Luciferase Reporter Assay

The luciferase reporter plasmids (Bcl2-W and MRP4-W, containing wild-type binding site; Bcl2-M and MRP4-M, containing mutant binding site) were synthesized by the GenScript Co. (Piscataway, NJ, USA). HEK293T cells were cotransfected with the luciferase reporter plasmids and microRNAs, respectively. The Luciferase Reporter Kit (Beyotime Institute of Biotechnology, Beijing, China) was applied to detect the luciferase activity after 48 h in accordance with manufacturer's instructions.

#### Statistical Analysis

Each experiment was repeated five times, the data were presented as mean ± standard deviation (SD), and analyzed with Student's t-test and one-way analysis of variance (ANOVA) using SPSS 22.0

TABLE 1 | The correlation analysis between the expression of ANRIL gene and the clinicopathological characteristics of 87 EC tissues.


\**P* < *0.05,* \*\**P* < *0.01.*

software (IBM, Armonk, NY, USA). If the P < 0.05, the difference was statistically significant.

#### RESULTS

#### The Expression Level of CDKN2B-AS Was Up-Regulated in EC

CDKN2B-AS gene in EC specimens was up-regulated in comparison with matched PNET specimens (**Figure 1A**). In addition, the expression level of CDKN2B-AS gene in Ishikawa and HEC-1A cells was significantly higher than that in HEC-251 cells (**Figure 1B**).

As shown in **Table 1**, the high-expression of CDKN2B-AS showed a positive correlation with high pathological grade of EC, however, that was not correlated with other clinical characteristics, including patient's age, estrogen receptor, and progesterone receptor. These findings suggested that CDKN2B-AS gene was involved in the progress of EC.

### High-Expression of CDKN2B-AS Was Correlated With Poor Response to Paclitaxel in EC

Furthermore, the over-expression of CDKN2B-AS was associated with the low paclitaxel sensitivity of EC patients (**Table 1** and **Figure 1C**), and the expression level of CDKN2B-AS in Ishikawa/PA and HEC1A/PA cells was significantly higher than that in Ishikawa and HEC1A cells (**Figure 1D**), which preliminarily confirmed that CDKN2B-AS participated in the genesis of chemotherapy resistance in EC cells.

To verify the roles of CDKN2B-AS on chemotherapy resistance, the expression level of CDKN2B-AS was knockdown in Ishikawa/PA and HEC1A/PA cells to carry out lossof-function assays (**Figure 1E**). Knockdown of CDKN2B-AS decreased the IC50 of paclitaxel from 81.29 and 62.37 mg/L to 38.41 and 29.74 mg/L in Ishikawa/PA and HEC-1A/PA cells, respectively, (**Figure 1F**), which certified that knockdown of CDKN2B-AS sensitized Ishikawa/PA and HEC-1A/PA cells to paclitaxel. Moreover, under treating with paclitaxel (10 mg/L), knockdown of CDKN2B-AS depressed cell viability (**Figure 1G**) and promoted apoptosis in Ishikawa/PA and HEC-1A/PA cells (**Figure 1H**). Besides, knockdown of CDKN2B-AS gene significantly increased paclitaxel-induced cytotoxicity.

#### CDKN2B-AS Silenced the Expression Level of miR-125a-5p in EC Cells

The online databases (TargetScan 7.1 and Starbase 2.0) predicted a specific combination between CDKN2B-AS and miR-125a-5p (**Figure 2A**). Then, the expression level of miR-125a-5p in EC specimens was down-regulated in comparison with matched PNET specimens (**Figure 2B**), and analysis of the co-expression patterns showed a negative correlation between CDKN2B-AS and miR-125a-5p in EC cells (**Figure 2C**, r = −0.5609, P < 0.001). Knockdown of CDKN2B-AS significantly up-regulated the expression level of miR-125a-5p in Ishikawa/PA and HEC-1A/PA cells (**Figure 2D**). In addition, RNA pull-down assay identified CDKN2B-AS could be combined with bio-miR-125a-5p-W probe, except for bio-miR-125a-5p-M (**Figure 2E**). Similarly, miR-125a-5p could be combined with bio-CDKN2B-AS-W probe, while that was not observed for bio-CDKN2B-AS-M (**Figure 2F**). Furthermore, RIP experiments confirmed that CDKN2B-AS and miR-125a-5p were both enriched in anti-Ago2 group (**Figures 2G,H**), and they were in a RNA-induced silencing complex (RISC). These results indicated that CDKN2B-AS and miR-125a-5p were associated with Ago-2 protein in a RISC complex, and CDKN2B-AS decreased the expression level of miR-125a-5p in a RISC-dependent manner, which was a classical regulatory manner of lncRNAs, regulating microRNAs.

#### Up-Regulation of the Expression Level of miR-125a-5p Inhibited Paclitaxel Resistance in EC

The expression level of miR-125a-5p in resistant EC patients was down-regulated in comparison with sensitive patients (**Figure 3A**). Similarly, compared with Ishikawa and HEC1A cells, the expression level of miR-125a-5p was also downregulated in Ishikawa/PA and HEC1A/PA cells (**Figure 3B**).

To identify the role of miR-125a-5p on paclitaxel resistance, the expression level of miR-125a-5p was up-regulated in Ishikawa/PA and HEC1A/PA cells to carry out gain-of-function assays (**Figure 3C**). Additionally, up-regulation of miR-125a-5p decreased the IC50 of paclitaxel from 78.68 to 63.24 mg/L to 36.17 and 27.52 mg/L in Ishikawa/PA and HEC-1A/PA cells, respectively, (**Figure 3D**), which certified that up-regulation of miR-125a-5p sensitized Ishikawa/PA and HEC-1A/PA cells to paclitaxel. Moreover, under treatment with paclitaxel (10 mg/L), up-regulation of miR-125a-5p depressed cell viability (**Figure 3E**), and promoted apoptosis in Ishikawa/PA and HEC-1A/PA cells (**Figure 3F**). Up-regulation of miR-125a-5p also significantly increased paclitaxel-induced cytotoxicity.

#### MiR-125a-5p Mediated the Chemoresistance-Suppressive Effects of Knockdown of CDKN2B-AS in EC Cells

To determine whether the chemoresistance-suppressive effects of CDKN2B-AS knockdown were mediated by miR-125a-5p, antagomiR-125a-5p was transfected into stable ss-CDKN2B-AS cells (**Figure 4A**). Co-transfection of ss-CDKN2B-AS with antagomiR-125a-5p showed the lower paclitaxel sensitivity compared with co-transfection of ss-CDKN2B-AS with antagomiR-NC, and transfection with antagomiR-125a-5p rescued the inhibitory effects of ss-CDKN2B-AS on paclitaxel resistance (**Figures 4B–D**). Based on the above-mentioned results, we confirmed that miR-125a-5p mediates suppressive effects of CDKN2B-AS knockdown on paclitaxel resistance in EC cells.

#### Bcl2 and MRP4 Are Target Genes of miR-125a-5p in EC Cells

Based on information obtained from the online database TargetScan 7.1, Bcl2 and MRP4 might be targets of miR-125a-5 (**Figure 5A**). Firstly, up-regulation of miR-125a-5p remarkably

depressed the expression level of Bcl2 and MRP4 in Ishikawa/PA and HEC-1A/PA cells (**Figure 5B**). Then, the specific binding sites of miR-125a-5p in the 3′UTR of Bcl2 and MRP4 were confirmed by luciferase reporter assay. In the Bcl2-W group, the luciferase activity of co-transfection with agomiR-125a-5p was inhibited, while no significant change was observed in their NC group; in the Bcl2-M group, the luciferase activity remained unchanged (**Figure 5C**). In addition, similar results were observed in MRP4 (**Figure 5D**). The results confirmed our prediction that Bcl2 and MRP4 are targets of miR-125a-5p in EC cells.

# Enhancement of Bcl2 and MRP4 Partially Reversed the Suppressive Effects of Up-Regulation of miR-125a-5p on Paclitaxel Resistance in EC Cells

To uncover whether Bcl2 and MRP4 could reverse the chemoresistance-suppressive effects of up-regulation of miR-125a-5p in EC cells, the cells were co-transfected with agomiR-125a-5p, as well as pUC-Bcl2 or pUC-MRP4. Western blot analysis demonstrated that pUC-Bcl2 could up-regulate the expression level of Bcl2 in agomiR-125a-5p + pUC-Bcl2 group compared with agomiR-125a-5p + pUC-NC group (**Figure 6A**). In addition, similar results were achieved with pUC-MRP4 transfection (**Figure 6B**).

Drug sensitivity assay showed that up-regulation of miR-125a-5p sensitized Ishikawa/PA and HEC-1A/PA cells to paclitaxel, and over-expression of Bcl2 promoted the paclitaxel resistance and notably decreased miR-125a-5p-triggered increasing in paclitaxel sensitivity (**Figure 6C**). Similarly, over-expression of MRP4 also partially rescued the promotive effects of miR-125a-5p's on paclitaxel sensitivity (**Figure 6D**). Enhancement of Bcl2 and MRP4 could partially reverse the suppressive effects of upregulation of miR-125a-5p on paclitaxel resistance in EC cells. These results illustrated that miR-125a-5p suppressed paclitaxel resistance by targeted silencing Bcl2 and MRP4 in EC cells.

#### DISCUSSION

Recent studies reported that lncRNAs were involved in the formation and maintenance of chemotherapy being resistance in

almost all malignant tumors, including EC. For example, KCNQ1 Opposite Strand/Antisense Transcript 1 (KCNQ1OT1) was closely associated with insensitivity of lung adenocarcinoma, and also knockdown of KCNQ1OT1 depressed the expression level of Multidrug Resistance Protein 1 (MDR1) and the paclitaxel resistance (28). Urothelial carcinoma associated 1 (UCA1) is an independent prognostic biomarker, which contributed to adriamycin resistance in gastric cancer (29).

Paclitaxel is a broad-spectrum antitumor drug that is used in chemotherapy for a variety of tumors, including EC. Paclitaxel is isolated from the bark of the Pacific yew tree, and it can bind to tubulin and inhibit the disassembly of microtubules, thereby causing the obstacle of cell division and the death of tumor cells. However, paclitaxel resistance is a major factor associated with treatment failure.

In this study, CDKN2B-AS gene was highly expressed in EC tissues and cell lines, and its over-expression was positively associated with high pathological grade of EC, which suggested that CDKN2B-AS gene was involved in the progress of EC. Furthermore, the over-expression of CDKN2B-AS was associated with the low sensitivity to paclitaxel of EC patients, which preliminarily confirmed that CDKN2B-AS participated in the genesis of chemotherapy resistance in EC. Lan et al. reported that knockdown of CDKN2B-AS inhibited the development of multidrug resistance in gastric cancer cells to paclitaxel or 5 fluorouracil (30). CDKN2B-AS also contributed to paclitaxel resistance of lung adenocarcinoma A549 cells (31).

To verify the roles of CDKN2B-AS on paclitaxel resistance, the impacts of CDKN2B-AS on paclitaxel resistance in EC were examined by a series of loss-of-function assays. Knockdown of CDKN2B-AS decreased IC50 of paclitaxel in Ishikawa/PA and HEC-1A/PA cells, and also promoted the cytotoxicity induced by paclitaxel, which showed that knockdown of CDKN2B-AS increased paclitaxel sensitivity in EC cells.

Recently, lncRNAs were proposed to act as miRNA sponges or competitive endogenous RNAs (ceRNAs), forming extensive regulatory networks, thereby negatively regulating miRNA expression (32). For instance, LINC00161 sensitized the osteosarcoma cells to cisplatin through sponging miR-645 (33); lncRNA AC023115.3 acted as a ceRNA for miR-26a to inhibit cisplatin resistance of glioma cells (34). Using online bioinformatics databases, we predicted a negative regulatory relationship between CDKN2B-AS and miR-125a-5p. Furthermore, non-coding RNAs can ordinarily form ribonucleoprotein (RNP) complexes with their partner proteins to exert their functions and miRNAs assembling with Argonaute (Ago) family proteins into an effective complex called RISC, mediating silencing the target gene (35, 36).

A limited number of researches on miR-125a-5p and chemotherapy resistance were conducted. To date, only one literature reported that up-regulation of miR-125a-3p could advance docetaxel sensitivity of breast cancer cells through modulation of BRCA1 signaling (37). Little attention has been paid to the miR-125a-5p-associated chemotherapy effect in EC, especially to paclitaxel. Therefore, the correlation between the expression of miR-125a-5p and paclitaxel resistance in EC was here examined, and the results revealed that its lowexpression was related to poor paclitaxel response of EC patients. Furthermore, enhanced miR-125a-5p decreased IC50 of paclitaxel in Ishikawa/PA and HEC-1A/PA cells, and promoted the paclitaxel-induced cytotoxicity, which showed that overexpression of miR-125a-5p could increase paclitaxel sensitivity in EC cells.

We attempted to explore whether the influence of CDKN2B-AS on paclitaxel resistance is mediated by miR-125a-5p in EC. In detail, the regulatory relationship between CDKN2B-AS and miR-125a-5p was confirmed based on the following results: (1) Knockdown of CDKN2B-AS significantly increased the expression level of miR-125a-5p in Ishikawa/PA and HEC-1A/PA cells; (2) RNA pull-down assay revealed CDKN2B-AS function via interaction with miR-125a-5p; (3) RIP experiment confirmed that CDKN2B-AS and miR-125a-5p were associated with Ago-2 protein in a RISC complex, and suggested that CDKN2B-AS decreased the expression level of miR-125a-5p in a RISC-dependent manner; (4) miR-125a-5p deficiency rescued the inhibitory effect of knockdown of CDKN2B-AS on paclitaxel resistance. Based on the above-mentioned achievements, we confirmed that miR-125a-5p mediates the suppressive effects of knockdown of CDKN2B-AS on paclitaxel resistance in EC cells.

As microRNAs play their roles through regulating their target genes, such as miR-16-1 and FUBP1 (38), we predicted that Bcl2 and MRP4 genes might be miR-125a-5p's target genes based on the online bioinformatics database. In addition, Tong et al. reported that miR-125a-5p could regulate cytobiological phenotypes of colon cancer via targeting Bcl2 (20). A series of following gain-of-function experiments, such as luciferase reporter assay and Western blotting, demonstrated that Bcl2 and MRP4 genes were the target genes of miR-125a-5p. To sum up, we speculate that miR-125a-5p might modulate the paclitaxel resistance of EC cells through targeted silencing Bcl2 and MRP4.

To verify this hypothesis, knockdown of Bcl2 and MRP4 by miR-125a-5p enhancement was rescued by transfection with pUC-Bcl2 and pUC-MRP4. The following experiments found that up-regulation of Bcl2 and MRP4 separately reversed the regulatory roles of miR-125a-5p on paclitaxel resistance in Ishikawa/PA and HEC-1A/PA cells. Besides, miR-125a-5p might promote cell apoptosis and reduce paclitaxel discharge

FIGURE 5 | Bcl2 and MRP4 are target genes of miR-125a-5p in EC cells. (A) The predicted miR-125a-5p binding site in the 3′UTR of Bcl2 and MRP4 mRNA. Short vertical lines indicated complementary nucleotides. (B) The expression of Bcl2 and MRP4 protein in Ishikawa/PA and HEC-1A/PA cells. \**P* < 0.05 vs. agomiR-NC group. \*\**P* < 0.01 vs. agomiR-NC group. (C) The relative luciferase reporter assay of HEK 293T cells co-transfected with Bcl2-W or Bcl2-M and agomiR-125a-5p or agomiR-NC. \*\**P* < 0.01 vs. Bcl2-W + agomiR-NC group. (D) The relative luciferase reporter assay of HEK 293T cells co-transfected with MRP4-W or MRP4-M and agomiR-125a-5p or agomiR-NC. \*\**P* < 0.01 vs. MRP4-W + agomiR-NC group.

through silencing targeted Bcl2 and MRP4, and then re-sensitize Ishikawa/PA and HEC-1A/PA cells to paclitaxel.

To sum up, CDKN2B-AS is specifically combined with miR-125a-5p and also down-regulates its expression level in EC, weakening the ability of miR-125a-5p to silence its target genes, and then up-regulates the expression of Bcl2 and MRP. Additionally, Bcl2 can inhibit apoptosis and promote cell survival from paclitaxel-induced cytotoxicity. Moreover, an overexpressed MRP4 pumps paclitaxel out of EC cells, and reduced intracellular paclitaxel concentration and paclitaxelinduced chemotherapy damage. Accordingly, CDKN2B-AS can desensitize EC cells into paclitaxel, and contribute to the formation of chemotherapy resistance through miR-125a-5p/Bcl2 and miR-125a-5p/MRP4 pathways.

In conclusion, high-expression of CDKN2B-AS is associated with a poor response to paclitaxel of EC patients, and knockdown

#### REFERENCES


of CDKN2B-AS inhibits paclitaxel resistance through miR-125a-5p-Bcl2/MRP4 pathway in EC. Our findings helped elucidate the underlying mechanism of chemotherapeutic resistance in EC patients.

#### AUTHOR CONTRIBUTIONS

LM and CS conceived and designed the project. CS and CA completed the experiments and acquired the data. CA and CC analyzed the data. All authors read and approved the final version of this manuscript.

#### FUNDING

This study was funded by the Foundation of Macau Polytechnic Institute (RP/ESS-03/2016).

an antisense non-coding RNA whose expression coclusters with ARF. Cancer Res. (2007) 67:3963–9. doi: 10.1158/0008-5472.CAN-06-2004


resistance in castration-resistant prostate cancer. Oncol Lett. (2017) 14:1748– 56. doi: 10.3892/ol.2017.6357


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Shang, Ao, Cheong and Meng. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Telomerase and Telomeres in Endometrial Cancer

Rafah A. A. Alnafakh1,2, Meera Adishesh1,2, Lucy Button1,2, Gabriele Saretzki <sup>3</sup> and Dharani K. Hapangama1,2 \*

*<sup>1</sup> Liverpool Women's Hospital NHS Foundation Trust, Liverpool, United Kingdom, <sup>2</sup> Department of Women's and Children's Health, Institute of Translational Medicine, University of Liverpool, Liverpool, United Kingdom, <sup>3</sup> The Ageing Biology Centre and Institute for Cell and Molecular Biosciences, Newcastle University, Newcastle upon Tyne, United Kingdom*

Telomeres at the termini of human chromosomes are shortened with each round of cell division due to the "end replication problem" as well as oxidative stress. During carcinogenesis, cells acquire or retain mechanisms to maintain telomeres to avoid initiation of cellular senescence or apoptosis and halting cell division by critically short telomeres. The unique reverse transcriptase enzyme complex, telomerase, catalyzes the maintenance of telomeres but most human somatic cells do not have sufficient telomerase activity to prevent telomere shortening. Tissues with high and prolonged replicative potential demonstrate adequate cellular telomerase activity to prevent telomere erosion, and high telomerase activity appears to be a critical feature of most (80–90%) epithelial cancers, including endometrial cancer. Endometrial cancers regress in response to progesterone which is frequently used to treat advanced endometrial cancer. Endometrial telomerase is inhibited by progestogens and deciphering telomere and telomerase biology in endometrial cancer is therefore important, as targeting telomerase (a downstream target of progestogens) in endometrial cancer may provide novel and more effective therapeutic avenues. This review aims to examine the available evidence for the role and importance of telomere and telomerase biology in endometrial cancer.

#### Edited by:

*Paola Gehrig, University of North Carolina at Chapel Hill, United States*

#### Reviewed by:

*Marilyn Huang, Sylvester Comprehensive Cancer Center, United States Saori Furuta, University of Toledo, United States*

\*Correspondence:

*Dharani K. Hapangama dharani@liv.ac.uk*

#### Specialty section:

*This article was submitted to Women's Cancer, a section of the journal Frontiers in Oncology*

Received: *08 January 2019* Accepted: *15 April 2019* Published: *17 May 2019*

#### Citation:

*Alnafakh RAA, Adishesh M, Button L, Saretzki G and Hapangama DK (2019) Telomerase and Telomeres in Endometrial Cancer. Front. Oncol. 9:344. doi: 10.3389/fonc.2019.00344* Keywords: endometrial cancer, telomere, telomerase, endometrium, TERRA, TRAP, hTERT, hTERC

# INTRODUCTION

Telomeres are specialized structures that are found at the ends of linear chromosomes, containing a tandemly repeated specific DNA sequence and associated protective proteins. The protective function of telomeres in preventing the loss of genomic DNA in proliferating cells is well-established (1–3). As telomeres shorten with each cell division, critically short telomeres initiate cellular senescence or an apoptotic pathway, leading to cessation of cell division, therefore telomere shortening is a major tumor suppressor mechanism (4, 5). In addition, oxidative stress is an important additional cause for telomere shortening (6, 7). Telomerase is a unique reverse transcriptase enzyme (8) that is able to add repetitive telomeric sequences de novo onto telomeric ends (9) that are continually lost during DNA replication due to oxidative stress and the "end replication problem" in mitotic cells. Thus, telomerase prevents shortening and maintains telomeres. However, most human somatic cells do not have significant levels of telomerase activity whereas cells, such as embryonic stem cells and most cancer cells exhibit high telomerase activity while adult tissue stem cells are potentially able to up-regulate telomerase upon activation (10–12).

Human endometrium is a unique somatic organ that contains a relatively high yet dynamic pattern of telomerase activity that changes according to the menstrual cycle, correlating with endometrial cellular proliferation (13, 14). Further evidence from benign endometrium also suggests that telomerase activity is a fundamental requirement for endometrial cell proliferation and survival (15). The involvement of telomerase in most cancerrelated cellular abnormalities in cell fate regulatory pathways prompted many studies into telomerase and telomeres in a variety of cancers including endometrial cancer (16–18).

Endometrial cancer is the fourth common cancer in women in the UK and is the commonest gynecological cancer (CRUK). Increasing obesity and longevity have both caused the incidence of EC to increase at an alarming rate. For example, in the United Kingdom, the incidence of EC increased by more than 40% since 1993. European estimates predict a 100% increase in the incidence by 2025 not only in older post-menopausal women but also in younger women (19). Figures from the UK report that mortality associated with EC has risen by 21% over the last decade in an era of improving survival rates for most other cancers, highlighting the inequality and lack of translation of advances in cancer research to EC (CRUK) (20). The survival rates for highgrade EC are exceptionally poor, similar to ovarian cancer; and the traditional surgical treatment is associated with significant morbidity and mortality for many women even when presented with early disease due to frequently occurring co-morbidities and obesity (21). Urgent novel therapeutic options are therefore needed to prevent, treat as well as to avoid progression of EC.

Although EC is an important disease with a significant clinical and economic consequence, the molecular biology of endometrial carcinogenesis is not well-described or understood when compared with other female-specific malignancies, such as breast or ovarian cancer. Human endometrium is a unique organ with a massive regenerative potential (22) and is the main target organ for ovarian steroid hormone action (23). While being a hormonally responsive tissue, endometrium responds rather differently to the same steroid hormones than other hormone responsive organs, such as breast tissue (23, 24). This has made it difficult to translate the pioneering discoveries made in other cancers to EC management and therapy. Unlike most other somatic tissue, benign endometrial tissue demonstrate high telomerase activity, and telomerase has a pivotal functional role in healthy endometrial cell proliferation (14, 15). High telomerase activity is observed in most epithelial cancers, and the carcinogenesis process in those tissues involved ectopic expression of telomerase components and genetic alterations, such as activation mutations in promotors of the vital genes. In the endometrium however, the high telomerase activity is a feature even without being associated with driver mutations. It is therefore intriguing to explore the distinctive endometrial telomerase biology relevant to EC and we hypothesize EC to have a unique telomerase biology that is different to the other cancers. Furthermore, EC is a hormone driven disease and advanced and recurrent ECs are treated with progesterone which regress these tumors albeit without extending survival (24). It is therefore of particular interest to examine telomerase as a downstream progesterone target in the endometrium (15) which can be manipulated for therapeutic utility in progesterone resistant ECs. This review therefore focuses on the significance and role of telomerase and telomere biology in EC, highlighting recent findings proposing some aspects of telomerase biology as potential therapeutic targets for EC (25).

# METHOD

We performed systematic PubMed (Medline) and Ovid searches using a combination of relevant controlled vocabulary terms and free-text terms related to telomeres and telomerase. The key words used included: telomerase, telomeres, telomere length, telomerase reverse transcriptase (TERT), telomeric RNA component (TERC), shelterin proteins, telomerase associated proteins, with endometrium, endometriosis, endometrial hyperplasia, endometrial cancer (EC), endometrial carcinomas, uterine cancer, cancers. All studies investigating telomerase or telomere biology in endometrium in women or animals or respective cell lines, either primary cells or tissue explants in culture, and published from database inception until December 2018, were included in this review.

# TELOMERES

#### Structure

Human telomeres consist of a repetitive TTAGGG hexanucleotide sequence bound by six-proteins forming the shelterin complex [(26) **Figure 1**]. In normal somatic cells the average length of telomeres is around 5–15 kilobases and they shorten in vitro by 30–200 base pairs (bp) during every cell division depending on the cell type and environmental conditions (34). Under increased oxidative stress telomere shortening rate per cell division can increase substantially, up-to 500 bp (6).

Most of the non-coding telomeric DNA is double-stranded whilst the terminal nucleotides (nt) form the single stranded 3′ Grich overhang, which serves as the primer for telomerase action (35) and also protect telomeres from being recognized as DNA damage. This forms a D-loop (Displacement loop) facilitating repetitive DNA sequences to be added by telomerase (36).

Another mechanism to protect telomeres from being recognized as DNA damage is the formation a t-loop, which is a specific higher order conformation. This large duplex loopback structure is formed via invasion of the single-stranded telomeric 3′ overhang into the double stranded telomeric repeat array (37). The authors suggested that the t-loops are the basic mechanism by which the telomeric nucleoprotein complex sequesters chromosome ends from the DNA damage pathway, preventing inappropriate DNA repair and telomerase action (37).

The shelterin complex (**Figure 1**) includes telomeric repeat binding factor 1 and 2 (TRF1 and TRF2), which are homodimeric proteins that bind specifically to double-strand telomeric DNA (27, 37). In contrast, Protection of telomeres 1(POT1) binds to the single-stranded region of the telomere (28) and forms a heterodimer with TPP1 (38). The Repressor/activator protein 1 (RAP1) is recruited through its relation with TRF2 (31) and TRF1-interacting protein 2 (TIN2) is the central part of

the shelterin complex (29) and it interacts with TRF1, TRF2 (30), and POT1/TPP1 (32) to assure structural integrity of the complex. Removal of individual shelterin proteins has been shown to stimulate a DNA damage response (DDR) pathway: TRF1 prevents the stimulation of both ataxia-telangiectasia mutated (ATM) and ataxia telangiectasia and Rad3 related (ATR) pathways (39); TRF2 and RAP1 inhibit the activation of the ATM pathway (40, 41) and homology-directed recombination (HDR) (42) while TPP1 bound POT1 (POT1a/b in mouse) inhibit the ATR pathway (43). TRF2 plays a vital role in facilitating this t-loop formation (44). Super-resolution fluorescence light microscopy visualization of the t-loop has shown that the strand invasion point can be located at almost any point along the duplex DNA, resulting in highly variable t-loops sizes (45).

#### Functions of Telomeres

The main function of telomeres is to protect chromosomal ends from degradation and end-to end-fusion (1) as well as to prevent the ends of chromosomes being recognized as DNA damage by the DNA damage response machinery of the cell (37). However, when telomeres are critically short, they activate the apoptosis/senescence pathways, thereby preventing genetic material being lost by inhibiting inappropriate continuous DNA replication in the context of short telomeres. The telomere structure described above, prevents inappropriate DNA repair at these sites, for example the loop conformation (D-loop) masks the single stranded terminal DNA and enables its protection from the DNA damage response pathway (37).

The shelterin complex supports the chromosome protective function of telomeres and stabilization of telomere lengths, and the complex interaction of shelterin proteins at the chromosomal ends have a key role in telomere maintenance via a negative feedback loop which also has an inhibitory effect on the telomerase enzyme (46).

In cells which have replicative capability, telomere shortening can lead to chromosomal instability by promoting end-to-end fusions leading to multiple chromosomal aberrations, such as breakages, fusions, and translocations rendering the genome aneuploid and therefore promoting carcinogenesis. To maintain telomere length, the homeostasis mechanism that involves telomerase, uses both TRF1 and TRF2 as negative regulators that stabilize and limit telomere length elongation (47, 48). Overexpression of both TRF1 and TRF2 was reported to cause telomere shortening (47) and this could be due to the binding of TRF1 and TRF2 along the length of the double stranded telomeric repeat array which measures telomere length as demonstrated in yeast (47, 48). POT1 can either facilitate or inhibit telomerase accessing telomeres depending on its position relative to the DNA 3 ′ -end (49). Examining the high-resolution crystal structure of the human POT1-TTAGGGTTAG complex suggested that it would not be elongated by telomerase. When POT1 is bound at one telomeric repeat before the 3′ -end, leaving an 8-nucleotide 3 ′ -tail, the resulting complex is elongated with increased activity and processivity (50). Replication protein A (RPA) is another ssDNA binding protein which has an important role in telomere replication by facilitating telomerase enzyme at the telomeres (51, 52). It also recruits the ATR-ATRIP protein kinase complex to DNA damage sites and initiates the checkpoint signaling (53, 54). Collectively, the available evidence demonstrates that shelterin and other telomere-binding proteins are involved in the regulation of telomere length.

**Gene regulation** is another reported function of telomeres but with limited evidence available for it. Telomeric attrition extensively alters expression of some genes, and the difference in expression of genes proximal to telomeres may result from chromatin modifications, a conserved phenomenon termed as telomere position effect (TPE). TPE is a silencing mechanism spreading from the telomeres toward subtelomeric regions (55). In humans, only a limited number of endogenous genes (e.g., ISG15) has been mentioned to be affected by TPE (56, 57), however, microarray data suggests that the expression of many other genes close to telomeres to be also altered with the aid of a telomere length-dependent and DNA damage-independent mechanism, and this is known as telomere position effect–over long distance (TPE-OLD) (58). For example, the looping of chromosomes brought long telomeres closer to some genes which are over 10 Mb away from the telomere, but these same loci were completely separated from the telomeres when the telomeres were short (58). Further microarray data supports the notion that telomere length-dependent chromosome conformation can affect the transcription of non-subtelomeric genes (58). At the genomewide level, the effect of this mechanism on gene expression has been proposed to occur earlier than replicative senescence and that could potentially explain the increased incidence of age-related pathologies that are associated with old age without necessarily imposing a DNA damage signal from a critically-short telomere (59, 60).

Telomere length is the main determinant of a cell's replicative life span. Dysfunctional telomeres which result from either progressive telomere shortening, internal DNA damage (61) or shelterin complex loss, provoke a strong DNA damage response and genomic instability (62). A plethora of experimental data has shown that tumorigenesis can be caused by genome instability resulting from telomere shortening (4, 63). Nevertheless, in late generation telomerase knock-out mouse models, telomere attrition was also a tumor suppressor mechanism through the induction of replicative senescence or apoptosis that repress tumorigenesis. Telomere shortening and telomere uncapping in metazoans stimulate ATM/ATR kinases to phosphorylate downstream kinases CHK1 and CHK2, which initiate p53 dependent replicative senescence and apoptosis pathways which inhibit tumor formation (4).

#### TERRAs (Telomeric Repeat Containing RNAs)

Telomeres were initially thought to be transcriptionally silent, but recently they have been found to be transcribed into telomeric repeat containing, long non-coding RNAs, termed TERRAs (64). TERRAs have a role on telomere regulation and also regulate telomeric access of telomerase as described below in more detail.

# REGULATION OF TELOMERE LENGTH AND TELOMERE MAINTENANCE MECHANISMS (FIGURE 2)

The most widely known classical telomere maintenance mechanism is dependent on telomerase reverse transcriptase activity. However, another telomerase-independent telomere maintaining pathway has been described in cells that do not have measurable telomerase activity, termed alternative lengthening of telomeres (ALT) pathway (69). TERRAs also have a role in telomere length regulation by mainly managing telomeric access of telomerase.

#### Telomerase

#### Structure of Telomerase (Figure 1)

Telomerase, the only RNA dependent DNA polymerase in mammals, was first discovered in protozoans in 1985 (70), and subsequent studies demonstrated mammalian/human species in 1989 (71). The telomerase holoenzyme contains three core components: the RNA component harboring the template region for telomere synthesis (hTR or hTERC), a catalytic protein with reverse transcriptase activity, hTERT (72) as well as dyskerin (**Figure 1**). However, only the RNA component (TERC) and the catalytic subunit (TERT) are necessary and sufficient for in vitro telomerase activity (73). **Table 1** lists some of the well-known telomerase associated proteins.

#### **Telomerase RNA component (hTERC or hTR)**

The human telomerase RNA (TERC or hTR) consists of 451 nt and is an essential constituent of the telomerase catalytic core complex. Although the length is variable among eukaryotes, the structure of TERC remains conserved. For example, the length ranges from ∼150 nt in ciliates, 400–600 nt in vertebrates to ∼1,300 nt in yeast (114). Additionally, in ciliates, polymerase III transcribes the telomerase RNA (115), whereas it is RNA polymerase II in yeast and vertebrates (116).

Vertebrate TERC's secondary structure has four conserved elements: a pseudoknot domain (CR2/CR3), a CR4/CR5 (conserved region 4 and conserved region 5) domain, box H/ACA (CR6/CR8) domain and a CR7 domain (114, 117). The proximal template/pseudoknot domain and the distal CR4/5 domain represent the essential regions of TERC for telomerase activity (118).

As mentioned before, an active telomerase enzyme can be generated by combining the two RNA domains from the TERC subunit with the TERT protein on oligodeoxynucleotide substrates in vitro (73, 119–121). The human/vertebrate TERC has a third, conserved component, the H/ACA domain located at the 3′ end that has homologies to small nucleolar (sno) and small Cajal body-specific (sca) RNAs. The TERC H/ACA region binds to telomerase associated proteins, such as dyskerin, NOP10, NHP2, and GAR1 (14), and this region is essential

for telomerase biogenesis, and are important for RNA stability. Additionally, in the 3′ stem-loop of the H/ACA, there is another domain, the Cajal body localization box (CAB), for binding the telomerase Cajal body protein 1 (TCAB1) (33). Mutations in the H/ACA region decrease TERC accumulation, whereas mutations in the CAB cause TERC to accumulate in nucleoli instead of Cajal bodies (122, 123). Although this mutant TERC has the capacity of forming catalytically active telomerase in vivo, it is highly impaired in telomere elongation because of the decreased association of telomerase with telomeres (124). This result emphasizes that sub-nuclear localization of telomerase as an important regulatory mechanism for the homeostasis of telomere length in human cells (124). TERC therefore, not only provides the template, which identifies the telomere repeat sequence, but it also comprises motifs, which are crucial to reconstitute telomerase activity (125). Furthermore, it plays a role in stability, maturation, accumulation, and functional assembly of the telomerase holo-enzyme.

#### **hTERT**

TERT is the catalytic component of the telomerase enzyme and as described above, together with TERC, it is essential for telomerase activity and thus for the maintenance of telomere length, chromosomal stability, and cellular immortality. The human TERT gene (hTERT) is located at chromosome 5p15, and encompasses more than 37 kb and contains 16 exons (126). The TERT protein consists of four conserved structural domains, the telomerase essential N-terminal (TEN) domain, the telomerase RNA binding domain (TRBD), the central catalytic reverse transcription (RT) domain, and the C-terminal extension (CTE). Mutations in the RT conserved residues prevent telomerase enzymatic activity in vitro (127). These mutated TERT proteins fail to maintain telomere lengths in vivo (128), and many of these mutations have been identified in individuals with telomere-mediated disorders or telomeropathies (129). As already stated above, telomerase activity can be reconstituted by hTERC and hTERT co-expression in yeast and mammalian extracts (73, 130). Telomerase activity is established in Saccharomyces cerevisiae via reconstitution of telomerase by hTERC and hTERT co-expression (130). Therefore, hTERC and hTERT are the minimal requirement for telomerase activity (72). However, biochemical telomerase activity as measured by the telomere repeat amplification protocol (TRAP) assay does not always mean that the enzyme has necessarily telomere elongation capacity in vivo. This was demonstrated when the hTERT protein was modified by attaching a hemagglutinin (HA) epitope tag to the C terminus: while the catalytic activity of telomerase enzyme remained unaffected telomere maintenance function was lost in vivo due to loss of access to the telomere (131). Telomerase associated proteins are also essential for the full biological function of the enzyme but hTERT is the primary determinant of enzyme activity in most cells (120, 132).

#### **Dyskerin**

Dyskerin is a highly conserved, nucleolar, 514-amino-acid long protein, also known as NAP57 in rat (133) or Cbf5 in yeast (134) and has been proposed to be the third core component of the telomerase holoenzyme. Dyskerin is an essential member of the telomerase complex (but not required for biochemical telomerase activity as stated above); it binds to the telomerase RNA component (TERC) and participates in stabilizing the telomerase enzymatic complex (135). It is a pseudouridine synthase, encoded by the DKC1 locus at Xq28 (136), which is

#### TABLE 1 | Telomerase associated proteins [adapted from Hapangama et al. (14)].


*Hsp90, heat shock protein 90; CLL, chronic lymphocytic leukemia.*

responsible for the conversion of uridine to pseudouridine in non-coding RNAs, a vital step in rRNA and ultimately ribosomal synthesis (103).

Complete dyskerin depletion is lethal in mice, Drosophila (they do not have telomerase activity therefore a non-telomerase related function) and yeast (137–139). In humans, germline mutation in the DKC1 gene is the causative factor for X-linked dyskeratosis congenita (140).

#### Functions of Telomerase

Telomerase is a specialized reverse transcriptase, which maintains and elongates telomeres at the 3′ -single strand in the absence of a DNA template while using the inherent RNA (TERC) for the template function and is thus a RNA dependent DNA polymerase. In the subsequent S-phase of the cell cycle, the conventional DNA replication machinery can then replicate the complementary C-rich strand. Thus, telomerase ascertains chromosomal stability and cellular proliferation in proliferative somatic cells, tissue progenitor cells and in cancer cells (141). When telomeres shorten beyond a critical threshold length, normal healthy cells in humans which are devoid of telomerase activity, will assimilate a cellular senescence phenotype with an irreversible growth arrest and the classical morphological alterations (142). Somatic human cells lacking measurable telomerase yet expressing certain viral oncoproteins can overcome the senescence checkpoint and continue to proliferate, but they then accumulate chromosomal instability including aneuploidy, polyploidy and chromosomal fusions. On these grounds, high telomerase activity has been assigned a role in maintaining genome stability by preventing telomere shortening. Telomerase fulfills this important role via interaction with many key cellular pathways as detailed below.

#### **ATM/ATR pathway**

Ataxia-Telangiectasia Mutated (ATM) and ATM and Rad3 related (ATR) DNA damage response kinases have essential roles in telomerase-mediated telomere maintenance (66). The conserved ATM and ATR family of serine-threonine kinase proteins mediates DNA damage and replication stress checkpoint responses (143, 144), therefore, play a crucial role in DNA repair, cell apoptosis, and cell senescence, and are closely associated with the development and progression of cancer in humans (145, 146). ATM is required for the addition of new repeats onto telomeres by telomerase (147) and evaluation of bulk telomeres in both immortalized human and mouse cells showed that ATM inhibition suppressed elongation of telomeres while ATM stimulation through PARP1 led to an increase in telomere length (147).

Stalled replication forks increased telomerase localization to telomeres in an ATR-dependent manner (66). Additionally, increased telomerase recruitment was observed upon phosphorylation of the shelterin component TRF1 at an ATM/ATR target site (S367) (66) and this led to TRF1 loss from telomeres and may therefore increase replication fork stalling (148). ATM and ATR depletion reduced assembly of the telomerase complex, and ATM was required for telomere elongation in cells expressing POT11OB, an allele of POT1 that causes disruption in telomere length homeostasis (66). Hence from this data it can be concluded that ATM and ATR are involved in triggering telomerase recruitment and facilitating its assembly (66).

#### **WNT pathway**

Wnt family proteins are essential for regulating cell proliferation, cell polarity, and cell fate determination during embryonic development and tissue homeostasis (149). A dysregulated Wnt/β-catenin signaling pathway is also associated with human tumourigenesis (149). Due to the intricate relationship of telomeres and telomerase with similar cellular functions, their close interaction is not a surprise. An activated Wnt signaling pathway can reinforce the stability of telomeres by coupling and enhancing the two main telomere maintenance pathways: telomerase-dependent and ALT pathways. A Wnt-mediated telomere protective effect is particularly expected to have an important role during development, in adult stem cell function and oncogenesis (65).

The Wnt pathway may regulate telomere maintenance via its effect on several essential shelterin components, including TRF2 and POT1. Recently, in human somatic and cancer cells as well as in mouse intestinal tissue, activation of canonical Wnt/β-catenin pathway activated TRF2 and also increased telomere protection were demonstrated (65). In mice lacking telomerase, apoptosis of the Wnt-dependent intestinal crypt stem cell niche could be rescued by administration of Wnt agonists (150). Additional evidence demonstrates that the Wnt pathway triggers APC- and β-catenin induced regulation of TRF2 and TCF4 which further regulate TRF1 and POT1 (150, 151).

Further to the enhancement of shelterin protection, the Wnt/β-catenin signaling pathway also activates TERT (152). Importantly, the use of Wnt pathway agonists can rescue telomere uncapping, suppress apoptosis and lead to elevated Ascl2 transcripts as well as Sox9 protein levels (150) suggesting a therapeutic strategy for some conditions with aberrations in telomerase.

#### **Non-canonical functions of TERT**

Non-canonical functions of TERT have been discovered later than telomerase activity, and they also play a role in tumorigenesis, for example via TERT's role in regulating the Wnt signaling as a cofactor for the β-catenin pathway (153). TERT has been shown to be inducible in ischemic brain cells and to prevent apoptosis via a non-telomeric action via shift of the cytosolic free Ca2<sup>+</sup> into the mitochondria (154). Despite having normal telomere lengths, lack of hTERT impairs the cellular capability to repair damaged DNA and fragmented chromatin (155). TERT also is demonstrated to have RNA dependent RNA polymerase function by interacting with the RNA component of mitochondrial RNA processing endoribonuclease (RMRP) and forming ribonucleoprotein complexes. These complexes produce double-stranded (ds) RNAs that serve as substrates for the generation of siRNA which may regulate the expression of other genes related to stem cell biology (156). Further to the above, there are many other additional non-telomeric functions of TERT active in cancer, such as improved DNA repair, increased apoptosis resistance, changes in chromatin structure and altered gene expression (157).

# Hormone Regulation of Telomerase in Hormone Responsive Tissues

There is evidence from multiple studies that telomerase is under the regulation of steroid hormones in hormone responsive tissues. This corroborates with the known direct regulation of cell fate and proliferation in such tissues by steroid hormones, for example the ovarian hormone, estradiol, induces a mitotic response in endometrial epithelial cells (23, 158). In different studies, telomerase is induced by estrogen in various macaque and human cell lines (15, 159, 160). Androgens also upregulate telomerase in an ovarian cancer cell line (161) but progesterones down regulate telomerase in the endometrium (15). ATM silencing also down regulated proteins, such as ChK2, p53, and caspase 3, which were stimulated by the synthetic progestogen, medroxyprogesterone acetate (MPA) (162). This result suggested that MPA exerts its effects via the ATM-Chk2 p53-caspase-3 pathway protecting against carcinogenesis (162). The progestagenic effect on telomerase may also be mediated through this pathway. Hormonal regulation of telomerase in the healthy endometrium was recently reviewed in detail (14).

### Telomerase-Related Telomere Regulation by TERRAs

Telomerase regulation by TERRAs has initially been examined in yeast although recent work also suggests a similar regulation in human cells. In yeast cells, TERRAs were found to sequester and direct telomerase to the specific telomeres which were the shortest (68). In addition, TERRA was found to bind to hTERC and hTERT components of telomerase independently, to function as an inhibitor of human telomerase enzyme (67). In telomerase negative cells with shortened telomeres, increase in TERRA levels trigger homology directed repair (HDR) whereas in telomerase positive cells, it results in recruitment of telomerase to the short telomeres (163). Absence of both telomerase and HDR accelerates the cell senescence pathway (164). Due to loss of Rat1 function, in yeast free TERRA accumulates at critically short telomeres which helps in recruiting the telomerase enzyme to that telomere and elongation of that telomere (165).

TERRA was found to be induced in cells with short telomeres and acted as a scaffold for spatial organization of the telomerase components forming a TERRA-telomerase complex which helped in recruitment of telomerase to the telomere of its origin hence TERRA was proposed to be a recruiter of telomerase enzyme rather than an inhibitor (68). Contrary to some in vitro studies, in human cancer cells, telomerase-led telomere elongation was not affected by the transcription of the telomere. In these cells, it was suggested that shortening of telomeres may not have been due to telomerase inhibition, but due to impaired replication due to integrity of the chromosomes affected by high levels of TERRAs (166). In general, the interaction of TERRAs and telomerase is complex and might depend on cell type and conditions, such as cell cycle phase, or telomere length.

# Telomere Maintenance by Alternative Lengthening of Telomeres (ALT)

Cells can maintain their telomeres via a telomerase dependent pathway or a telomerase independent ALT pathway (69). New telomeric DNA is synthesized from a DNA template in ALT (167) by homologous recombination (HR) (168). The template could either be the telomere of another chromosome, another region of the same telomere by t-loop formation or sister telomere recombination.

The first evidence for the presence of an ALT mechanism was described in several immortalized human cell lines that did not have telomerase activity but maintained telomere lengths for hundreds of population doublings, and this mechanism occurs in ∼15% of cancers including osteosarcomas, soft tissue sarcoma subtypes, and some glial brain tumors (169, 170).

In human cells, where ALT activity is elevated to a degree sufficient for telomere length maintenance, telomeres are characterized by their highly heterogeneous length, but the average length (>17 kb) is about double that of most cells where telomeres are elongated by telomerase (171).

Mutations in the ATRX/DAXX chromatin remodeling complex have been observed in cancers and cell lines that use the ALT mechanism, suggesting that ATRX may suppress the ALT pathway (172). In mortal cells or immortal telomerasepositive cells, knockout or knockdown of ATRX does not stimulate ALT (172). However, ATRX loss in SV40-transformed fibroblasts together with one or more unidentified genetic or epigenetic alterations was attributed to either a marked increase in the proportion of cells with an activated ALT (instead of telomerase) or significant decrease in the time taken for ALT activation (172). Loss of ATRX protein and mutations in the ATRX gene are also characteristic features of ALT-immortalized cell lines (172). In addition, ALT is associated with marked genome rearrangements, extensive micronucleation, a defective G2/M checkpoint and alteration in double-strand break (DSB) repair (173).

#### ROLE OF TELOMERES AND TELOMERASE IN PRE-MALIGNANT AND MALIGNANT PROLIFERATIVE DISORDERS

#### Alteration of Telomere Biology in Premalignant Conditions and in Cancers

Limitless proliferation is a cardinal feature of cancer cells, whist increased proliferation is common to all premalignant changes including hyperplasia. The excessive proliferation observed in these malignant/premalignant conditions is maintained by avoiding senescence and crisis/apoptosis. Senescence/apoptosis exist as barriers for mitosis, thus they are tumor suppressor mechanisms in normal cells, which are regulated intricately by telomeres and checkpoint activation (**Figure 3**). The unrestricted proliferation of cancer cells is therefore thought to be sustained by telomere maintenance mechanisms which were detailed above. Since high telomerase activity is reported in over 85% of cancers, telomerase dependent telomere lengthening is believed

to be the most common telomere maintenance mechanism relevant to carcinogenesis.

#### Evidence for Altered Telomere Lengths in Cancers

During ongoing proliferation in normal somatic cells without telomerase or other telomere-maintenance mechanisms, telomeres shorten until reaching a certain minimal length. Beyond this, when tumor suppressor checkpoints, such as p53 are functioning, senescence or apoptosis can be induced. In contrast, when p53 or other important DNA damage checkpoints are not functioning, cells can enter a crisis state where ongoing proliferation promotes further telomere shortening and telomere dysfunction (174). This can cause various genomic instabilities, such as end-to-end fusion of telomeres resulting in anaphase bridges in subsequent cell division cycles. Most of these cells usually die due to apoptosis and gross genomic instabilities. However, some rare cells acquire mutations in the TERT promoter that increase telomerase activity resulting in restabilization of telomeres, Importantly, as long as telomeres are capped and protected, they can be rather short and this situation is frequently found in epithelial cancer cells compared with adjacent healthy tissue. Several studies using telomere PNA-FISH have shown that breast, prostate, and pancreatic cancers are associated with telomere shortening (175–177). Furthermore, around 40 to 97% of colorectal tumors have shorter telomeres compared with normal tissue, and telomere shortening is therefore considered to be one of the early events in tumourigenesis (178, 179).

However, importantly, acquiring telomerase activity can stabilize even short telomeres in genetically unstable cells and provide sufficient capping for them to attain an unlimited proliferation potential. Thereby, telomerase re-activation conserves genomic mutations and instabilities and contributes further to tumourigenesis (**Figure 2**).

Significant telomere length shortening results in end-to-end fusion, thus increasing the potential for genome instability and carcinogenesis. There are few other generic associations which lead to telomere attrition, such as oxidative stress, lifestyle choices, environmental factors, smoking and obesity (180) and some of these also increase the risk of developing a variety of cancers. Telomere shortening can influence the progression of premalignant breast tissue to malignancy and premalignant breast lesions had short telomeres leading to nonclonal chromosome aberrations (181).

Meta-analyses of available studies also revealed that shorter peripheral blood mono-nucleocyte (PBMC) telomeres are associated with a significant increase in the risk of developing cancer (OR = 1.35, 95% CI = 1.14–1.60) than longer telomeres (182, 183). Shorter PBMC telomeres could be related to oxidative stress endured by an organism, which is in agreement with the established mediatory role that oxidative stress plays between inflammation and cancer (184). When PBMC mean telomere lengths were prospectively studied in the general population in Denmark, shorter telomere lengths were also associated with decreased survival after cancer rather than the cancer risk itself (185). Another systematic review has also reported a consistent inverse relationship between age and PBMC telomere length (186).

Telomere dysfunction may also be a resultant of altered telomere-associated proteins that are also essential for regular end-capping function (187, 188). For example, mutations in the C-terminal of POT1 can initiate genomic instability permissive for tumourigenesis (189). TRF1 flox/flox × K5-Cre transgenic mice, do not have TRF1 in stratified epithelia. These mice demised perinatally and showed skin hyperpigmentation and epithelial dysplasia and were associated with telomere initiated DNA damage, p53/p21 and p16 pathway activation and in vivo cell cycle arrest. Deficiency of p53 rescues mouse survival but causes increase in the incidence of squamous cell carcinomas (39). Alteration of the levels of TRF1, TRF2, TIN2, and POT1 has also been described in some human tumors (190). A dysregulated expression of TRF1, RAP1, and TPP1 has been reported in patients with chronic lymphocytic leukemia (191). Likewise, TIN2, TRF1, and TRF2 mutations have been associated with some cases of Dyskeratosis congenita and aplastic anemia (192– 195) and both these conditions increase the risk of developing some cancers. Defects in shelterin components naturally cause dysregulation of telomere homeostasis as explained above. This may operate as a tumor suppressor mechanism when it initiates the p53/pRb pathways which in turn triggers senescence and prevents the tumorigenesis process. Alternatively, it can contribute to carcinogenesis with the fusion of dysfunctional telomeres or fusion between dysfunctional telomeres and double strand breaks which trigger breakage-fusion-bridge cycles (196). In hepatocellular carcinomas, longer telomeres, increased hTERT expression and higher levels of TRF2 protein as "stemness markers" were associated with poorer prognosis and more chromosomal instability (197). Further studies have confirmed that different causal factors, such as hepatitis B and C, and alcohol lead to telomere dysfunction in hepatic cells hence initiating the carcinogenesis process (198). A significant decrease in POT1 and RAP1 protein levels are described in familial papillary thyroid cancers (199). TP53 disruption in hematological malignancies has been associated with the downregulation of expression in shelterin genes and severe telomere dysfunction and genomic instability (200). Therefore, genetic mutations resulting in functional alterations in the essential components of the telomerase enzyme or shelterin components may repress telomerase activity and thus shorter telomeres will be the consequence. The available evidence also suggests a concerted dysregulation in the expression of shelterin genes and protein levels with the commonly observed removal of cellular tumor suppressor mechanisms in premalignant conditions can lead to alteration in telomere lengths that can trigger the tumourigenesis process.

# Evidence for Altered Telomerase in Cancers

#### **Polymorphism in genes of the telomerase complex**

Such as hTERT and hTERC has been reported to affect individual susceptibility to cancers (201, 202). Variants in chromosome 5p15, the region that harbors the hTERT gene, have been identified by Genome-wide association studies (GWAS) to be associated with the risk of bladder, pancreas, brain, testicular, breast, prostate, skin, and lung cancers (203–207).

#### **hTERT promotor mutations**

Tumors with high hTERT promoter mutation frequencies have almost always originated in tissues with relatively low cell turnover rates. Contrastingly, tissues with rapid cell turnover seem to have different mechanisms to elongate telomeres and seem less likely to benefit from activating hTERT expression by mutations (208). Mutations that result in increased hTERT expression, telomerase activity or longer telomere lengths have been identified in cancers of the central nervous system, thyroid, bladder, liver, tongue, adipose tissue and skin (208–210). In thyroid cancers, when hTERT and BRAF mutations coexist, such tumors express high levels of hTERT (211).

Common inherited variants of telomere related genes, such as TERC, TERT, and rare POT1 mutations have been found to be associated with higher risk of developing gliomas. TERT promoter and ATRX mutations were found to be the most recurrent somatic events which led to glioma associated lengthening of telomeres (212).

A high frequency of hTERT promoter mutations was also reported in follicular cell-derived thyroid carcinomas (213). An over-representation of hTERT promoter mutations had been detected in advanced thyroid cancers and these mutations were more prevalent in advanced disease (51%) compared with welldifferentiated tumors (22%). Thus, hTERT promoter mutations have been suggested as biomarkers of tumor progression (213). hTERT promoter mutations usually cause an increased expression of the hTERT gene and paradoxically, these mutations were reported to occur together with short telomeres in tissues with low-rates of self-renewal and were also associated with poor patient survival in primary melanomas (210). Tissue stem cells are reported to have active telomerase and daughter cells produced by these switch off telomerase upon differentiation, and subsequent reactivation of telomerase in these tissues have been proposed to be the reason for the observed short telomeres in thyroid cancers with high telomerase expression (210). Rachakonda et al. showed that mutations of the hTERT promoter were also the most common somatic lesions in bladder cancer (214). The authors also found that a common polymorphism rs2853669 in the hTERT promoter acts as modulator of the mutations effect on survival and disease recurrence. The patients with the mutations had poor survival outcome in the absence but not in the presence of the variant allele of the polymorphism. The mutations without the presence of the variant allele were markedly correlated with tumor recurrence in patients with non-invasive and invasive T1 bladder tumors (214). Polymorphisms in the hTERT gene were also associated with an increased lung cancer risk in the Chinese Han population (215).

#### **Telomerase activity in cancers**

The early observation that telomerase activity is absent in most human somatic tissues during differentiation but strongly upregulated in tumors, agrees with the hypothesis that telomerase playing an important role in the carcinogenesis process (216). In pancreatic ductal cell carcinoma, levels of telomerase activity were higher compared to other types of pancreatic cancer and benign pancreatic tissues (217). In gastric cancers, tumors with high telomerase activity had poorer prognosis and the authors concluded that detecting telomerase activity might be useful as a prognostic indicator of clinical outcome (217). Telomerase activity was also detected in 90% of head and neck squamous cell cancers, in 100% hyperplastic squamous epithelium but not in normal mucosa (218). Colorectal cancers with high telomerase activity had poorer prognosis in spite of curative surgery in apparently disease free patients, thus the survival seems to have been associated with the level of telomerase activity (219). A systematic analysis of telomerase activity levels in many cancer types performed by Bacchetti and Shay in 1997 demonstrate high telomerase being a common observation in most of them (220).

**hTERC alterations in cancer**: Recent work has proposed that hTERC maturation involves the poly(A)-specific ribonuclease (PARN) which is localized in the nucleolus and in the Cajal body (CB). The enzyme trims hTERC precursors by removing poly (A) tails and may be involved in impairment of telomerase activity (221). Individuals with biallelic PARN mutations and PARN-deficient cells showed a reduction of expression of genes encoding several key telomerase components, such as TERC, and DKC1. These cells also have critically short telomeres (222). Improper hTERC processing and telomere dysfunction in premalignant diseases, such as Pontocerebellar Hypoplasia 7 (PCH7) and dyskeratosis congenita had been proposed to have a mechanistic link (221). hTERC amplification was associated with the aggressive progression of cervical cancer, and authors suggested that hTERC may serve as a surrogate marker for cancer progression and form a potential therapeutic target for cervical cancer (223). However, it is important to appreciate that most cervical cancers initiated in a background of persistent papilloma virus infection in the transformed epithelial cells. hTERC over-expression has been reported in many other cancers including prostate (224); breast (225); and oral squamous cell carcinoma (226).

**Dyskerin alterations in cancer:** Dyskeratosis congenita is a rare multisystemic syndrome characterized by low telomerase activity already during development and consequently, shorter telomeres in many tissues resulting in a high susceptibility to develop a subset of cancers, therefore, wild type dyskerin protein has been suggested to act as a tumor suppressor. Conversely, wild-type dyskerin protein is upregulated in a number of human cancers, such as in breast, prostate, colon and hepatocellular carcinomas (108, 227–229) and in these cancers, high levels of dyskerin were associated with an aggressive histopathological feature and poor prognosis (229). Acute loss of dyskerin function by RNA interference led to marked reduction of steady-state levels of H/ACA RNAs, disruption of the morphology and repression of anchorage-independent growth of telomerasepositive and telomerase-negative human cell lines. The levels of dyskerin in cancer cells modulate telomerase activity through the regulation of TERC levels, independently of TERT expression (227). The function of telomerase associated proteins in cancer is summarized in **Table 1**. Dyskerin might also contribute to tumor development through mechanisms where the presence of cellular telomerase activity is not essential, and which may be only partially dependent upon the protein's role in rRNA processing (104).

#### ENDOMETRIUM

The endometrium is the inner mucosal lining of the uterus that contains several cell types including tissue specific epithelial and stromal cells, as well as leucocytes and blood vessels (22, 230–233). It is the primary target organ for ovarian steroid hormone action (24) and healthy human endometrium is characterized by its regenerative and remodeling capacity that undergoes repetitive monthly cycles of proliferation, secretory changes, break-down and regeneration. These cycles of changes occur ∼400 times in a female's reproductive life (22, 230) and are regulated by ovarian steroid hormones (23). Telomerase activity as well as mean telomere length change according to ovarian cycle in whole healthy endometrial samples (15, 234) suggesting an ovarian regulation and correlation with proliferative activity (15). Epithelial cells demonstrated significantly higher telomerase activity, but contrastingly, shorter telomeres compared with stromal cells across the cycle (14, 15) (**Figure 4**). In the endometrium, Estrogen upregulates telomerase activity. Whilst progesterone inhibits telomerase activity and hTERT expression (15). The telomere and telomerase biology of normal endometrium has recently been reviewed in detail (14).

#### The Role of Telomeres and Telomerase in Benign Endometrial Disorders (Table 2)

The role of telomeres and telomerase in benign endometrial disorders was recently reviewed in detail in Hapangama et al (14). There are various benign gynecological disorders, such as endometriosis (243), recurrent reproductive failure, subfertility with reported abnormal telomerase activity and telomere length aberrations (13, 235). High telomerase activity, high hTERT mRNA and protein levels with longer mean endometrial telomere lengths are characteristics of the eutopic secretory endometrium (13, 235, 242, 244), whereas epithelial cells of ectopic lesions also demonstrated longer mean telomere length (15).

The progesterone dominant window of implantation in healthy women has shown virtually no hTERT immunoreactivity (235) and lowest telomerase activity (13, 234). However, immunostaining for hTERT was significantly and differentially increased in various endometrial cellular compartments in women with recurrent reproductive failure (235). These observations suggest that particular aberrations in cellular proliferation or causative dysregulation of telomerase to be important in endometrial pathologies. Furthermore, normal telomerase activity seems to play a pivotal functional role in ensuring normal endometrial function.

# Alteration of Telomere Biology in Endometrial Premalignant Conditions and in Endometrial Cancer

#### Endometrial Hyperplasia

Endometrial epithelial hyper-proliferation with increased glandular to stromal cell ratio is defined as endometrial hyperplasia. Pathogenesis of endometrial hyperplasia is virtually always associated with relative predominance of the mitotic estrogen signal, due to direct excess of Estrogen or due to insufficient levels of progesterone (24). Anovulatory cycles in premenopausal women, extra-ovarian aromatization of adrenal androgens in to estrogenic compounds in obese women and iatrogenic interventions, such as Tamoxifen and Estrogen only hormonal replacement therapy are common examples

group 2. (C) Epithelial cells maintained in long-term culture had a phenotype consistent with senescence. Note the enlarged cells and positive blue stain for β-galactosidase in the micrographs (representative of *n* = 5). (D) Epithelial cells were isolated from an adenocarcinoma of the human endometrium and maintained in culture as a cell line. The cells were treated with the concentrations of Imetelstat indicated for 72 h prior to TRAP. TRAP activity is expressed as a percentage relative to the activity of the mismatch control (mean ± SEM for *n* = 3 separate experiments). (E) Epithelial cells were maintained in culture for up to 3 days and then treated with 1µM Imetelstat or mismatch control oligonucleotide for a further 72 h prior to TRAP assay. TRAP activity is expressed as a percentage of the mismatch control (*n* = 4). *T*-test, \**p* = 0.02. (F) EEC (*n* = 5) were directly seeded into 96-well dishes, allowed to attach and treated the next day with Imetelstat or the mismatch control at the concentrations indicated for 72 h. Cell viability was assessed by MTT assay. Note significant loss in cell viability at 100µM (Mann Whitney test, *p* = 0.002). (G) Cultures of normal epithelial cells and an adenocarcinoma of the endometrium treated with Imetelstat or mismatch control as before, and immunoblotted for phospho-H3 [phosphohistone H3 (Ser10)]. Histone H3 is only phosphorylated on Ser 10 during mitosis. Shown is a representative blot (top) of normal epithelial cells (*n* = 5) and the adenocarcinoma (representative of two separate experiments) and densitometric analysis (bottom). *T*-test, \*\**p* = 0.009. (H) Stromal cells were grown for 24 h and then treated with 1µM Imetelstat or mismatch control oligonucleotide for 72 h prior to TRAP assay. Telomerase activity is expressed as a percentage of the mismatch control. *T*-test, \*\*\**p* = 0.0004. This previously published figure in human reproduction (15) is reused with permission.

of conditions related to endometrial hyperplasia. Importantly, the premalignant endometrial hyperplasia, which includes the category of atypical hyperplasia/endometrial intraepithelial neoplasia according to the 2014 World Health Organization (WHO) classification is the typical precursor of endometrioid endometrial cancers (245).

#### **Alterations in telomere lengths in endometrial hyperplasia**

The involvement of telomere shortening in chromosomal instability has been associated with the initiation of carcinogenesis (246). There are only 2 studies that have examined telomere lengths in endometrial hyperplasia. A study using a telomere-FISH (telo-FISH) assay to measure telomere lengths, compared chromosomal arm loss or gain in premalignant endometrial lesions with normal endometrium, and reported telomere lengths to be stable with the pathological transformation in endometrial hyperplasia and in endometrial carcinoma (247). Albeit using a small sample size, the authors conclude that unlike in cervical precancerous lesions, endometrial hyperplasia did not have widespread chromosomal alterations, implying that endometrial carcinogenesis involves mechanisms distinct from those of cervical carcinogenesis, which is almost always induced by viral infection (247). However, close scrutiny of the data presented on different endometrial hyperplasia subtypes suggested that atypical endometrial hyperplasia may be associated with higher telomere length heterogeneity. This may be also suggestive of the involvement of ALT mechanism in this premalignant condition, but larger studies are needed to confirm the ALT mechanism in the true pre-malignant endometrial hyperplasia subtype with atypia. Importantly, the analysis method utilized in the Maida study did not allow inter-patient comparison of tissues samples (of different women) but was only suitable to compare adjacent cells of a single tissue sample. Therefore, the study presented insufficient data to conclude if there was a definite change in the telomere length in precancerous endometrial hyperplasia when compared with either normal or cancerous endometrium.

By using a three-dimensional (3D) imaging technique, a specific 3D arrangement of telomeres was revealed in tumor cell nuclei (248). Unlike the non-overlapping nature of telomeres in normal nuclei, telomeres of cancer nuclei have the tendency to form aggregates (248). Different numbers and sizes of such telomere aggregates can be found in tumor nuclei (248). Telomere aggregate formation does not depend on telomere length and telomerase activity (249).

The existence of telomere aggregates in precancerous lesions, such as in human cervical intraepithelial neoplasia supports the notion that changes in the organization of the 3D nucleus may facilitate tumorigenesis (250). The "telomere-driven genomeinstability" can happen as a result of the close contiguity of telomeres forming aggregates of different numbers and sizes that increase the risk of end-to-end telomeric fusions followed by cycles of breakage-bridge-fusion (249). A significantly increased number of telomere aggregates was observed in atypical hyperplastic cells in a mouse models which is also a specific feature of cancer cells. Moreover, the PTEN heterozygous mouse model further demonstrated that 3D telomere architectural changes occur before the complete loss of PTEN and prior to the development of histological characteristics of atypical hyperplasia and endometrial carcinoma (251). Therefore, the presence of telomere aggregates in hyperproliferative lesions with atypical nuclei may render them to be precancerous changes. Further studies including larger sample size and both types of endometrial hyperplasia are warranted to examine and conclude on changes in telomere length in precancerous endometrial hyperplasia lesions.

#### **Telomerase in endometrial hyperplasia**

High hTERT levels and elevated telomerase activity were reported in all types of endometrial hyperplasia, including simple, complex and complex with atypia subtypes (252– 256). This early observation prompted some investigators to propose that telomerase activity measured by TRAP assay to be a suitable tool to screen the endometria of postmenopausal women with post-menopausal bleeding (257). The authors proposed that this method will determine endometrial premalignant and malignant conditions (257) from benign endometrium, since telomerase activity was rarely detected in normal post-menopausal women, while the majority of endometrial hyperplasia and cancers contained high telomerase activity. However, there are other studies that reported a lack of measurable telomerase activity by TRAP assay in benign endometrial hyperplasia (258). Further work also found that it was possible to use hTERT immunohistochemical (IHC) analysis (259) as a marker for premalignant (atypical) endometrial hyperplasia. However, it is difficult to conclude on the diagnostic feasibility of telomerase activity or hTERT protein (IHC) in endometrial hyperplasia considering these studies, because of the inadequate sample sizes which were only n = 12 atypical endometrial hyperplasia in Brustmann (259) and n = 18 simple and atypical endometrial hyperplasia in Maida et al. (257) and Brustmann (259). In addition, the studies did not clarify whether the existence of endometrial hyperplasia cells were confirmed in the analyzed samples, particularly with TRAP assay and since endometrial hyperplasia can co-exist with either normal or cancerous endometrium, this may affect the results. Progesterone is one of the main current pharmacological therapies for treating endometrial hyperplasia (24) and telomerase being a (albeit indirect) downstream target of progesterone in the endometrium is of interest. This justifies future studies exploring the therapeutic utility of directly targeting telomerase in the treatment of endometrial hyperplasia.

#### Endometrial Cancer

Traditionally, EC had been divided into two major groups: estrogen-dependent, type-I (endometrioid type) and estrogenindependent, type-II (non-endometrioid), with the former accounting for 80% of ECs. Five-years survival rates are exceptionally poor for advanced type-I and type-II (high grade) EC at 23% which is a far worse rate than for most other common cancers, such as breast cancer (CRUK). However, the recent trend had been to apply for an alternative classification system that more accurately defines ECs into prognostically distinct molecular subtypes that reflect the underlying molecular alterations with well-described underlying genomic aberrations (260). EC is a disease of post-menopausal women, however, obesity associated unopposed estrogen action is an established cause for the trend toward increasing incidence of this cancer even in younger women (23, 24, 261). ECs are hormone responsive tumors and even high grade ECs retain some



*(Continued)*

TABLE

2


 biology in benign endometrial

 disorders, telomerase,

 and telomere length.


#### **Evidence for telomere alterations in endometrial cancer (Figure 5) (Table 3)**

A study in 1992 found that endometrial adenocarcinomas have reduced telomeric repeat sequences suggesting shorter telomeres compared with normal tissue (262). A decade later a second study demonstrated changes in telomere lengths in 17/23 (73.9%) of endometrial cancers using a Southern blot technique (269). Another study by Menon and Simha (273), using the same telomere restriction fragment (TRF) measurement, found that mean TRF lengths became shortened when normal endometrium underwent neoplastic changes (273). A study which used a telomere-oligonucleotide ligation assay demonstrated erosion of the telomere overhang length, rather than overall telomere length, and proposed that this might play a role in endometrial carcinogenesis and may be related to tumor aggressiveness (274). All these studies utilized techniques that assess the average telomere length values of a tissue sample. However, when endometrial samples were harvested and frozen, they did not examine if the proportion of the endometrial sample examined for telomere length actually contained cancerous cells. Subsequently, 12 years ago, Maida et al. (247) employed a telomere-FISH (telo-FISH) assay that assessed the relative telomere length in normal and pathological cells in intact tissue at the cellular level and no significant difference was found between the telomere length of normal endometrium and endometrial cancer (247). That study however did not specify the normal cell type that they used as the control (stromal/epithelium) and included only adenocarcinomas (Type I). A similar, but slightly modified version of telomere chromogenic in situ hybridization method was subsequently used by Akbay et al. and the authors demonstrated a significant telomere shortening in both type I and type II endometrial cancers in comparison with normal stromal cells (270). They also reported that the adjacent normal stromal cells were compared with epithelial cancer cells to demonstrate telomere shortening only in type II cancers. The authors expanded the study to confirm their hypothesis in a rodent model. These animals were generated with shortened telomeres to show that telomere attrition contributes to the initiation of type II endometrial cancers and progression of Type I endometrial cancers (270). This is of interest, but caution should be taken when interpreting these results, as the endometrial stromal cells are known to possess longer telomeres when compared even with healthy epithelial cells (14, 15) and that has been hypothesized to be due to the difference in the cell proliferation rates, telomerase activity levels and different regulation of telomere maintenance in these two cell types (14). Therefore, the data may simply reflect cell type specific difference in relative telomere lengths but not demonstrating a true endometrial cancer associated change in telomere lengths. Hashimoto et al. (274) found that endometrial cancers show short 3 ′ single-strand telomeric overhang length compared to normal endometrium (274). They also found that poorly differentiated cancers or deeply invading endometrial cancers had a longer overhang length in comparison with well-differentiated cancers

*glyceraldehyde*

 *3-phosphate*

*dehydrogenase;*

 *EECs, endometrial*

 *epithelial cells; RIF, recurrent implantation*

 *failure.*

or superficial invading cancers and this may suggest that the 3′ overhang may have a role in tumor progression (274).

A recent paper that considered germline genetic variants in a genome wide association study (GWAS) as instrumental variables to appraise the causal relevance of telomere length for the risk of cancer, demonstrated that their predicted increase in telomere lengths was strongly associated with some specific cancers, such as gliomas, low grade serous ovarian cancers, lung adenocarcinomas, neuroblastomas, bladder cancers, melanomas, testicular cancers, and also endometrial cancers (275). However, this study did not measure the exact telomere length of the tissue of origin of cancers but assumed the particular genetic variance may promote longer telomere lengths. With that assumption, the authors calculated a stronger association of presumed longer telomere lengths and rarer cancers and cancers with a lower stem cell division rate (275). However, this data should be considered with caution, since age associated tissue/cell specific telomere length change is a well-established fact but that was not considered by the authors. Therefore, the postulated prediction in telomere length change may be relevant to the effect of genetic variants that were examined, in increasing cancer risk, but it does not provide direct or compelling evidence for a role for tissue telomere length change in endometrial carcinogenesis. When telomere lengths were estimated for cancer cohorts in The Cancer Genome Atlas (TCGA) dataset; sarcomas, testicular germ cell tumors and low grade gliomas were associated with longer telomeres whilst cervical and endometrial cancers had shortest average telomere length (276). This observation has also been explained as a result of some tumors having high telomerase activity, thus shorter telomere lengths that are stabilized [e.g., in testicular tumors (277)], and others have long telomere lengths accompanied by increased activity of the ALT mechanism (e.g., in low grade gliomas and sarcomas). Longer telomere length in PBMC has also been associated with a significantly increased risk of endometrial cancer in a group of Caucasian Americans (272). Since endometrial cancers are known to have high telomerase activity, the ALT mechanism is less likely to be active in those cancers. Considering the above evidence, it is likely that endometrial cancers have relatively shorter telomere lengths that are maintained by high telomerase activity compared with normal tissue. Further studies are warranted to examine subtype specific telomere length aberrations and the relationship of telomere lengths with the telomerase activity in the different types of endometrial cancers.

The protein and/or mRNA levels of the most conserved out of all shelterin proteins, POT1 (26) were increased in many different cancers including gastric, thyroid, breast (199, 278, 279) and in endometrial cancers (280). Higher levels of point mutations in the POT1 gene were observed in endometrial cancers, revealing that genetic variations in POT1 may lead to carcinogenesis in the endometrium (280). Simultaneous conditional inactivation of the shelterin protein POT1a with the tumor suppressor p53 in endometrial epithelial cells in a murine model, induced type II metastatic adenocarcinomas in 100% of the animals by 15 months (281). This suggests that telomere dysfunction and loss of tumor suppressor genes can produce Type II endometrial cancers. This will obviously need to be accompanied by telomerase re-activation observed in most endometrial cancers supporting the cancer-associated increased cellular proliferation. The loss of POT1 proteins activates ATR (282) and ATR activation requires Replication Protein A (RPA), which binds single stranded (ss) DNA (282); the POT1-TPP1 heterodimer protects telomere ends from being detected as DNA damage by excluding RPA from binding telomeric ssDNA. Therefore, the loss of POT1 described in endometrial cancer may cause inappropriate telomere access of telomerase resulting in compromised telomere capping and sustained telomere dysfunction facilitating genetic instability.

There are no published studies examining the expression or function of other shelterin proteins or TERRAs in EC to date.

#### **Evidence for a role of telomerase in endometrial cancer (Figures 5, 6)**

Kyo et al. examining 13 endometrial cancers and 5 cell lines derived from endometrial cancers using a Telomerase Repeated Amplification Protocol (TRAP) assay reported that 92% of cancer samples displayed detectable telomerase activity (263). At that point in time, the general consensus was that only specialized cells or cancer cells would have detectable telomerase activity. A year later, the same group increased their endometrial samples to 17, included 60 normal endometrial samples, and reported that being a somatic organ, the benign human endometrium, expresses dynamic levels of telomerase activity (measured by TRAP assay), with the highest levels observed in the late proliferative phase endometrium which was comparable to endometrial cancer. They also indicated that endometrial telomerase levels are closely associated with proliferation and likely to be regulated by estrogen (264). During the same year, Saito et al. examined a larger and more diverse endometrial cancer sample set and reported that activation of telomerase was found in most of these cancers, similar to the reports on gastric, prostate, bladder, and skin cancers (252, 283–286). Saito et al. further confirmed the earlier work by Kyo et al. that 28/30 endometrial cancers had high telomerase activity and late proliferative phase to have the highest telomerase activity levels in the benign endometrial samples. Additionally, the authors found that endometrial hyperplasia demonstrated high telomerase activity similar to cancer, whereas no activity was detected in healthy post-menopausal endometria with or without bleeding problems, indicating telomerase activity to be a suitable diagnostic test for identifying post-menopausal endometrial pathology (252). The authors also noted that telomerase activity was increased by estrogen which induced cell proliferation and was reduced in progesterone dominant conditions, indicative of an ovarian steroid hormonal regulation. The finding of high telomerase activity in endometrial cancers has been subsequently confirmed by many other groups (15, 25, 159, 239, 247, 267– 269). In addition to the high telomerase activity measured by the gold standard test, the TRAP assay, some authors studied expression levels of components of the telomerase holoenzyme using qPCR to detect gene expression levels. They concluded that hTERT levels correlated well with TRAP assay data (159, 268) and both seem to be related to endometrial epithelial proliferation (15). In a relatively small study, Bonatz et al. (287) have shown a significant correlation between higher telomerase activity and higher International Federation of Gynecology and Obstetrics (FIGO) stage and grade, suggesting that telomerase activity is increased in advanced stages of endometrial cancer (287). In their study, Wang et al showed that 82% of their endometrial cancer samples had telomerase activity but they did not find any correlation between telomere lengths and telomerase activity in different gynaecologic cancers (cervical, ovarian and endometrial) (269).

Detection of hTERT mRNA in peripheral blood (PBMCs) has been reported to be significantly higher in women with EC compared to patients with benign uterine diseases and healthy controls. Using a relatively moderate sample size (n = 56 patients with endometrial cancer, n = 40 patients with benign uterine diseases and n = 40 healthy control) the authors claimed that the exact levels of hTERT mRNA will demarcate those with metastatic disease thus may be useful in stratifying patients for adjunctive therapy (288). This claim needs to be confirmed in a future study which includes an adequate sample size.

Recently, in two progesterone responsive and progesteroneinsensitive human endometrial cancer cell lines (162), ATM protein was shown by reverse-phase protein array (RPPA) to participate in progesterone stimulation to suppress carcinogenesis in the endometrium (162). Additionally, a progressive loss of ATM levels from hyperplasia to the lowest levels was observed in type 1 endometrial cancer lesions and there was a negative relationship of the pathological grades and ATM levels (162).

Activating hTERT promotor mutations do not usually occur in a background of loss of the tumor suppressor protein ARID1A (289). Recent data suggest that ARID1A negatively regulates hTERT transcription and telomerase activity; while induction of ARID1A represses transcription and histones via occupying SIN3A and H3K9me3 sites (290). ARID1A is a member of the SWI/SNF chromatin remodeling complex, and it is frequently mutated in endometrial adenocarcinoma (291), therefore it is conceivable how hTERT might be upregulated in the endometrial cancer with loss of ARID1A.

In endometrial cancer cell lines, telomerase activity and expression of hTERT were both increased by estrogen in an estrogen receptor alpha (ERα) dependent and estrogen responsive element (ERE) dependent effect in the hTERT promoter (292). Additionally, a previous study showed that estrogen also induced telomerase activity via post-transcriptional Akt dependent phosphorylation of hTERT in human ovarian cancer cell lines (293).

Lehner et al. (268) compared hTERT mRNA levels and telomerase activity using TRAP assay in normal endometrium with endometrial cancer and they concluded that the levels and activity were significantly higher in cancer and low in normal endometrium during the secretory phase of the menstrual cycle as well as in atrophic endometrium (268). Thus, they suggested that quantitative analysis of these parameters may be useful as markers for diagnosis of endometrial cancer.

PTEN regulates telomerase activity, most likely through its known effects on the PI3-kinase/Akt pathway (294). Reconstitution of PTEN in the PTEN-null Ishikawa endometrial cancer cells resulted in inhibition of cell growth and suppression of Akt phosphorylation as well as a parallel decrease in telomerase activity and hTERT mRNA levels (294). At present, there are no reports of different expression levels of other telomerase associated proteins. Interestingly DC, which is associated with an


3|Publishedliteratureontelomerasebiologyinendometrialcancer:telomeraseactivityandtelomere

 or

 content

*(Continued)*


*(Continued)*

TABLE

3


Continued

**101**


TABLE 3 |

Continued bar 10µm.

increase in the risk of developing some cancer types, has not been reported to be linked with an increased incidence in EC. There are no published studies examining the role of dyskerin in EC to date.

# ANTI-TELOMERASE THERAPY

Telomerase was thought to be a suitable target for anticancer agents due to the high activity levels seen in most cancers. Available anti-telomerase strategies can be grouped into three main categories: (1) Telomerase inhibitors, (2) telomerase targeted immunotherapy and (3) telomerase directed viral therapy. Imetelstat (GRN163L) is the only clinically applicable specific oligonucleotide telomerase inhibitor (**Figure 4**), which is a water soluble, acid and nuclease resistant compound that forms stable RNA duplexes (295). It prevents the 13-nucleotide region of TERC to form a complex with hTERT. Unfortunately, clinical data for Imetelstat has been disappointing with high toxicity (296). The other anti-telomerase agents are also undergoing clinical trials yet there are no conclusive data yet available for their clinical effectiveness in cancer. For those cancers harboring activating TERT promotor mutations, directed immunotherapies have been proposed as part of a personalized treatment (297). Anti-telomerase therapy and its relevance to cancer was reviewed in detail in several reviews recently (298, 299).

Progestogens remain to be one of the main hormonebased chemotherapeutic agents that are used in early, advanced and recurrent EC with only modest benefit (24). The loss of response to progesterone or progressive disease despite progestogens has been alluded to progesterone-induced down regulation of progesterone receptor (261) and the lack of progesterone receptor expression is a feature of advanced ECs (261). Since telomerase levels are high in most ECs and since telomerase seem to be a downstream target of progesterone in the endometrium, direct telomerase inhibition may have an added benefit in some women with EC. Those with recurrent disease despite progesterone treatment or having PR negative advanced ECs may particularly respond to telomerase inhibition. However, the available limited in vitro data may suggest that Imetelstat may reduce telomerase activity but may not cause cell death (**Figure 4**) (15). Since the in vitro data has been generated in a mono-cellular 2D culture system comprising of only epithelial cells, thus it may not accurately reflect the in vivo response to the medication (158). Further studies using either physiologically more relevant 3D culture systems containing epithelial and stromal cells or animal models are warranted to explore this avenue further before embarking on clinical studies.

# CONCLUSION

Telomere and telomerase have an intricate relationship with cancer-related multiple cellular functional pathway aberrations. Collectively, the available evidence suggests that endometrial cancer tissues have relatively short telomeres that are maintained by high telomerase activity. Further studies should shed light into different endometrial cancer subtype-associated changes in telomere length, which might facilitate exploring alternative therapeutic strategies to prevent occurrence and progression or recurrence of this devastating disease. Future studies examining the involvement of various telomere and telomerase associated proteins as prognostic markers that potentially could be used in stratifying patients for adjuvant therapies in endometrial cancer are also warranted. In addition, a comprehensive understanding of the telomere and telomerase biology in endometrial cancer will facilitate assessment of targeting telomerase as a personalized therapeutic strategy in endometrial cancer.

# AUTHOR CONTRIBUTIONS

DH conceived the manuscript. RA, MA, LB, and DH prepared the first draft. GS, RA, MA, and DH revised the manuscript critically for important intellectual content and RA, MA, and DH prepared the figures and references. All authors revised and read the manuscript and approved the submitted final version.

# FUNDING

The authors would like to acknowledge the support from Wellbeing of Women project grant RG1487 (DH), Higher Committee for Education Development in Iraq (RA), Liverpool Women's Hospital cancer charity (MA).

# ACKNOWLEDGMENTS

The authors are grateful for department of Women's and Children's Health of University of Liverpool in supporting this work (RA, MA, and DH).

#### REFERENCES


the canonical Wnt signalling pathway. EMBO Rep. (2013) 14:356– 63. doi: 10.1038/embor.2013.16


Cajal body localization and telomere synthesis. Science. (2009) 323:644– 8. doi: 10.1126/science.1165357


risk loci at 5p15, one of which associates with TERT expression. Hum Mol Genet. (2013) 22:2520–8. doi: 10.1158/1538-7445.AM2013-2546


cascade in human ovarian cancer cell lines. Oncogene. (2004) 23:4505– 15. doi: 10.1038/sj.onc.1207582


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Alnafakh, Adishesh, Button, Saretzki and Hapangama. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# PP2A: A Promising Biomarker and Therapeutic Target in Endometrial Cancer

#### Michiel Remmerie and Veerle Janssens\*

*Laboratory of Protein Phosphorylation and Proteomics, Department of Cellular and Molecular Medicine, KU Leuven, Leuven, Belgium*

Over the last decade, the use of targeted therapies has immensely increased in the treatment of cancer. However, treatment for endometrial carcinomas (ECs) has lagged behind, although potential molecular markers have been identified. This is particularly problematic for the type II ECs, since these aggressive tumors are usually not responsive toward the current standard therapies. Therefore, type II ECs are responsible for most EC-related deaths, indicating the need for new treatment options. Interestingly, molecular analyses of type II ECs have uncovered frequent genetic alterations (up to 40%) in *PPP2R1A,* encoding the Aα subunit of the tumor suppressive heterotrimeric protein phosphatase type 2A (PP2A). *PPP2R1A* mutations were also reported in type I ECs and other common gynecologic cancers, albeit at much lower frequencies (0–7%). Nevertheless, PP2A inactivation in the latter cancer types is common via other mechanisms, in particular by increased expression of Cancerous Inhibitor of PP2A (CIP2A) and PP2A Methylesterase-1 (PME-1) proteins. In this review, we discuss the therapeutic potential of direct and indirect PP2A targeting compounds, possibly in combination with other anti-cancer drugs, in EC. Furthermore, we investigate the potential of the PP2A status as a predictive and/or prognostic marker for type I and II ECs.

#### Edited by:

*Massimo Nabissi, University of Camerino, Italy*

#### Reviewed by:

*Alessandro Igor Cavalcanti Leal, Johns Hopkins Medicine, United States Luigi Formisano, University of Naples Federico II, Italy*

#### \*Correspondence:

*Veerle Janssens veerle.janssens@kuleuven.be*

#### Specialty section:

*This article was submitted to Women's Cancer, a section of the journal Frontiers in Oncology*

Received: *07 December 2018* Accepted: *14 May 2019* Published: *04 June 2019*

#### Citation:

*Remmerie M and Janssens V (2019) PP2A: A Promising Biomarker and Therapeutic Target in Endometrial Cancer. Front. Oncol. 9:462. doi: 10.3389/fonc.2019.00462* Keywords: phosphatase targeted therapy, PPP2R1A, endometrial cancer, type II endometrial carcinoma, serous endometrial carcinoma, CIP2A, PME-1, PP2A activating drug

#### INTRODUCTION

Treatment options for cancer have advanced immensely throughout history, with mainly one goal: to specifically target the tumor with as little harm as possible for the patient (1–4). During the last decade, molecular characterization of many tumors has brought cancer research another step closer toward this goal (5), providing the keys to unlock the door toward personalized medicine. However, treatment for endometrial cancer seems to lag behind, although potential markers for this disease have been identified and successful precedents using such markers for targeted therapy have been set in other cancers (e.g., lung cancer, chronic myeloid leukemia, breast cancer, melanoma) (6–13).

In this review, we will discuss the potential of the tumor suppressive protein phosphatase type 2A (PP2A) as a new biomarker and therapeutic target for both type I and type II endometrial carcinomas (ECs). We will mainly focus on the potential predictive and prognostic value of PPP2R1A, encoding the Aα subunit of PP2A, which is mutated in up to 40% of type II ECs,

**112**

while largely being unaffected in type I ECs and other common gynecologic cancers. In the latter cancer types, PP2A dysfunction commonly occurs, however, by other mechanisms of inactivation, stressing the importance of functional PP2A for preventing tumor development and/or progression. Overall, we propose that current, unsatisfactory (type II) EC treatments could be largely improved by taking the PP2A status of the tumor into account since it could be a potential useful indicator for prognosis and therapy response.

#### WHY TARGETED THERAPIES SHOULD BE THE FOCUS IN TYPE II EC

To date, the standard treatment for all ECs is surgery, followed by adjuvant therapy if necessary (14). This treatment protocol is usually sufficient for the clinically indolent and hormoneresponsive type I ECs, which comprise 80% of all ECs. However, the other 20% of ECs are aggressive type II cancers, with serous histology as the most prominent subtype (15, 16). Unfortunately, most of these high-grade tumors are resistant to conventional chemo- and radiation therapies, underscoring the major clinical need for improved treatment regimens for this EC subgroup (17– 23). Additionally, due to their late stage detection and metastatic character, surgery is usually not an option for type II ECs, since the cancer has often already spread outside the uterus at the time of diagnosis (24–26). Therefore, not surprisingly, most ECrelated deaths are due to the type II cancers, with dismal overall survival rates of generally <30% (25, 27–29).

Despite the above knowledge, patients with type II EC are still treated with largely ineffective chemotherapy regimens, thereby leading to unnecessary physical and economic burdens for the patient. In Japan, for example, one cycle of the commonly used carboplatin/paclitaxel protocol consists of three courses with a cost of ∼2,000 Euros per course (30). Additionally, the medical care costs for managing the side effects were 1.6 times as much as the cost for one course, accumulating in a total cost of ∼9,200 Euros per chemotherapy cycle. A study in the United Kingdom further demonstrated that the total costs (diagnosis/surgery, adjuvant therapy, and further treatment) increased with increasing EC grade (31). Hence, more preclinical studies and clinical trials should be focusing on targeted therapies in order to provide more adequate treatment options for patients with high-grade type II EC (32). This view seems particularly justified in light of the fact that successful results have already been obtained in other cancer types and several molecular markers have been identified in EC (33–35). One of the most promising markers in this context is certainly PPP2R1A, encoding the Aα subunit of the tumor suppressive phosphatase PP2A.

#### THE TUMOR SUPPRESSIVE PROTEIN PHOSPHATASE PP2A

Reversible phosphorylation is the key pillar on which signal transduction is built. The enzymes responsible for these post-translational modifications are the protein kinases, which catalyze phosphorylation of proteins, and the protein phosphatases, which, in turn, remove the phosphate group from their substrates (36–38). Importantly, the presence or absence of a phosphate can affect the biological activity of the modified protein, either positively or negatively, depending on the substrate. Like that, kinases and phosphatases act as on/off or off/on switches in cellular signaling. The balanced activities between both enzymes ensures that cellular homeostasis is preserved and that cells can generate the appropriate responses (e.g., proliferation, differentiation, survival, apoptosis. . . ) to specific external stimuli. However, in cancer cells, this balance is genetically disrupted by mutations in key signaling molecules that often directly or indirectly affect kinases and phosphatases, so that signaling pathways will be constitutively activated or inhibited, eventually leading to overall uncontrolled cell growth and survival (39, 40).

Initially, in cancer research, protein kinases got the bulk of the attention, since their over-activation commonly drives oncogenic signaling and their pharmacologic inhibition showed promising clinical potential (6, 38, 41, 42). Furthermore, with more than 500 genes encoding protein kinases, they were thought to be the specific regulators of important oncogenic signaling pathways (43). However, if one accepts that a change in phosphorylation often just reflects an altered balance between kinase and phosphatase activities, kinases and phosphatases seem equally attractive therapeutic targets. Nevertheless, compared to kinases, phosphatase research, and phosphatase-directed therapies have lagged behind for a long time. This has in part been due to the fact that the first phosphatase was discovered 20 years after the first kinase, and that protein phosphatases were for a long time regarded as significantly less specific and less amenable to regulation by external stimuli, rendering them less attractive as therapeutic targets (44, 45). However, more and more attention has been brought to the phosphatases nowadays, hopefully resulting in more clinical applications in the near future (37, 46, 47).

The large majority of protein phosphorylation occurs on serine (Ser) and threonine (Thr) residues (48). The Ser/Thr phosphatase PP2A constitutes about 1% of the total cellular protein content and is together with Protein Phosphatase 1 (PP1) responsible for more than 90% of all Ser/Thr phosphatase activity in the cell (49, 50). PP1 and PP2A are both holoenzymes, consisting of different subunits. PP2A consists of a dimeric core enzyme composed of a catalytic C subunit and a scaffolding A subunit (**Figure 1**). In humans, each of these subunits have two isoforms, α and β, of which the α isoform is the most commonly expressed in most cell types (51, 52). However, in

**Abbreviations:** Bad, Bcl2-associated death promotor; CIP2A, Cancerous inhibitor of PP2A; EC, Endometrial carcinoma; ERK, Extracellular signal-regulated kinase; GSK-3β, Glycogen synthase kinase 3β; HPV, Human papilloma virus; MAPK, Mitogen-activated protein kinase; MEK, Mitogen-activated protein kinase kinase; mTOR, Mammalian target of rapamycin; PI3K, Phosphatidylinositol-4,5 bisphosphate 3-kinase; PME-1, PP2A methylesterase 1; PP2A, Protein phosphatase 2A; pRb, Protein retinoblastoma; PTPA, Phosphatase 2A phosphatase activator; SET, Suvar/Enhancer of zeste/Trithorax; SMAP, Small molecule activator of PP2A; SV40, Simian virus 40; TKI, Tyrosine kinase inhibitor.

order to target specific protein substrates, a third subunit needs to be associated with the AC core dimer, resulting in the formation of the trimeric PP2A holoenzyme (**Figure 1**). This third subunit is referred to as the regulatory B subunit and determines the subcellular localization and substrate specificity (53). The human genome encodes four different families of B subunits, which mutually exclusively bind the AC core dimer: PR55 (B/B55), PR61 (B'/B56), PR72/130 (B"), and B"' (Striatins). Furthermore, each family of B subunits consists of several isoforms (α up to ε) and splice variants, allowing for the formation of many different PP2A holoenzymes (53).

This huge structural diversity of PP2A holoenzymes forms the basis for its diverse functions in cellular signaling by allowing PP2A to act on various components within important signaling pathways (54). The main pathways affected by PP2A are the PI3K (Akt), mTOR (p70S6K) and MAPK (MEK/ERK) pathways (**Figure 1**). Additionally, PP2A also targets the oncoprotein cMyc as well as components involved in Wnt (GSK-3β, β-catenin) signaling, apoptosis (Bcl2, Bad, FOXO), cell cycle regulation (cdc25, WEE1, pRb) and DNA damage response (p53, ATM, Chk) (51, 55–60). All of these pathways are key regulators of processes imbalanced in tumorigenesis (e.g., protein synthesis, cell proliferation, cell survival, cell migration, and invasion). Since PP2A usually negatively affects these pathways, it was denoted as a potential tumor suppressor (61).

The tumor suppressive properties of PP2A were first demonstrated in in vitro experiments using the tumor promoting agent and selective PP2A inhibitor, okadaic acid (OA), as well as the simian virus 40 (SV40) small T antigen. These experiments showed that PP2A inactivation is an absolute requirement in order to achieve oncogene-induced transformation of immortalized human epithelial cells (e.g., by oncogenic H-Ras). OA is able to inhibit PP2A by acting on the catalytic C subunit, while SV40 small T antigen inhibits PP2A by binding the Aα subunit, thereby replacing specific B subunits (61–64). The tumor suppressive nature of PP2A was further corroborated by in vivo evidence in mice. For example, mice completely lacking the B56δ subunit, or showing ≥50% decreased expression of the phosphatase 2A phosphatase activator (PTPA), spontaneously developed tumors (65–67). Additionally, the general physiologic importance of PP2A function was demonstrated in several mouse models (68). For example, PP2A Aα or PP2A Cα knockout mice are embryonically lethal, indicating the importance of PP2A already during development. Furthermore, a vast array of pathological phenotypes were observed in mice with different genetic PP2A dysfunctions, stressing the importance of functional PP2A in many crucial signaling pathways and in tissue homeostasis (68).

In line with these studies, many human cancers have shown to be associated with PP2A dysfunction (69–71). The main mechanism of PP2A inactivation in cancer is via the overexpression of the endogenous PP2A inhibitors SET (Suvar/Enhancer of zeste/Trithorax) and CIP2A (Cancerous Inhibitor of PP2A) (69, 72, 73). However, PP2A can also be inactivated via aberrant post-translational modifications, mostly via PP2A methylesterase (PME-1) upregulation, thereby stabilizing inactive PP2A complexes through binding and/or demethylation of the C subunit C-terminal tail (74–76). Another way of PP2A inactivation is via mutations in one of its subunits (77), or via mutations or heterozygous loss of the cellular PP2A activator PTPA (PPP2R4) (66). Interestingly, PP2A dysfunction is very common in endometrial cancer, as well as in other gynecologic malignancies, such as ovarian and cervical cancer. In the following part, we will give an overview of how PP2A is specifically inactivated in these gynecologic cancers.

#### THE PP2A STATUS IN ENDOMETRIAL AND OTHER COMMON GYNECOLOGIC CANCERS

#### PP2A Status in Endometrial Cancer

Inactivation of PP2A is observed in both type I and type II ECs. However, the way PP2A is inactivated seems to be quite different in both EC subtypes.

In type I endometrioid ECs, PP2A inactivation is likely indirect via an upregulation of the endogenous PP2A inhibitors CIP2A or PME-1. Immunohistochemical analysis of paraffinembedded EC tissue revealed positive CIP2A staining in 79% of the cases. Additionally, increased CIP2A mRNA levels were observed in fresh human EC tissue compared to healthy endometrial tissue (78). CIP2A depletion in endometrioid cancer cells decreased cell proliferation and invasion, while apoptosis was increased, indicating the oncogenic role of CIP2A in type I EC and its potential as a therapeutic target (78). Detailed information on the mechanism of CIP2A overexpression in ECs is still lacking. However, in estrogen receptor (ER)-positive breast cancer cells, estradiol (E2) was able to increase CIP2A protein levels through the ERα (79). Therefore, it can be hypothesized that the same could be true for the estrogen-dependent type I ECs. Additionally, it could be one of the explanations why CIP2A overexpression is rare in the estrogen-independent type II ECs.

In addition, Wandzioch et al. demonstrated PME-1 overexpression in endometroid EC cell lines as well as patient samples (80). PME-1 expression was about 20 times higher in tumor tissue compared to healthy tissue, indicating PME-1 could also be a new potential biomarker for type I ECs. In case of PME-1 upregulation, PP2A activity was significantly reduced, resulting in an increased oncogenic phenotype via upregulation of the PP2A targets Akt and ERK (80).

In contrast to type I ECs, PP2A inactivation in type II ECs is, to our knowledge, not associated with CIP2A or PME-1 overexpression. Instead, up to 40% of type II EC tumors are associated with heterozygous missense mutations in PPP2R1A (34, 35, 81–95). PPP2R1A mutations also occur in type I ECs, albeit at very low frequencies (2.5–6.9%) (32).

PPP2R1A encodes the Aα subunit of PP2A and is an established tumor suppressor gene (96, 97). The structure of Aα is characterized by 15 Huntingtin-Elongation-A subunit-TOR (HEAT) repeats (98) (**Figure 1**). Each of these HEAT repeats consists of a pair of anti-parallel alpha helices connected via intrarepeat loops. These intra-repeat loops are responsible for the interaction with the C and B subunits. More precisely, HEAT repeats 1–10 are able to bind the regulatory B subunits while HEAT repeats 11–15 bind the catalytic C subunit. Remarkably, most of the PPP2R1A mutations cluster together in HEAT repeats 5 and 7, which are involved in B subunit binding. Another intriguing fact is that these PPP2R1A mutations almost always occur at the same residues across several cancer types, forming so called hotspot mutations. These hotspot mutations include p.P179R/L, p.R183G/P/Q/W (HEAT repeat 5), and p.S256F/Y (HEAT repeat 7) (93, 94, 99–101). Remarkably, in type II EC, p.P179L/R hotspot mutations are a lot more abundant than in most other cancer types (**Figure 2**). Biochemical studies in endometrial cancer cells demonstrated that C subunit binding for these hotspot mutations was significantly reduced. Strikingly, loss of C subunit binding was more severe for p.P179R (80% less C binding compared to wild type) than for p.R183W and p.S256F (50–60% less C binding compared to wild type). This

biochemical difference between the hotspot mutations could result in distinct functional consequences, although further research is needed to investigate this. Moreover, PPP2R1A hotspot mutations resulted in deficient binding to specific regulatory B subunits (B55α, B56α,β,ε, B"/PR72), while some B subunits (i.e., B56δ and B56γ, and the striatins) retained binding to the mutated Aα subunit (102, 103). Nevertheless, despite the retained binding of these B subunits, the phosphatase activity of the Aα mutated PP2A-B56δ, γ trimers was reduced. Therefore, it was suggested that PPP2R1A mutations are not simply loss-of-function (i.e., by their inability to bind specific B-type subunits), but may also lead to a dominant-negative inhibition of specific PP2A complexes (i.e., by their ability to retain binding to specific B-type subunits, but capturing them in trimeric complexes with decreased phosphatase activity). This hypothesis was further mechanistically underbuilt by mass spectrometrybased identification of the (mutant) Aα interactomes. These interactomes indeed revealed that the Aα mutants had an increased binding to the endogenous PP2A inhibitor TIPRL1, which could explain the decreased PP2A activities of retained mutant Aα-PP2A trimeric complexes (102). In accordance, ectopic expression of several Aα mutants, among which p.P179R and p.S256F, in the wild-type PPP2R1A-expressing HEC-1A EC cell line, resulted in increased anchorage-independent cell growth and increased xenografted tumor growth in nude mice, and correlated with increased Akt and mTOR/S6K oncogenic signaling (102).

Besides the PPP2R1A mutations, also mutations in PPP2R1B, encoding the PP2A Aβ isoform, have been reported for ECs in the cBioportal database, albeit with occurrences of <1% (mainly deletions) in both EC subtypes (93, 94).

#### PP2A Status in Ovarian and Cervical Cancer

Inactivation of PP2A has also been reported in other gynecologic cancers. In contrast to type II ECs, and similar to type I ECs, ovarian (0–7%), and cervical cancers (0–1%) hardly present any mutations in PPP2R1A (91, 93, 94, 104–112). Interestingly, however, the few number of PPP2R1A mutations that are present mainly occur at the hotspot residue p.R183, which is most frequently affected across all cancer types (**Figure 2A**).

Likewise, mutations in PPP2R1B, were also reported in ovarian and cervical cancers, albeit at very low frequencies, and without any hotspot mutations. Furthermore, the presence of PPP2R1B mutations or loss of heterozygosity was not relevant for ovarian and cervical tumorigenesis (113–116).

Instead, the main way of PP2A inactivation in both ovarian and cervical cancers is indirect. Indeed, a retrospective analysis of serous ovarian cancer demonstrated 40.4% of the specimens to have strong CIP2A immunoreactivity and another 42.4% had weak positive staining (117). Another retrospective study of 152 ovarian cancer specimens (serous, endometrioid, mucinous and clear cell) further corroborated this by showing CIP2A overexpression in 65.79% of samples tested (118). Likewise, recent studies have shown SET overexpression in ovarian cancers (119).

Increased CIP2A expression levels have also been reported in cervical cancer. For example, one study reported on the expression of CIP2A in 60.8% of samples from patients with squamous cervical cancer while this was only 5.7% in normal cervical epithelial tissue. Furthermore, five cervical cancer cell lines harbored elevated CIP2A levels (120). This was further corroborated by a study of Huang et al. in which CIP2A expression was observed in cervical cancer cell lines but not in normal epithelial cells. Additionally, cervical cancer tissue had higher CIP2A mRNA levels compared to healthy adjacent tissue (121). In cervical cancer, CIP2A is mainly upregulated via the E6 and E7 oncoproteins expressed by the human papilloma virus (HPV) type 16 (122–124), the most common type of HPV in cervical cancers (125). In addition to the indirect PP2A inhibition via CIP2A, Pim et al. also proposed a direct way of PP2A inhibition in cervical cancers. They observed that the E7 oncoprotein is able to bind the PP2A Aα and Cα subunits, thereby displacing the B subunit. This way E7 is probably acting in the same way as the SV40 small T antigen rendering PP2A unable to dephosphorylate and inhibit its oncogenic targets (126, 127). In contrast, however, White et al. were not able to demonstrate this interaction between the E7 oncoprotein and PP2A, despite some similarities between E7 and the SV40 small T antigen (128). Finally, another way of decreased PP2A activity during cervical carcinogenesis might be via reduced PP2A Cα expression, potentially by a microRNA-dependent mechanism (129).

Summarized, based on the mechanism of PP2A inactivation, two groups can be distinguished within gynecologic cancers. The first group comprises type I endometrioid EC, ovarian and cervical cancers and is characterized by PP2A inactivation mainly via CIP2A or PME-1, and perhaps also via SET. This group also has a very low frequency of PPP2R1A mutations. However, when present, these mutations recur mostly at hotspot residue p.R183. The second group comprises the type II ECs. This cancer type lacks CIP2A and PME-1 overexpression, but harbors frequent heterozygous missense mutations in PPP2R1A. Moreover, these missense mutations are different from the ones associated with the first group. More precisely, PPP2R1A mutations in type II ECs most frequently recurred at residues p.P179 and p.S256 (**Figure 2B**). Indeed, when looking at the cBioportal database, almost all hotspot mutations in p.P179 and p.S256 were associated with type II ECs, while hotspot mutations in p.R183 are more frequently observed across other cancer types (**Figure 2A**) (90, 93, 94, 130). The reasons for this distinct mutational pattern remain currently unclear. However, distinct PPP2R1A missense mutations might affect B subunit binding and PP2A activity in a slightly different way (102), thereby, in part, contributing to a different tumor biology in type I and II ECs as well as other cancers. However, further research is warranted in order to fully understand the involvement of specific PP2A holoenzymes in different cancers. Nevertheless, the distinct PP2A inactivating mechanisms between type I and type II ECs, as well as other gynecologic cancers, open up specific opportunities for direct or indirect, personalized therapeutic targeting of PP2A, in order to (re)-activate this phosphatase.

#### PP2A AS A POTENTIAL THERAPEUTIC TARGET IN ENDOMETRIAL CANCER

In the last few years, the notion that phosphatases as opposed to kinases could also be useful therapeutic targets has gained more and more attention (54, 131–134). In contrast to therapeutic inhibition of kinases, the focus with targeting phosphatases, and in particular the tumor suppressive phosphatase PP2A, is on the development of activating or reactivating compounds (133). These compounds are either able to relieve the inhibition by endogenous inhibitors, or directly bind and activate PP2A. In contrast, some studies also reported on the benefit of PP2A inhibition rather than activation. Despite the promising therapeutic potential of these compounds in multiple pre-clinical studies in several other human cancer types, pre-clinical studies, specifically in EC, are however, mostly lacking. In the following four subsections, we will provide an overview of the most promising PP2A-targeted therapies, which—pending additional dedicated studies—may become applicable to EC as well.

#### Compounds Indirectly Activating PP2A CIP2A Targeting Compounds

Since CIP2A is commonly overexpressed in several cancers, it has become an interesting therapeutic target in order to re-activate PP2A. Specifically in type I EC, CIP2A depletion decreased proliferation and invasion, and increased apoptosis in vitro (78), indicating it could be a valuable therapeutic target. Furthermore, depletion of CIP2A also showed anti-tumorigenic potential in ovarian and cervical cancer cells (118, 120, 135).

Most of the currently described CIP2A targeting compounds are able to increase PP2A activity by reducing the CIP2A protein levels, either via downregulation of CIP2A expression or by promoting its degradation. In cervical and endometrial cancer cells, CIP2A expression is mainly regulated by two transcription factors, Elk1 and Ets1, which are both necessary for regulating CIP2A protein levels (136). Additionally, in cervical cancer, the transcription factor E2F1 has also been implicated in the regulation of CIP2A expression via the E7 oncoprotein (124, 137). Therefore, compounds targeting one of these factors could have potential therapeutic value. For example, several erlotinib derivatives were able to reduce CIP2A levels and increase PP2A activity in breast cancer and hepatocellular carcinoma cells via disrupting the interaction between the transcription factor Elk1 and the CIP2A promotor (138, 139). On the other hand, lapatinib downregulated CIP2A through regulation of protein stability in breast cancer cells (140). Also, bortezomib, a US Food and Drug Administration (FDA)-approved proteasome inhibitor, was able to reduce CIP2A expression levels in several cancer cell lines, although the mechanism of action is not elucidated yet (141–143). Furthermore, increased CIP2A degradation through autophagy was seen in breast cancer cells upon mTORC1 inhibition (e.g., using rapamycin) (144). Additionally, several natural compounds have demonstrated PP2A re-activating potential via downregulation of CIP2A. Despite the lack of studies testing these compounds in EC, positive results have already been obtained in other cancer studies. For example, rhabdocoetsin B, arctigenin and the red wine component ellagic acid were able to reduce CIP2A transcription levels in breast and lung cancer cells (145–147). On the other hand, celastrol and gambogenic acid promoted CIP2A degradation in lung and liver cancer cells (148, 149). Additionally, the compounds genistein and ethoxysanguinarine promoted both transcriptional suppression and proteasomal degradation of CIP2A (150, 151). Lastly, fusogenic-oligoarginine peptide-mediated delivery of siRNA targeting CIP2A has also appeared as a new therapeutic strategy, showing anti-tumorigenic potential in vitro and in vivo in oral cancer cells (152, 153). It would be extremely interesting to test whether any of these known CIP2A inhibiting compounds would have therapeutic benefits in CIP2A-overexpressing endometrioid EC models.

#### PME-1 Targeting Compounds

PME-1 has also emerged as a potential therapeutic target in endometrioid EC, especially since PME-1 depletion using RNA interference resulted in increased PP2A activity, thereby reducing the oncogenic phenotype of type I EC cells in vitro and in xenograft assays (80). Additionally, PME-1 depletion in HeLa cells, a cervical cancer cell line, also led to decreased proliferation and colony formation by increasing PP2A activity and thereby inhibiting MAPK pathway activity (154).

So far, two classes of pharmacologic PME-1 inhibitors have been discovered, the ABL (Aza-β-lactam) inhibitors and the sulfonyl acrylonitrile inhibitors, which both irreversibly bind to PME-1 and inhibit PME-1 esterase activity (155, 156). Pusey et al. tested two of these PME-1 inhibitors, ABL-127 and AMZ-30, of which ABL-127 was the most potent one in EC models. However, in vivo testing of this compound in xenograft assays could not corroborate the in vitro data, implying that inhibition of solely the PME-1 esterase activity may be insufficient to inhibit PME-1's oncogenic characteristics (157).

#### SET Targeting Compounds

Although the relevance of SET overexpression (if any) in EC is currently unclear, SET inhibitors can mainly be divided into three groups according to their origin. The first group comprises sphingolipid-based compounds, such as ceramide and FTY720 (also called Fingolimod), as well as their derivatives. The second group resembles the apolipoprotein E (ApoE) and these compounds are denoted as SET interfering peptides (e.g., COG112 and OP449). More recently, potent cytotoxic effects were reported for cell penetrating peptides, the third group of SET inhibitors, which constitute the precise SET-PP2A interaction interface (158). Although the mechanism of action of all these SET inhibitors is not always well understood, they most likely increase PP2A activity in the same way, i.e., via disruption of the interaction between SET and PP2A (53, 59, 133). Recently, the interaction between the sphingolipidbased compounds (i.e., ceramide and FTY720) and PP2A were investigated in more detail using NMR spectroscopy (159). In this study, they observed that the sphingolipid compounds probably work by disrupting the dimerization of SET, which is thought to be important for its PP2A binding and inhibiting activity.

#### Compounds Directly Activating PP2A

Recently, small molecule activators of PP2A (SMAPs) have been developed (160). These SMAPs are derived from the anti-psychotic phenothiazines and are predicted to directly activate PP2A. Although a direct binding to the PP2A Aα subunit has been demonstrated (161), the mechanism by which these SMAPs are able to activate PP2A remains unknown. Nevertheless, several pre-clinical studies have shown the promising anti-proliferative potential of SMAP treatment, for example in T-cell acute lymphoblastic leukemia, castration-resistant prostate cancer, KRAS mutant lung adenocarcinoma, and tyrosine kinase inhibitor (TKI) resistant lung adenocarcinoma (161–164). Hence, it would be of amazing interest to test these promising compounds in pre-clinical EC models with intact as well as impaired PP2A functionality.

### The Therapeutic Potential of Combination Therapies

Besides the therapeutic potential of single agent targeting of PP2A, also combination therapies of PP2A activators with other drugs have gained attention (47, 165). The combination of a PP2A activator with a kinase inhibitor seemed particularly beneficial in cases where oncogenic kinase activation simultaneously resulted in PP2A inhibition, and therapy resistance to a single agent kinase inhibitor occurred (166–168). For example, KRAS-mutant lung cancer and pancreatic ductal adenocarcinoma cells showed resistance to MEK inhibitors and mTOR inhibitors, respectively. This resistance occurred due to cross-talk with the PI3K/Akt/mTOR pathway and cMyc oncoprotein upregulation, probably via PP2A inhibition. Hence, they tested the combination of a direct PP2A activator (SMAP) with a MEK or mTOR inhibitor, which resulted in significantly increased anti-cancer effects in vitro as well as in vivo (169, 170). Such combinatorial benefit was further demonstrated in myeloid leukemia where the combination of a SET inhibitor (indirect PP2A activation) with a tyrosine kinase inhibitor resulted in synergistic anti-cancer effects (171, 172). In TKI-resistant lung adenocarcinoma cells, the synergistic effects of a SMAP and the TKI afatinib were in part also contributed to a downregulation of CIP2A (164).

The combination of PP2A activators with chemotherapy has also been investigated, although not yet in EC. Several studies tested the effect of combining SET inhibitors (FTY720, OP449) with different chemotherapy regimens (e.g., doxorubicin, cisplatin). These studies demonstrated synergistic anti-cancer effects in myeloid leukemia, breast cancer cells, colorectal cancer cells as well as in cisplatin-resistant melanoma and lung cancer cells (171, 173–175).

Overall, these successful precedents open up possibilities to test these PP2A activating compounds in EC models, possibly in combination with kinase inhibitors or chemotherapeutics. This could be specifically interesting for the type II serous ECs, in which therapeutic combinations with PP2A activators might sensitize these cancer cells toward the current, mainly ineffective, therapies.

# Exploiting PP2A Inhibition for Therapeutic Purposes

In contrast to the therapeutic potential of PP2A activation, some studies also reported on the therapeutic relevance of PP2A inhibition, when applied together with a DNA damaging treatment, or when combined with immunotherapy (176, 177). The anti-cancer effect of PP2A inhibition in combination with DNA damaging agents can be explained by the enabling role of PP2A in DNA damage response and repair pathways as well as in cell cycle regulation. Hence, PP2A inactivation in this situation (i.e., combined with chemo-or radiation therapy) leads to aberrant cell cycle progression and checkpoint activation, resulting in mitotic catastrophe and, consequently, cell death (133, 178, 179). Likewise, PP2A is also involved in the immune response by negatively regulating the function of cytotoxic Tlymphocytes (176, 180). Therefore, PP2A inhibition combined with immunotherapy could enhance the immune-mediated antitumor response.

The small molecule LB-100 is one of the best studied PP2A inhibitors so far, without any apparent toxicities in animals and with promising results in a first human clinical trial (181, 182). Pre-clinical studies demonstrated LB-100 was able to sensitize many different solid tumor cells to DNA damaging agents. For example, LB-100 enhanced cisplatin-mediated cytotoxicity in ovarian carcinoma cells in vitro and in vivo in xenografts (178, 183, 184). Likewise, the combination of LB-100 with the immune checkpoint inhibitor aPD-1 in colon and melanoma cancer cells resulted in an enhanced and durable T-celldependent anti-tumor response, with more effector T-cell and less suppressive regulatory T-cell infiltration (176). On a critical note, it needs to be mentioned here though, that recent evidence has suggested that LB-100 is not entirely specific for PP2A, and also inhibits the catalytic activity of the related Ser/Thr phosphatase PP5 (182). As PP5 is considered as tumor promoting, PP5 inhibition could contribute to the anti-tumor activities of LB-100. This was further corroborated in vitro in ovarian cancer cells, where knockdown of PP5 resulted in decreased cell proliferation and colony formation.

To conclude, further research is warranted to fully understand how both PP2A activation and inhibition can be therapeutically viable as anti-cancer treatment for EC.

# PP2A DYSFUNCTIONS AS PREDICTIVE BIOMARKERS FOR TARGETED THERAPIES IN EC

#### CIP2A-Mediated PP2A Inhibition

So far, a wealth of studies reported on the role of CIP2A overexpression, and thereby likely PP2A inactivation, as a potential predictive biomarker for diverse therapies (targeted and untargeted), in a large variety of solid cancers (54). For example, in lung and breast cancer, the overexpression of CIP2A resulted in resistance to the EGFR inhibitors lapatinib and erlotinib, while RNA interference-mediated CIP2A depletion sensitized the cells toward these compounds (140, 185). CIP2A overexpression also resulted in resistance toward Chk1 kinase

inhibitors in gastric adenocarcinoma and breast cancer cells (186). Furthermore, CIP2A overexpression conferred resistance to chemotherapy in several solid tumor types, including cervical and ovarian cancers (187–190). Liu et al. further demonstrated that in HeLa cells, treated with several chemotherapeutics such as paclitaxel, doxorubicin and cisplatin, CIP2A expression was significantly associated with drug insensitivity through increased expression of p-glycoprotein drug efflux pumps (190). Additionally, the use of siRNA, targeting CIP2A in vitro, resulted in sensitization of HeLa cells to different chemotherapeutics (190). Finally, the natural CIP2A inhibitors, ethoxysanguinarine and gambogenic acid, sensitized lung and hepatocellular cancer cells to chemotherapy (149, 151).

While there is a general lack of studies on the predictive potential of CIP2A expression in EC models, it is very likely that, based upon the evidence obtained in other solid (gynecologic) tumors, CIP2A expression could mediate therapeutic resistance in EC cells as well. This further implies that CIP2A status of the EC tumors should better be taken into account in clinical trial set-ups. Additionally, the data illustrate the potential advantages of combining PP2A activators with EC therapies that are mainly ineffective on their own.

#### Recurrent PPP2R1A Hotspot Mutations

Whether PPP2R1A mutations are present in EC tumors or not, could have consequences for the efficacy of targeted therapies. For example, kinase inhibitors targeting the PI3K/Akt/mTOR pathway or MEK/MAPK pathway, commonly affected in ECs, could be less effective when PP2A, counteracting the targeted kinase, is mutated. The rationale behind this is that a kinase inhibitor can only work to its full potential when the opposing phosphatase is not inactivated. In case the phosphatase is dysfunctional, the net phosphorylation would be largely unaffected and the pathway would remain activated. Simply put, the use of certain kinase inhibitors in case of PP2A dysfunction would be the equivalent of pouring water into a bucket with holes in it. This biochemical logic was nicely underscored by Kauko et al. who showed that PP2A inhibition achieved by siRNAmediated knockdown of Aα, conferred resistance to a MEK inhibitor in KRAS-mutant lung cancer cells (169).

On the other hand, PPP2R1A mutations in type II ECs could also be predictors of positive outcome to certain kinase inhibitors. For example, Haesen et al. showed hyperactivation of the PI3K/Akt/mTOR pathway in PPP2R1A mutated EC cells, while the MAPK pathway was actually downregulated (102). This indicates single agent kinase inhibitors targeting the PI3K pathway might be effective in PPP2R1A mutated ECs, since cross-talk to the MAPK pathway would possibly be absent and PPP2R1A mutant EC cells might be dependent on PI3K/Akt/mTOR signaling for growth and survival (32). Patient stratification based on PPP2R1A status of the tumor could also be applied to other gynecologic cancer types, even when PPP2R1A mutations are rare. For example, Papp et al. demonstrated ovarian cancer cell lines harboring PPP2R1A mutations to be more sensitive to a PI3K/mTOR inhibitor (108). However, in this experiment they took cell lines with mutations in PPP2R1A and PARP1 into account, which could bias the results.

Finally, the response of type II ECs to SMAPs could also be dependent on the PPP2R1A mutational status of the cancer cells, especially since these SMAPs bind to the Aα subunit in close proximity to the PPP2R1A hotspot mutations (161). Therefore, mutations in this subunit could disturb the interaction with the SMAPs and consequently render the compound ineffective. Targeted pre-clinical studies addressing these possibilities should provide further insights in these issues in the near future.

#### Others

Although no studies have yet addressed the predictive role of PME-1 overexpression in type I EC, overexpression of PME-1 in glioma drives resistance to various multikinase inhibitors. Consequently, PME-1 depletion resulted in enhanced sensitivity to these inhibitors in vitro and in vivo in xenografts (191).

As identified through a large siRNA screen, decreased expression of the PP2A activator PTPA conferred significantly increased resistance of cervical HeLa cells to several cytotoxic agents, including cisplatin, taxol, and etoposide (192), perhaps suggestive for a similar dismal predictive role for heterozygous loss or mutation of PTPA in type II endometrial carcinosarcoma (66).

Likewise, increased SET expression has been associated with resistance to TKI's, cisplatin, paclitaxel, oxaliplatin, and 5-fluorouracil in diverse cancer types (54, 193, 194). Whether this would be relevant for EC, remains, again, to be determined.

#### PP2A DYSFUNCTIONS AS PROGNOSTIC BIOMARKERS FOR EC

#### CIP2A Overexpression

In several cancers, PP2A inactivation is associated with significantly worse prognosis (54). For example, overexpression of the PP2A inhibitor CIP2A is correlated with worse prognosis in several solid tumors as well as in myeloma (195–197). Specifically, for gynecologic cancers, several studies reported on the prognostic potential of CIP2A in cervical and ovarian cancer. In serous ovarian cancers, strong CIP2A immunoreactivity correlated with worse prognosis (117), and the same was observed in a retrospective study on 152 ovarian cancer specimens, including serous, endometrioid, mucinous, and clear cell subtypes (118). In cervical cancer, CIP2A was found to associate with H-Ras to promote epithelial-mesenchymal transition, resulting in increased migration and invasion of cervical cancer cells in vitro and in vivo (120). Furthermore, cervical cancer tissue analysis revealed that CIP2A expression correlated with lymph node metastasis and high-grade and advanced stage cervical cancer (120). However, these results are in contrast with the human protein atlas database (www. proteinatlas.org) which does not put CIP2A forward as an unfavorable prognostic marker in both ovarian and cervical cancers (198).

Data concerning the prognostic potential of CIP2A in EC are scarce. One study of Yu et al. demonstrated that CIP2A expression in type I endometrioid EC correlated with increased FIGO stage and tumor grade (78). Furthermore, according to the human protein atlas, CIP2A expression correlated with worse prognosis in patients with EC. Nevertheless, more studies are necessary to further demonstrate the prognostic potential of CIP2A in EC.

#### Recurrent PPP2R1A Hotspot Mutations

In the absence of any published work on the prognostic marker potential of PPP2R1A in type II EC so far, we used the cBioportal database to analyze the survival data of type II ECs. However, survival data were only available for 44 patients (UCEC-TCGA study) with type II serous EC, of which only 12 presented with mutations in PPP2R1A (93, 94). Analysis of this limited data set revealed no significant difference in overall survival between patients with and without PPP2R1A mutations (P = 0.39) (**Figure 3**). However, more patient data are definitely required in order to obtain more conclusive results. Longer patient follow-up and centralized data collection in multi-institutional centers could boost the data collection for patients with this rare endometrial cancer subtype.

#### Others

Similarly, PP2A inactivation via PPP2R4 (PTPA) haploinsufficiency leads to a worse prognosis in many cancer types, including endometrial carcinosarcomas (66). Also, lower PP2A/C expression in cervical cancer was closely associated with the nodal status of cervical cancer patients (129).

Although there is a lack of studies on the role of SET in EC, the protein atlas database (www.proteinatlas.org) indicates that there is no correlation between SET expression and a worse prognosis in gynecologic cancers (198). On the other hand, a strong correlation was seen between SET expression levels and decreased survival of ovarian cancer patients (119).

#### FUTURE STEPS

In this review, we discussed the therapeutic potential of PP2A targeting as well as the biomarker potential of PP2A dysfunctions in EC, and other gynecologic cancers. Specifically, in type II

EC, PPP2R1A mutations are remarkably common, while in type I EC, PP2A dysfunction rather occurs through non-genomic mechanisms, involving increased expression of PP2A inhibitors CIP2A and PME-1. However, in order for PPP2R1A to become a clinically relevant biomarker for type II ECs, reliable and fast detection of somatic mutations in this gene will be necessary. Therefore, future development of methods able to detect somatic mutations in tumor samples, or preferably in liquid biopsies, will be crucial.

Over the last few years, several promising methods have been developed for the detection of mutations in oncogenes. For example, Spaans et al. designed a Matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF) based somatic mutation panel able to detect hotspot mutations in 13 genes frequently mutated in gynecologic cancers (199). This method is able to provide reproducible, highthroughput data based on low quality and quantity DNA from formalin-fixed paraffin-embedded (FFPE) samples. Additionally, next-generation sequencing (NGS) has come forward as a potential technique to detect mutations in the clinic. NGS is able to rapidly detect all mutations in the complete gene of interest. However, data analysis is still rather complex and differentiating between somatic passenger and driver mutations can be time consuming. However, optimizations of this technique could lead to a more user-friendly method for the detection of specific mutations. For example, Cottrell et al. validated a NGS assay (WuCaMP) which targets a specific panel of genes with known clinical importance (200). This highly specific and sensitive assay allowed for a fast analysis of the target genes, thereby reducing time and costs of NGS.

More recently, also non-or minimally-invasive liquid biopsies have been investigated as a way to detect somatic mutations in patients with EC (201). For example, circulating tumor cells (CTCs) and circulating tumor DNA (ctDNA) extracted from blood samples or uterine aspirates showed potential as a way to screen for somatic mutations. However, whether CTCs can be useful for patients with high-risk EC is still unclear due to its debatable prevalence in blood samples. Bogani et al. reported a low prevalence of CTCs in pre-operative blood samples of patients with high-risk EC and even an absence of CTCs in patients with type II EC (202). In contrast, Alonso-Alconada et al. demonstrated the presence of CTCs in patients with high-risk EC (203). Thus, further research is warranted in order to prove the potential usefulness of CTCs in type II ECs. On the other hand, also cell-free DNA can be used to detect mutations. NGS was able to detect specific endometrioid EC-associated mutations in the cell-free DNA derived from peripheral blood samples of patients with early and late stage endometrioid EC (204). These promising results indicate the potential of this technique for the detection of type II ECs, when implementing PPP2R1A in the targeted gene panel.

Furthermore, also DNA obtained during a routine Pap (Papanicolaou) test can be analyzed for the detection of oncogenic mutations. For example, Wang et al. designed a test called PapSEEK, which is able to detect mutations in 18 commonly mutated genes in endometrial cancer (205). This is particularly interesting since with this method, oncogenic

mutations could be detected already during a routine checkup, making early stage detection of (type II) EC possible. This is especially important since patients with type II EC show few symptoms until the disease is already in late stages and therefore less amenable to the current therapies.

In contrast, detection of CIP2A or PME-1 overexpression in (type I) EC cannot rely on genetic methods, but should rather focus on the mRNA or protein level, and is therefore, much more dependent on the availability of tumor biopsies or resected tumor material. In this respect, the use of reliable, validated antibodies, able to specifically detect these oncoproteins via immunohistochemistry techniques is crucial. However, the fact that CIP2A is always found expressed at high levels in the tumor tissue, while being nearly undetectable in the corresponding non-proliferating normal tissue, virtually eliminates the issues associated with the poorer suitability of immunohistochemistry as a technique to reliably quantify protein expression in tissues. In addition, the use of autoantibodies as serum biomarkers for CIP2A showed promising results in breast cancer patients (206). Therefore, it might also be interesting to investigate this in the setting of EC.

Despite the identification of the phosphatase PP2A as a promising molecular marker for ECs, few pre-clinical studies have investigated its potential as a direct therapeutic target, nor as a stratification marker for targeted kinase inhibitor treatments in this cancer type. Nevertheless, a plethora of studies in other solid tumor types suggest PP2A to have potential as a new therapeutic target for both type I and, more importantly, for the more aggressive type II ECs. We put several potential therapeutic compounds forward that could be tested in EC studies, potentially in combination with chemotherapy or targeted therapy.

So far, hormonal intervention and the immunotherapeutic pembrolizumab are the only two FDA-approved targeted therapies for hormone-dependent type I ECs, while there are none for the type II ECs (207). Nevertheless, molecular analyses of ECs have revealed that, in the large majority of ECs, the PI3K pathway is overactivated, which led to a number of (pre-)clinical studies investigating kinase inhibitors targeting this pathway (e.g., several mTOR and PI3K inhibitors) (32). The outcome of these studies was largely disappointing, not only due to the development of inherent resistance mechanisms (e.g., cross-talk to MEK/MAPK pathway), but also in a big part due to the complete lack of patient stratification in clinical studies (208, 209). The importance of the latter was further illustrated by the clear therapeutic benefit of the HER2 inhibitors trastuzumab (anti-HER2 antibody) and lapatinib (tyrosine kinase inhibitor, TKI), in type II serous EC cells stratified based on HER2 amplification vs. normal HER2, or based on HER2 amplification and functional PI3K vs. those with HER2 amplification and mutant PIK3CA (32, 210, 211). Likewise, we hypothesized here that the PP2A status of the endometrial tumor, should be an important additional stratification marker for testing these targeted kinase inhibitors, given that PP2A mainly acts as a negative regulator of PI3K and HER2 downstream signaling, and hence its functional or dysfunctional state could co-determine kinase inhibitor therapy outcome. In case dysfunctional PP2A would mediate therapy resistance, the use of SMAP combination therapy could be a valuable solution. Also type I ECs could benefit from the combination of PP2A activators with standard therapies, since sensitization of the cells to chemo- or radiation therapy could result in a lower dose and duration of the therapy, required to treat type I ECs. This in turn, would reduce the physical and economic burden associated with chemotherapy.

Furthermore, since the mechanism of PP2A inactivation is different between both subtypes of EC, it will be important to stratify patients in those having type I and those having type II tumors within (pre)-clinical trials. This way, existing, or new therapeutic compounds will be tested on a more rational basis and no bias will occur toward the biggest group of indolent type I ECs. Therefore, the presence of certain PP2A dysfunctions in the EC tumor could indicate whether the patient is eligible for certain (targeted) therapies.

In conclusion, we highlighted the therapeutic potential of PP2A activating as well as inactivating compounds in several gynecologic cancers. However, it has to be noted that more studies should focus on these promising compounds in the specific context of type I and type II ECs. Furthermore, we demonstrated, based on studies in several other cancers, among which ovarian and cervical cancer, that PP2A dysfunction, due to mutations or cellular PP2A inhibitors, could be an indicator for worse prognosis as well as a predictor for therapeutic outcome in EC. Therefore, stratification of patients with type II EC based on their PPP2R1A mutational status, or of patients with type I EC based on their CIP2A or PME-1 status could help to establish more reliable testing of current and future targeted therapies in clinical trials. Furthermore, the presence of CIP2A or PME-1 expression could also broaden the therapeutic possibilities for the type I ECs. Dedicated pre-clinical studies in EC cells, with functional vs. dysfunctional PP2A status, should address these issues in the near future. Additionally, the presence of PPP2R1A mutations could also help to diagnose patients with type II EC in earlier stages (e.g., via liquid biopsies), thereby also contributing to a better patient outcome. However, further research is warranted in order to confirm this marker potential of PPP2R1A in type II ECs. In the end, only such dedicated studies, will help treatments for patients with type I and type II EC to catch-up with the emerging personalized medicine and targeted therapies already established in many other cancers.

#### AUTHOR CONTRIBUTIONS

MR: writing. VJ: review and Editing.

# FUNDING

MR is a recipient of an F.W.O.-SB fellowship of the Research Foundation-Flanders. VJ is supported by grants from the Belgian Foundation against Cancer (FAC-F/2016/822), the Research Foundation-Flanders (F.W.O. G.0B01.16N and F.W.O. G.0B17.19N) and the KU Leuven Research Fund (C24/17/073).

# REFERENCES


in uterine papillary serous carcinoma. Gynecol Oncol. (2009) 114:293–8. doi: 10.1016/j.ygyno.2009.04.036


molecular characterization of cervical cancer. Nature. (2017) 543:378–84. doi: 10.1038/nature21386


of castration-resistant prostate cancer. Cancer Res. (2018) 78:2065–80. doi: 10.1158/0008-5472.CAN-17-0123


endometrial cancer: a NCI clinical trials planning meeting report. Gynecol Oncol. (2018). In press. doi: 10.1016/j.ygyno.2018.02.005


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Remmerie and Janssens. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

*Integrated Hospital*

# Hepatoma-Derived Growth Factor and DDX5 Promote Carcinogenesis and Progression of Endometrial Cancer by Activating β-Catenin

Chunhua Liu1†, Lijing Wang1†, Qingping Jiang2†, Junyi Zhang3†, Litong Zhu<sup>1</sup> , Li Lin<sup>1</sup> , Huiping Jiang<sup>1</sup> , Dan Lin<sup>1</sup> , Yanyi Xiao<sup>1</sup> , Weiyi Fang1,3 and Suiqun Guo<sup>1</sup> \*

*<sup>2</sup> Department of Pathology, Third Affiliated Hospital of Guangzhou Medical University, Guangzhou, China, <sup>3</sup>*

*of Traditional Chinese Medicine, Southern Medical University, Guangzhou, China*

*<sup>1</sup> Department of Obstetrics and Gynecology, The Third Affiliated Hospital, Southern Medical University, Guangzhou, China,*

#### Edited by:

*Massimo Nabissi, University of Camerino, Italy*

#### Reviewed by:

*Saori Furuta, University of Toledo, United States Farhad Jadidi-Niaragh, Tabriz University of Medical Sciences, Iran Guang-Rong Yan, Third Affiliated Hospital of Guangzhou Medical University, China*

> \*Correspondence: *Suiqun Guo guosq2005@126.com*

*†These authors have contributed equally to this work*

#### Specialty section:

*This article was submitted to Women's Cancer, a section of the journal Frontiers in Oncology*

Received: *14 November 2018* Accepted: *11 March 2019* Published: *11 April 2019*

#### Citation:

*Liu C, Wang L, Jiang Q, Zhang J, Zhu L, Lin L, Jiang H, Lin D, Xiao Y, Fang W and Guo S (2019) Hepatoma-Derived Growth Factor and DDX5 Promote Carcinogenesis and Progression of Endometrial Cancer by Activating* β*-Catenin. Front. Oncol. 9:211. doi: 10.3389/fonc.2019.00211* Background: Our previous work determined the correlation between high nuclear expression of hepatoma-derived growth factor (HDGF) and clinicopathological data of endometrial cancer (EC); however, the modulatory mechanisms and biological role of HDGF in EC have not been reported.

Methods: Lentiviral particles carrying human HDGF short hairpin RNA (shHDGF-1, -2, and -3) vector and plasmids for HDGF, DDX5, and β-catenin expression were, respectively introduced into EC cells to evaluate the effects and molecular mechanisms underlying EC cell proliferation, migration, invasion, and metastasis. Quantitative real time reverse transcription polymerase chain reaction (qRT-PCR) and western blotting were used to determine HDGF and DDX5 expression. Co-immunoprecipitation (co-IP), mass spectrometry, and an immunofluorescence co-localization study were conducted to explore the relationship between HDGF, DDX5, and β-catenin. Immunohistochemistry was used to analyze the clinical associations between HDGF and DDX5 in EC.

Results: Knocking down HDGF expression significantly decreased EC cellular proliferation, migration, invasion *in vitro*, as well as tumorigenesis and metastasis *in vivo*. Conversely, HDGF overexpression reversed these effects. Stable knockdown-based HDGF suppression activated the PI3K/AKT signaling pathway, along with downstream β-catenin-mediated cell cycle and epithelial-mesenchymal transition signaling. Furthermore, co-IP combined with mass spectrometry and an immunofluorescence co-localization study indicated that HDGF interacts with DDX5, whereas β-catenin was associated with DDX5 but not HDGF. Overexpression of DDX5 reversed the suppression of shHDGF. Immunohistochemistry analysis showed that high expression of DDX5 constituted an unfavorable factor with respect to the clinicopathological characteristics of EC tissues and that HDGF and DDX5 high expression (HDGF+/DDX5+) led to a worse prognosis for patients with EC (*P* < 0.001). In addition, we found that the

**128**

expression of HDGF and DDX5 was positively correlated in EC tissues (*r* = 0.475, *P* < 0.001).

Conclusion: Our results provide novel evidence that HDGF interacts with DDX5 and promotes the progression of EC through the induction of β-catenin.

Keywords: EC, HDGF, DDX5, β-catenin, PI3K/AKT

#### INTRODUCTION

Endometrial cancer (EC) comprises the most common malignancy involving the female genital tract and the fourth most common malignancy in women after breast, lung, and colorectal cancers (1). In 2012, approximately 320,000 new cases of EC were diagnosed worldwide and the incidence is increasing (2). Currently, endometrial carcinogenesis is thought to be a multi-step process involving the coordinated interaction of hormonal regulation, gene mutation, adhesion molecules, and apoptosis; however, the molecular mechanisms underlying the pathogenesis of EC have not been fully elucidated (3). Therefore, a better understanding of the molecular mechanism underlying the progression of EC will likely lead to new insights regarding novel therapeutic targets.

Hepatoma-derived growth factor (HDGF), the gene for which is located on chromosome 1q21–23, is a heparin-binding growth factor that was originally purified from media conditioned with the human hepatoma cell line, HuH7 (4). HDGF is ubiquitously expressed in normal tissues and tumor cell lines that exhibit growth factor properties. The most recent study reported that HDGF acts as a coactivator of SREBP1-mediated transcription of lipogenic genes (5). HDGF is characterized as a mitogen for many cell types and localizes to the nucleus, which is necessary for its mitogenic activity. Characteristics such as promoting growth, suppressing differentiation, and exhibiting angiogenic properties, suggest a role for HDGF in cancer induction and tumor progression (6–8). Accordingly, a number of studies have focused on the significance of HDGF as a prognostic marker and have demonstrated its clinical value for oral cancer (9), esophageal cancer (10), gastrointestinal stromal tumors (11, 12), meningiomas (13), hepatocellular cancer (14), and non-small cell lung carcinoma (15). Consistent with these findings, in our previous study (16), we determined a correlation between high nuclear expression of HDGF and clinicopathological data of EC; however, the functional significance of HDGF in EC remains unknown.

The DEAD-box RNA helicase, DDX5 (p68), constitutes a multi-functional protein with an important role in regulating transcription in conjunction with multiple, sequencespecific transcription factors (17). Recently, DDX5 has been demonstrated to act as a potent transcriptional co-activator of the estrogen receptor (18, 19), androgen receptor (20), tumor suppressor p53 (21), MyoD (22), and β-catenin (23). DDX5 has been implicated in cancer development and progression by functioning in several key cancer cell activities, such as proliferation, migration, cytoskeletal reorganization, and the epithelial-mesenchymal transition (EMT) (24–28). Recently, a small molecule inhibitor for DDX5, SupinoxinTM (RX-5902), has been developed for cancer therapy, and is currently in a clinical trial in patients with metastatic triple negative breast cancer (ClinicalTrials.gov identifier: NCT02003092) (29), which suggested the significance of DDX5 in the pathogenesis of tumor.

In the current study, we examined and characterized the interaction between HDGF and DDX5, and determined its effect on the activation of PI3K/AKT and downstream βcatenin-mediated cell cycle and EMT signaling to promote the proliferation, invasion, and metastasis of EC.

#### MATERIALS AND METHODS

#### Cell Culture

The EC cell lines Ishikawa and RL95-2 were purchased from the Chinese Academy of Sciences Cell Bank (Shanghai, China). All cell lines were maintained in RPMI-1640 medium supplemented with 10% fetal bovine serum (ExCell, Shanghai, China). Ishikawa and RL95-2 cell lines used in this study were incubated in a humidified chamber with 5% CO<sup>2</sup> at 37◦C.

#### Immunohistochemistry and Evaluation of Staining

One hundred and twenty two endometrial carcinoma (EC) paraffin sections (3 mm) samples from 2002 to 2008 were obtained in the Third Affiliated Hospital of Guangzhou Medical University, Guangzhou City, China. Detailed information and the IHC of HDGF about the 122 EC tissue specimens was performed in our previous study (16). Immunohistochemistry was performed according to standard procedures. The staining intensity of DDX5 (1:100, Abcam, Cambridge, MA, USA) was scored as previously described (20).

#### Establishment of EC Cell Lines With Stable Knockdown of HDGF

Lentiviral particles carrying human HDGF short hairpin RNA (shHDGF-1, -2, and -3; **Supplementary Table 1**) vector and empty vector controls (PLV-Ctr) were constructed by GeneChem (Shanghai, China). Ishikawa and RL95-2 cells were infected with lentiviral vectors, and cells with green fluorescent protein signals (**Supplementary Figure 1**) were selected for further experiments using qRT-PCR and western blotting analyses.

**Abbreviations:** co-IP, co-immunoprecipitation; DDX5, DEAD-box RNA helicase; EC, endometrial cancer; EMT, epithelial-mesenchymal transition; HDGF, hepatoma-derived growth factor; MTT, 3-(4 5-dimethylthiazol-2-yl)- 2, 5-diphenyltetrazolium bromide; PBS, phosphate buffered saline; PCNA, proliferating cell nuclear antigen; PI3K, phosphoinositide 3-kinase; qRT-PCR, quantitative real-time reverse transcription polymerase chain reaction.

# Transient Transfection Using Plasmids or Small Interfering RNAs or PI3K Inhibitor Ly294002

HDGF, DDX5, and β-catenin plasmids were generated by Biosense Technologies (Guangzhou, China). Small interfering RNA (siRNA) for DDX5 and β-catenin (named as siDDX5 and siβ-catenin, respectively) were designed and synthesized by Guangzhou RiboBio (RiboBio Inc., China). At 24 h before transfection, EC cells were plated onto a 6- or 96-well plate (Nest Biotech, China) at 30%–50% confluence. Plasmids were then transfected into cells using LipofectamineTM 2,000 (Invitrogen Biotechnology, Shanghai, China), according to the manufacturer's protocol. EC cells overexpressing HDGF were treated with or without Ly294002 according to a previous description (30). Cells were collected after 48–72 h for further experiments.

### RNA Isolation, Reverse Transcription, and qRT-PCR

RNA was extracted from EC cell lines and the empty vector controls using TRIzol (TaKaRa, Shiga, Japan). RNA was transcribed into cDNA (TaKaRa) and amplified with specific primers. The targeted HDGF sequences were: sense 5′ -GCT TCC GGC TAT CAG TCC TC-3′ ; antisense: 5′ -CTG CCT CCT TCT CCT CTC CT-3′ ; The targeted DDX5 sense: 5′ -GGC CTG ATC ACA GAA CCA TT-3′ ; antisense: 5′ -ACC ACC CTT ATT CCC AAA CC-3′ ; and ARF5 (used as an internal control) sense: 5′ - ATC TGT TTC ACA GTC TGG GAC G-3′ ; antisense: 5′ -CCT GCT TGT TGG CAA ATA CC-3′ . The assays were performed in accordance with the manufacturer's instructions (TaKaRa) and according to a previous description (31). The PCR reaction for each gene was repeated three times.

# Western Blotting Analysis

Western blotting was performed as previously described (32). The antibodies included anti-HDGF, DDX5, pRb, E2F1, CCND1, CDK4, c-Myc, P27, PI3K, p-PI3K, AKT, p-AKT, βcatenin, E-cadherin, N-cadherin, vimentin, Snail, and β-actin (**Supplementary Table 2**). β-actin was used as a loading control for all blots. The images were captured using a ChemiDocTM CRS+ Molecular Imager (Bio-Rad, Berkeley, CA, USA).

# MTT Assay

The 3-(4, 5-dimethylthiazol-2-yl)-2, 5-diphenyltetrazolium bromide (MTT) assay was used to evaluate the rate of in vitro cell proliferation. For MTT assay, cells were processed as described earlier (31). Briefly, after transfected with shHDGF or PLV-Ct, cells were incubated, dissolve and measured the absorbance value (OD) at 490 nm.

# EdU Incorporation Assays

Proliferating EC cells were examined using the Cell-Light EdU Apollo 567 in vitro Imaging Kit (RiboBio, Guangzhou, China), according to the manufacturer's protocol. Briefly, after incubation with 10 mM EdU for 2 h, EC cells were fixed with 4% paraformaldehyde, permeabilized in 0.3% Triton X-100, and stained with Apollo fluorescent dyes. A total of 5 mg/mL of DAPI was used to stain cell nuclei for 10 min. The number of EdU-positive cells was counted under a fluorescence microscope in five random fields. All assays were independently performed three times.

# Colony Formation Assay

Cells were plated in 6-well culture plates at 200 cells/well (3 wells/cell group). After incubation at 37◦C in a 5% CO<sup>2</sup> incubator for 15 days, cells were washed twice with phosphate buffered saline (PBS) and stained with hematoxylin solution. The visible colony numbers were counted. All experiments were repeated at least three times.

# Cell Cycle Analysis

A total of 5 × 10<sup>6</sup> EC cells were harvested after a 48-h incubation, and then washed with cold PBS. The cells were further fixed with 70% ice-cold ethanol at 4◦C overnight. Fixed cells were washed three times with cold PBS. After incubation with PBS containing 10 mg/mL of propidium iodide and 0.5 mg/mL of RNase A for 15 min at 37◦C. FACS caliber flow cytometry (BD Biosciences, San Jose, CA, USA) was used to obtain the DNA content of the labeled cells.

# In vivo Tumorigenesis in Nude Mice

The animal studies were approved by the Animal Ethics Committee of the Southern Medical University. A total of 5 × 10<sup>6</sup> logarithmically growing EC cells transfected with shHDGF or PLV-Ctr (n = 5 per group) in 0.1 mL of RPMI-1640 medium were subcutaneously injected into the left-right symmetric flank of 4–5-week-old male BALB/c-nu mice. The mice were maintained in a barrier facility on HEPA-filtered racks. The animals were fed an autoclaved laboratory rodent diet. After 21 days, the mice were sacrificed and tumor tissues were excised and weighed.

# Wound Healing Assay

EC cells were plated in 6-well plates and incubated overnight until 90% confluent. An injury line was made using a 10-µL plastic filter tip to create a wound approximately 10µm in diameter. Then we removed the culture medium and used PBS to eliminate dislodged cells. Subsequently, the wells were covered with serum-free medium to incubate for 48 h. "Wound closure" was observed at 0, 12, 24, 48 h under an inverted microscope.

# Transwell Migration and Invasion Assays

For cell migration assays, 1 × 10<sup>5</sup> cells in 100 µL of RPMI-1640 medium without serum were seeded on a fibronectincoated polycarbonate membrane insert in a Transwell apparatus (Corning, Armonk, NY, USA). In the lower chamber, 500 µL of RPMI-1640 with 10% serum was added as a chemoattractant. After the cells were incubated for 10 h at 37◦C in a 5% CO2 atmosphere, the insert was washed with PBS and cells on the top surface of the insert were removed with a cotton swab. Cells adhering to the lower surface were fixed with methanol, stained with Giemsa solution, and counted under a microscope in 5 pre-determined fields (200×). For the cell invasion assay, the procedure was similar to the Transwell migration assay, except that the Transwell membranes were pre-coated with 24µg/mL of Matrigel (R&D Systems, Minneapolis, MN, USA) for 4 h. All assays were independently repeated three times.

#### In vivo Metastasis Assays

In vivo metastasis assays were performed according to a previous study (30). A total of 5 × 10<sup>6</sup> EC-shHDGF and -PLV-Ctr cells were injected into nude mice (n = 5 for each group) through the liver membrane. Whole-body optical images were visualized to monitor primary tumor growth and formation of metastatic lesions. After 2 months, all mice were sacrificed, individual organs were removed, and metastatic tissues were analyzed by hematoxylin and eosin staining.

#### Co-immunoprecipitation (Co-IP)

Co-IP was carried out using a Pierce Co-Immunoprecipitation kit (Thermo Scientific, Waltham, MA, USA), according to the

manufacturer's instructions. The cells were lysed and the protein concentrations were measured. Then, 2000 µg of protein in 400 µL of supernatant was incubated with 10 µg anti-HDGF, anti-DDX5, and anti-β-catenin or anti-IgG antibodies coated on beads on a rotator overnight at 4◦C. The beads were washed, eluted in sample buffer, and boiled for 8 min at 95◦C. Immune complexes were subjected to Coomassie brilliant blue staining, mass spectrometry (Geneseed Biotech Co., Ltd, Guangzhou, China) and western blotting analysis. Anti-IgG was used as a negative control.

#### Immunofluorescence

Immunofluorescent staining was performed according to standard procedures as a previous study (33). Briefly, cells were cultured overnight, then fixed with 3.5% paraformaldehyde and permeabilization by 0.2% Triton X-100 at room temperature. Cells were incubated with mouse anti-β-catenin and rabbit anti-DDX5 antibody overnight at 4◦C. After three washes in PBS, the coverslips were incubated for 1 h with secondary antibody. Then, coverslips were mounted onto slides with mounting solution containing 0.2 mg/mL of DAPI and sealed with nail polish. Slides were stored in a dark box and observed using a scanning confocal microscope (Zeiss LSM 800, Oberkochen, Germany).

#### Statistical Analysis

SPSS 21.0 (Chicago, IL, USA) and Graph Pad Prism 5.0 software (LaJolla, CA, USA) were used to analyze all data for statistical significance. Comparisons between two groups were performed using Student's t-test, one-way analysis of variance (ANOVA) for multiple groups, and a parametric generalized linear model with random effects for tumor growth and the MTT assay. Data are expressed as the mean ± SD from at least three independent experiments. The chi-squared test was applied to determine the relationship between the level of DDX5 expression and clinicopathological characteristics. Analysis of HDGF and DDX5 expression in 122 EC tissues was performed using paired-samples t-tests. The relationships were analyzed using Spearman's correlation analysis. Survival analysis was performed using the Kaplan-Meier method. The multivariate Cox proportional hazards method was used for analyzing the relationship between variables and patient survival time. A prognostic value of < 0.05 (P < 0.05) was considered significant, and all tests were two-sided. Statistical significance was denoted as follows: <sup>∗</sup>P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001.

#### RESULTS

#### Stable Knockdown of HDGF Expression Inhibits Cell Proliferation in vitro and in vivo

To gain insight into the role of HDGF in EC, we first used a lentiviral vector to specifically and stably knockdown the expression of HDGF in the Ishikawa and RL95-2 cells. The levels of HDGF were assessed by western blotting (**Figure 1A**) and qRT-PCR (**Figure 1B**). The most efficient knockdowns of HDGF expression were shown in shHDGF-3-Ishikawa and shHDGF-1-RL95-2 cells compared to the PLV-Ctr (P < 0.001) (**Supplementary Figure 2**).

Next, we assessed the effect of decreased HDGF expression on EC cell growth in vitro. The growth curves determined by MTT assays showed that growth of the shHDGF-Ishikawa and shHDGF-RL95-2 cells was significantly slower than that of PLV-Ctr cells (P < 0.05; **Figure 1C**). Conversely, overexpression of

± SD for three independent experiments. \**P* < 0.05, \*\**P* < 0.01, \*\*\**P* < 0.001, statistically significant difference.

HDGF (**Figure 2A**) reversed these effects (P < 0.05; **Figure 2B**). The EdU incorporation assay revealed that the percentage of cells in S phase decreased following the downregulation of HDGF expression (P < 0.001; **Figure 1D**). Colony formation assays showed that knockdown of HDGF significantly decreased cell proliferation (P < 0.001; **Figure 1E**). Furthermore, cell

FIGURE 4 | HDGF controls cell cycle and EMT-associated gene expression in EC via the PI3K/AKT and β-catenin pathways. Knocking down endogenous HDGF expression reduced the expressions of pRb, E2F1, c-Myc, CDK4, and CCND1, and enhanced P27 in EC cells (A). Suppressing HDGF expression decreased the expression of EMT marker genes (N-cadherin, Vimentin, and Snail) and enhanced E-cadherin expression (B). Reduced HDGF expression decreased the expressions of p-PI3K, p-AKT, and β-catenin (C). Knocking down of HDGF suppressed both nuclear and cytosol protein expression of β-catenin (D). Suppressing the expression of p-PI3K by its specific inhibitor Ly294002 (50 nM) reduced p-AKT, β-catenin, Snail, N-cadherin, Vimentin, and upregulated E-cadherin in overexpressing HDGF EC cells (E,F). β-actin served as the internal control. Each experiment was repeated three times with similar results, and error bars represent the mean ± SD, \**P* < 0.05, \*\**P* < 0.01, \*\*\**P* < 0.001.

cycle analysis demonstrated that HDGF suppression dramatically reduced cell cycle progression from the G1 to S phase (P < 0.05; **Figure 1F**).

Subsequently, to confirm the growth effect of HDGF in vivo, we performed an in vivo tumorigenesis study by inoculating EC cells into nude mice. Mice in the shHDGF-EC and PLV-Ctr groups were sacrificed 21 days after inoculation, with average tumor weights of 0.187 g and 0.793 g, respectively (P < 0.001; **Figure 1G**). The mice injected with shHDGF-Ishikawa and shHDGF-RL95-2 cells had smaller tumor burdens (**Figure 1G**) and displayed lower expression of HDGF, Ki67 and proliferating cell nuclear antigen (PCNA) in tumor tissues relative to the controls (**Figure 1H**). These results suggested that HDGF significantly promotes tumorigenesis in vivo.

#### HDGF Downregulation Suppresses Cell Migration, Invasion, and Intrahepatic Metastasis of EC Cells in vitro and in vivo

To examine the effect of HDGF on cell migration and invasion, a Transwell apparatus, wound healing assay, and Boyden chamber coated with Matrigel were used. After incubation

(H). DDX5 mRNA expression after HDGF knockdown, normalized to the expression of ARF5. One-way ANOVA and Dunnett's multiple comparison test \*\*\**P* < 0.001 (I).

immunofluorescence under a scanning confocal microscope (D). DDX5 levels after HDGF knockdown, as assessed by western blotting (E). HDGF and β-catenin levels after DDX5 knockdown (F). Overexpression of β-catenin plasmid in shHDGF-EC cells increased DDX5 levels (G). DDX5 expression after β-catenin knockdown

for 10 h, a reduced number of migrated cells were observed for shHDGF-Ishikawa and shHDGF-RL9-2 cells compared with PLV-Ctr cells (P < 0.001; **Figure 3A**). In addition, the wound healing assay demonstrated that shHDGF-Ishikawa and shHDGF-RL95-2 cells inhibited the migration capacity (P < 0.05; **Figure 3B**). As shown in **Figure 3C**, the results of the Boyden chamber coated with Matrigel assays were similar to those of the Transwell assays (P < 0.01 for each); however, overexpressing HDGF reversed these effects (P < 0.05; **Figure 2C**).

To further assess the effect of HDGF on EC intrahepatic metastasis, shHDGF-Ishikawa and control cells were independently injected into the liver capsules of nude mice. Fluorescence imaging was used to identify scattered metastatic

Liu et al. HDGF/DDX5/β-Catenin Promotes Endometrial Cancer

nodules in the livers and intestines of nude mice that formed in the mice after 2 months. Only a few scattered metastatic cells were observed following injection of shHDGF-Ishikawa cells, whereas a variety of large clusters were observed in the PLV-Ctr group (**Figure 3D**). As can be seen from **Figure 3D**, control PLV-Ctr cells were more easily transferred to intestinal tissue, while shHDGF was less or not metastasized. The HE and IHC of the metastatic intestinal tissues and numbers of metastatic foci were shown in **Figures 3E–G**. Taken together, these results suggest that HDGF effectively promotes cell migration, invasion, and intrahepatic metastasis of EC cells in vitro and in vivo.

### HDGF Regulates the Expression of Cell Cycle- and EMT-Associated Genes via the PI3K/AKT and β-Catenin Signaling Pathways in EC Cells

To further study the mechanism by which HDGF regulates cell proliferation, migration, invasion, and metastasis, the protein levels of cell cycle- and EMT-associated genes were examined in Ishikawa and RL95-2 cells with stably suppressed HDGF. Knockdown of HDGF inhibited the activation of oncogenic cell cycle regulators, including pRb, E2F1, c-Myc, CCND1, and CDK4, and increased the level of P27 (**Figure 4A**). Further, we found that EMT markers (N-cadherin, Vimentin, and Snail) were suppressed, whereas the E-cadherin level increased (**Figure 4B**). Simultaneously, the levels of p-PI3K, p-AKT, and β-catenin were significantly decreased (**Figure 4C**). Further, we observed that knockdown of HDGF significantly suppressed both nuclear and cytosol protein expressions of β-catenin in EC cells (**Figure 4D**). In subsequent study, we used specific inhibitor (Ly294002) of PI3K to suppress the expression of p-PI3K and observed that the protein expression of p-PI3k, p-AKT, β-catenin, Snail, N-cadherin, Vimentin was decreased, and E-cadherin was upregulated in overexpressing HDGF EC cells (**Figures 4E,F**). These results suggest that HDGF regulates the expression of cell cycle- and EMT-associated genes via the PI3K/AKT and β-catenin signaling pathways in EC cells.

#### DDX5 Interacts With HDGF and β-Catenin

In previous study, our team had shown the interaction of βcatenin and HDGF or DDX5 in Lung Adenocarcinoma (33). To explore the precise molecular mechanisms of HDGF in EC, co-IP, combined with mass spectrometry, was used in Ishikawa cells. This analysis yielded 242 potential HDGF-interacting proteins (**Supplementary Table 3**), including DDX5 (69 kDa band) and β-catenin (92 kDa band). In addition, we used data sets from public domain data to draw a Venn diagram to show the proteins that interact with HDGF (34) and β-catenin (35) proteins, and observed that there were 22 overlapping proteins (**Figure 5A**). Exogenous and endogenous co-IP demonstrated that HDGF and DDX5 interact in Ishikawa cells, whereas β-catenin was associated with DDX5, but not HDGF (**Figures 5B,C**). Moreover, nuclear co-localization of DDX5 and β-catenin proteins, as well as that of HDGF and DDX5 proteins (33), was observed by immunofluorescence using a scanning confocal microscope (**Figure 5D**). Taken together, these results suggest that HDGF is associated with DDX5 in EC, as is the case with β-catenin and DDX5.

Subsequently, western blotting indicated that stable knockdown of HDGF expression resulted in DDX5 protein downregulation (**Figure 5E**); however, knockdown of DDX5 did not significantly affect HDGF expression, but resulted in β-catenin protein downregulation (**Figure 5F**). In addition,


we observed that β-catenin suppression decreased the level of the DDX5 protein (**Figure 5G**). Furthermore, overexpression of β-catenin from a plasmid in shHDGF-EC cells increased DDX5 expression (**Figure 5H**). qRT-PCR showed that the expression of DDX5 was reduced after knockdown of HDGF (**Figure 5I**), which indicated that HDGF affects DDX5 at the level of transcription.

#### DDX5 Overexpression Reverses the Suppression of shHDGF

Transiently transfecting DDX5 into shHDGF EC cells (**Figure 6A**) enhanced cell proliferation, as assessed by the MTT (**Figure 6B**) and EdU incorporation assays (**Figure 6C**). Transwell and Boyden chamber assays showed that EC cell migration and invasive ability were enhanced (**Figure 6D**). Furthermore, we found that DDX5 overexpression induced the expression of pRb, E2F1, c-Myc, CCND1, CDK4, Ncadherin, Vimentin, Snail, p-PI3K, p-AKT, and β-catenin, but reduced the expression of P27 and E-cadherin (**Figures 6E–G**). These results indicated that DDX5 overexpression can overcome the EC cell growth suppression induced by shHDGF.

#### Association of HDGF and DDX5 Expression With the Clinicopathological Characteristics of EC Tissues

Combined with data from our previous work (16), **Figures 7A–D** displays the expression of HDGF and DDX5. Immunohistochemical staining showed that HDGF and DDX5 positive signals were mostly located in the nuclei of EC cells, with minor cytoplasmic distribution. Our previous results indicated that 25.5% (31/122) and 74.5% (91/122) of cases exhibited high and low nuclear expression of HDGF, respectively (16). Here, we found that DDX5 protein was positively expressed in 32.8% (40/122) and not expressed in 67.2% of tumors (82/122; **Table 1**). DDX5 expression was significantly related to FIGO stage (I+II vs. III; P < 0.001; **Table 1**); however, no significant correlation existed with respect to age, menopausal status, histologic grading, depth of myometrial invasion, or lymph node status in patients with EC (P > 0.05; **Table 1**). The results were similar to those obtained for the HDGF protein.

## Correlation Analysis of HDGF and DDX5 With Overall Survival of Patients With EC

We stratified the data from patients with EC into four groups, according to the combination of different levels of HDGF and DDX5 expression: low HDGF/low DDX5 (HDGF–/DDX5–) expression; high HDGF/high DDX5 (HDGF+/DDX5+) expression; low HDGF/high DDX5 (HDGF–/DDX5+) expression; and high HDGF/high DDX5 (HDGF+/DDX5+) expression. Our previous work showed that 25.5% (31/122) and 74.5% (91/122) of cases exhibited high and low nuclear expression of HDGF, respectively (16). Here, we found that 22 EC tissues were HDGF+/DDX5+, whereas 73 cases were HDGF–/DDX5– (**Table 2**). The expression of HDGF+/DDX5+ correlated with the FIGO stage (P = 0.003; **Table 2**).

Patients with high expression of DDX5 had worse prognoses than those with low expression of DDX5 (P < 0.001; **Figure 7E**), and HDGF+/DDX5+ also correlated with a short survival time for patients with EC (P < 0.001; **Figure 7F**). We further


conducted survival analysis by strata analysis against FIGO stage. These results indicated that high DDX5 protein levels and HDGF+/DDX5+ are significantly associated with the survival time for patients with EC based on FIGO stage = I+II (P = 0.002, P = 0.011, respectively; **Figures 7G,I**) and FIGO stage = III (P = 0.012, P = 0.046, respectively; **Figures 7H,J**).

Univariate analyses showed that FIGO stage, histological grading, lymph node status, depth of myometrial invasion, high DDX5 expression, and post-operative hormone therapy significantly correlated with patient survival (P = 0.002, P = 0.001, P < 0.001, P < 0.001, P < 0.001, and P = 0.044, respectively; **Table 3**). Multivariate analysis showed that the level of DDX5 expression, FIGO stage, and depth of myometrial invasion were independent prognostic factors for EC (P = 0.038, P = 0.018, and P = 0.014, respectively; **Table 3**). However, HDGF expression, histological grading, lymph node status, and postoperative chemotherapy were not independent prognostic factors for EC (P = 0.984, P = 0.160, P = 0.702, and P = 0.631, respectively; **Table 3**).

#### Correlation Between HDGF and DDX5 Expression in EC

There was a significant positive correlation between HDGF and DDX5 protein levels in EC tissues (r = 0.475, P < 0.001; **Table 4**).

#### DISCUSSION

In previous studies, HDGF has been shown to promote the progression of tumors by activating the AKT–MAPK (36), Akt and TGF-β (37), and VEGF signaling pathways (38), and interacting with β-catenin as a positive feedback loop (39); however, the molecular mechanism involved in HDGFassociated EC cell proliferation, invasion, and metastasis has not been elucidated.

In the present study, we observed that HDGF knockdown markedly decreased cell proliferation, migration, invasion, and metastasis in vitro. Furthermore, subcutaneous tumor

TABLE 3 | Summary of univariate and multivariate Cox regression analysis of overall survival duration.


experiments in nude mice demonstrated that knockdown of HDGF inhibited tumorigenesis, invasion, and metastasis in vivo. These findings are consistent with an earlier report by Zhou et al. (40) in which downregulation of HDGF inhibited the proliferation and invasiveness of hepatocellular carcinoma cells.

Multiple studies have shown that PI3K/AKT constitutes a key signal mediator during carcinogenesis (41, 42) and that activation of PI3K/AKT may regulate β-catenin signaling through Akt phosphorylation and inactivation of GSK3-β (43). In the current study, we observed that decreased HDGF expression suppressed p-PI3K and p-AKT levels, and the downstream β-catenin-mediated cell cycle and EMT signal molecules, such as pRb, E2F1, c-Myc, CCND1, CDK4, Snail, Vimentin, and Ncadherin, while elevating the expression of P27 and E-cadherin in EC cells. In addition, using Ly294002 to treat overexpressing HDGF EC cells induced a decrease of p-PI3K, p-AKT, βcatenin and metastatic effect related proteins such as Snail, Vimentin, and N-cadherin. Therefore, we hypothesized that HDGF knockdown significantly suppresses EC cell proliferation, migration, invasion, and metastasis through the PI3K/AKT and β-catenin pathways in EC.

To better understand the molecular mechanisms underlying HDGF promotion of EC proliferation and metastasis, we searched public domain data and screened DDX5 as a candidate interaction protein of HDGF. DDX5 is a member of DEAD box family proteins that plays an important role in the progression of many tumors (44–48). Subsequently, we observed that HDGF not only combined with DDX5 but also induced the expression of DDX5 in EC. Furthermore, DDX5 could reverse the inhibitive effects on cell growth, migration, and invasion in shHDGF EC cells. Previously, HDGF was found to bind the promoter of βcatenin, resulting in β-catenin transcriptional activation (39). Phosphorylated p68 (DDX5) can enter the cytoplasm, which leads to its interaction with β-catenin and displacement of Axin (49). DDX5 (P68) forms a complex with β-catenin and facilitates its transcription activation to regulate both cell adhesion and gene expression (23). Notably, as β-catenin can, in turn, function as a transcription factor to stimulate DDX5 expression by binding to its promoter (50), we thus speculated that HDGF upregulated the expression of DDX5 by inducing β-catenin expression. Consistent with this speculation, we observed that DDX5 was significantly increased in shHDGF-EC cells after transfection of the β-catenin cDNA.

In previous reports, DDX5 had been documented to promote cell proliferation and EMT by interacting with Wnt-β-catenin signaling and inducing the nuclear translocation of β-catenin (23,



26, 28, 51). Consistent with these previous reports, we observed that DDX5 not only directly combined with β-catenin, but also induced the expression of β-catenin via activating PI3K/AKT signaling and stimulated its nuclear translocation, thus inducing cell cycle transition and EMT signaling and the promotion of cell growth and metastasis in EC. These data demonstrated that HDGF interacts with DDX5 to further induce β-catenin to participate in cell growth and metastasis in EC.

In our previous study of HDGF expression in 122 samples of EC, we found that high nuclear expression of HDGF was positively correlated with FIGO stage (P = 0.032) and that patients with high expression of HDGF had poorer overall survival rates compared to those with low expression (P = 0.001) (16). In the current study, we further showed that DDX5 protein was expressed in 32.8% (40/122) of EC samples. Similar to HDGF, DDX5 expression was also significantly associated with FIGO stage (P < 0.001), and patients with high expression of DDX5 had poorer overall survival rates than those with low expression (P < 0.001). Multivariate analyses showed that high expression of DDX5 was an independent predictor of prognosis for patients with EC (P = 0.038). Notably, our data showed a significant positive correlation between HDGF and DDX5 in EC tissues (r = 0.475, P < 0.001). Finally, we observed that high nuclear levels of HDGF and DDX5 led to the worst prognosis for patients with EC.

#### CONCLUSIONS

In summary, this study provides compelling evidence that HDGF in combination with DDX5 induces β-catenin to form a complex, which significantly promotes EC cell proliferation, migration,

invasion, and metastasis in vitro and in vivo. The underlying mechanism likely involves the activation of PI3K/AKT signaling and downstream β-catenin-mediated cell cycle and EMT signaling proteins (**Figure 8**). Our results suggest that HDGF– DDX5 and β-catenin work together to play important roles in EC carcinogenesis and progression.

#### ETHICS STATEMENT

For the use of the human tissue specimens for research purposes, prior consent from the patients and approval from the Ethics Committees of the Third Affiliated Hospital of Guangzhou Medical University were obtained. All animal studies were conducted in accordance with the principles and procedures outlined in the Southern Medical University Guide for the Care and Use of Animals.

#### REFERENCES


#### AUTHOR CONTRIBUTIONS

CL, LW, QJ, and JZ conducted the research. SG and WF designed the research study. CL, LZ, LL, HJ, DL, and YX performed the statistical analysis. CL, LW, LZ, and WF wrote the manuscript. All authors have read and approved the final manuscript.

#### FUNDING

This project was funded by the Nature Science Fund of Guangdong Province (2015A030313240) and the Medical Research Fund of Guangdong Province (A2015467).

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fonc. 2019.00211/full#supplementary-material


carcinogenesis of colorectal cancer. Oncotarget. (2015) 6:29357–74. doi: 10.18632/oncotarget.4982


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Liu, Wang, Jiang, Zhang, Zhu, Lin, Jiang, Lin, Xiao, Fang and Guo. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Anti-estrogen Treatment in Endometrial Cancer: A Systematic Review

Willem Jan van Weelden<sup>1</sup> \*, Leon F. A. G. Massuger <sup>1</sup> , ENITEC† , Johanna M. A. Pijnenborg1‡ and Andrea Romano2‡

*<sup>1</sup> Department of Obstetrics and Gynecology, Radboud Institute for Health Sciences, Radboud University Medical Center, Nijmegen, Netherlands, <sup>2</sup> Department of Obstetrics and Gynecology, GROW–School for Oncology and Developmental Biology, Maastricht University Medical Centre, Maastricht, Netherlands*

#### Edited by:

*Paola Gehrig, University of North Carolina at Chapel Hill, United States*

#### Reviewed by:

*Emma Rossi, University of North Carolina at Chapel Hill, United States Kevin Schuler, TriHealth, United States*

#### \*Correspondence:

*Willem Jan van Weelden willemjan.vanweelden@radboudumc.nl*

*†ENITEC: European Network for Individualised Treatment of Endometrial Cancer (within the European Society of Gynaecological Oncology: www.esgo.org; www.esgo. org/network/enitec/)*

*‡These authors share last authorship*

#### Specialty section:

*This article was submitted to Women's Cancer, a section of the journal Frontiers in Oncology*

Received: *14 January 2019* Accepted: *18 April 2019* Published: *07 May 2019*

#### Citation:

*van Weelden WJ, Massuger LFAG, ENITEC, Pijnenborg JMA and Romano A (2019) Anti-estrogen Treatment in Endometrial Cancer: A Systematic Review. Front. Oncol. 9:359. doi: 10.3389/fonc.2019.00359* Introduction: Hormonal therapy in endometrial cancer (EC) is used for patients who wish to preserve fertility and for patients with advanced or recurrent disease in a palliative setting. First line hormonal therapy consists of treatment with progestins, which has a response rate of 25% in an unselected population. Treatment with anti-estrogens is an alternative hormonal therapy option, but there is limited data on the effect and side-effects of anti-estrogens in EC. Therefore, we performed a systematic review to investigate the response rate and toxicity of anti-estrogenic therapy in patients with endometrial cancer.

Methods: A systematic search in electronic databases was performed to identify studies on selective estrogen receptor modulators (SERM) and down-regulators (SERD) and aromatase inhibitors that reported on response rates (RR) among EC patients. Outcome in estrogen receptor (ER) positive and negative disease was assessed independently.

Results: Sixteen studies on advanced stage and recurrent EC were included. Ten studies investigated anti-estrogen monotherapy and seven investigated a combination of anti-estrogenic drugs with either progestin or targeted treatment. Due to heterogeneity in patient population, no meta-analysis was performed. The median age of the patients in the included studies ranged from 61 to 71 years and the proportion of low grade tumors ranged from 38 to 80%. The RR for tamoxifen ranged from 10 to 53%, for other SERMs and SERDs 9–31%, for aromatase inhibitors from 8 to 9%, for combined tamoxifen/progestin treatment 19–58%, for combined chemo- and hormonal therapy 43% and for combination of anti-estrogenic treatment with mammalian target of rapamycin (mTOR) inhibitors 14–31%. Toxicity consisted mainly of nausea and thrombotic events and was higher in combination therapy of chemotherapy and hormonal therapy and hormonal therapy and mTOR inhibitors compared to other therapies.

Conclusion: Tamoxifen or a combination of tamoxifen and progestin should be the preferred choice when selecting second line hormonal treatment because the RRs are similar to first line progestin treatment and the toxicity is low. The response can be optimized by selecting patients with endometrioid tumors and positive estrogen receptor status, which should be based on a pretreatment biopsy.

Keywords: endometrial cancer, anti-estrogen, tamoxifen, fulvestrant, aromatase inhibitor, review

# INTRODUCTION

Endometrial cancer (EC) is the most common gynecologic malignancy in the Western world (1). The incidence of EC is increasing and is expected to rise further in the coming years (2). The most important risk factors for the development of EC are related to exogenous or endogenous estrogen exposure, including: estrogen medication, nulliparity, early menarche, late menopause, and obesity, which contributes to estrogen exposure by aromatase dependent conversion of androgen into estrogen (3–7). In general, two types of EC are identified based on tumor histology and presumed carcinogenesis. Endometrioid EC (EEC) represents 80% of EC cases and most EECs are caused by an excess estrogen exposure that, in the absence of counteractive effects of progesterone, induces endometrial proliferation and subsequent endometrial hyperplasia and cancer (8). Non-endometrioid EC (NEEC) is responsible for 20% of EC incidence and is assumed to develop independent of estrogen (8, 9). Standard therapy for EC consists of surgery followed by adjuvant radio- and/or chemotherapy depending on final tumor characteristics (10, 11). Hormonal therapy is an alternative treatment for patients who wish to preserve their fertility, and for those with metastatic or recurrent disease without curative options (12). Historically, progestin therapy has been the most widely applied hormonal treatment and it is still the preferred choice as first line hormonal therapy (10, 13). In addition to progestins, inhibition of estrogen-induced proliferation by anti-estrogens is used as an alternative to progestin treatment in EC (14). Currently used anti-estrogenic drugs are selective estrogen receptor modulators (SERM) or down-regulators (SERD) and aromatase inhibitors. SERMs and SERDs such as tamoxifen and fulvestrant have an antiproliferative effect by blocking the estrogen receptor (ER) through which estrogen effects are mediated. Within the group of SERMs, tamoxifen has both stimulatory and blocking effects on ER in the endometrium, while other SERMs like raloxifene and arzoxifene only block ER (15–17). Fulvestrant, the main SERD, only has antagonistic effects through down regulation of ER (18). Aromatase inhibitors like anastrozole, letrozole, and exemestane, limit the estrogen tumor exposure by aromatase in fat tissue, especially in postmenopausal women (12). The use of anti-estrogens is well established in breast cancer, but up till now, there is limited data on the response rates in EC. In one systematic review and meta-analysis, first and second line hormonal therapy in recurrent EC was evaluated, but the different types of hormonal therapy were not evaluated separately (19). Two separate reviews presented an overview of available (pre)clinical evidence on, respectively, fulvestrant and aromatase inhibitors. Unfortunately, no complete overview of anti-estrogenic treatment was given (20, 21). As a consequence, choice for anti-estrogenic drugs as second line hormonal therapy is based on experience of the treating physician, rather than on refined and up-to-date clinical data. Therefore, we performed this systematic review to determine the response rates and toxicity of anti-estrogenic therapy in patients with endometrial cancer and to relate it to the response rate of progestin therapy.

#### METHODS

#### Search Strategy

This review was performed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (22). An electronic search was performed in the following databases from inception until 3rd of October 2018: Pubmed, Embase, clinicaltrials.gov and Cochrane database of Systematic Review. The search string included "endometrial cancer," outcome measures like "response rate," "disease progression," or "survival" and drug terms like "estrogen antagonists," "aromatase inhibitors," "estrogen receptor modulators," "estrogen receptor down-regulator," and individual drug names. The full search string is shown in **Supplementary Table 1**. Citations of relevant articles and reviews were manually screened to ensure that no study was missed and that the search was complete.

# Study Selection

Studies were included if they reported on (1) women with endometrial cancer, who used anti-estrogenic therapy for fertility preservation or for advanced or recurrent disease. Studies investigating (2) estrogen receptor modulators, estrogen receptor down-regulators or aromatase inhibitors were eligible if (3) clinical outcome was reported. Studies reporting findings on patients with sarcomas or endometrial stroma sarcomas were excluded as well as conference papers, reviews and letters to the editor. Case reports and case series with <10 patients were excluded. Studies that combined anti-estrogen treatment with other therapy, i.e., progestins, chemotherapy, or targeted therapy were included and reviewed separately.

#### Data Extraction and Quality Assessment

Data from included articles was extracted using data collection forms with information regarding study design, in- and exclusion criteria, number of included patients, age, tumor stage and grade, estrogen (ER) and progesterone receptor status, previous treatment(s) and complete response (CR), partial response (PR), stable disease (SD), progressive disease, progression free survival (PFS), and overall survival (OS) was noted. Additional information was requested from study authors if necessary.

The quality of each individual study was assessed in five domains based of the National Institute of Health Quality Assessment Tool for Case Series Studies (23). Each full-text article was evaluated independently by three authors (WvW, JP, and AR) and risk of bias was subsequently discussed in a consensus meeting.

#### Outcome Assessment and Statistical Analyses

The primary outcome was the response rate (RR) to hormonal therapy and was defined as the proportion of patients with CR and PR. Other outcomes were the clinical benefit rate (CBR), which is defined as the proportion of patients with either CR, PR, or SD and toxicity which is defined as any adverse event occurring during treatment. Toxicity was ideally evaluated with a standardized measuring scale including grading of severity.

Individual treatment arms of randomized studies were analyzed separately. RR and CBR are reported for tamoxifen, other SERMs/SERDs, aromatase inhibitors, combination regimens and for ER positive and negative tumors separately. The specific expression of the two ER isoforms (ERα and ERβ) was not considered. Due to the large heterogeneity in the included studies, meta-analysis could not be performed. In case it was not reported in the study, the 95% confidence interval for RR and CBR was calculated using the normal approximation method of the binomial confidence interval (24).

# RESULTS

The search resulted in identification of 2,592 records. After removal of duplicates, 2,245 unique records were screened on title and abstract. For the systematic review, 2,208 records were excluded, leaving 37 articles for full text evaluation (**Figure 1**). A total of 21 articles were excluded from the final analysis due to: case reports or case series with <10 patients (n = 8), reports on the same patient cohort (n = 4) or studies that were outside the scope of the review (n = 6), including studies on endometrial stroma sarcoma and studies on chemotherapy and radiotherapy (25–30). Three other studies published between 1983 and 1990 could not be evaluated because the full text articles were not available (31–33). In addition, nine ongoing studies were identified (34–42).

#### Included Studies

Sixteen studies were included in the final systematic review. All included studies investigated patients with advanced stage and

#### TABLE 1 | Bias assessment.


*, Low risk of bias; , High risk of bias.*

recurrent EC. Ten studies described the use of monotherapy of which one reported outcomes on mono- and combined therapy, resulting in a total of seven studies on combined therapy (43– 58). There were two case series and 14 prospective studies. Bias was assessed as recommended by the National Institute of Health on five criteria (blinded treatment arms, robustness of outcome assessment, completeness of the data, selective outcome reporting and other biases) (23). Results of bias assessment and conflict of interest disclosures are shown in **Table 1**. Blinded treatment was not performed in any study included in the systematic review and was therefore regarded as high risk in all studies. Outcome assessment was performed with objective and reproducible criteria in all but one study (43). Two studies had a high risk of bias in three domains and were considered low quality studies (43, 48). All included studies investigated the effect of hormonal therapy among patients with advanced or recurrent EC.

#### Anti-estrogens as Monotherapy

An overview of the included studies that evaluated anti-estrogens as monotherapy in advanced and recurrent EC is shown in **Table 2**. Four studies investigated the use of tamoxifen, three studies investigated other SERMs or SERDs and three other studies reported on the use of aromatase inhibitors. Among all studies, the median age of included patients ranged from 61 to 71 years, and the proportion of patients with NEEC histology varied between 8 and 48%.

The overall RR of anti-estrogen monotherapy ranged from 8% (95% CI: 1–15) to 53% (95% CI: 29–78) among included studies (**Figure 2**). For tamoxifen the RR ranged from 10% (95% CI: 6– 18) to 53% (95% CI: 29–78), for the other SERMs and SERDs the RR varied between 9% (95% CI: 2–17) and 31% (95% CI: 15–51) and for aromatase inhibitors the RR ranged from 8% (95% CI: 1–15) to 9% (95% CI: 2–25). Results of the RR and CBR of all individual studies are illustrated in **Figure 2**.

Toxicity was scored according to a standardized scale in 6 out of 10 eligible studies. The remaining four studies did not report toxicity at all [n = 1, (43)] or did not report severity of complaints [n = 3, (44, 46, 49)] (**Table 2**). Nausea and thromboses were the most common side-effects. Thrombotic events were not reported in studies investigating tamoxifen or arzoxifene. The use of fulvestrant resulted in thrombosis in 6% of patients. Aromatase inhibitors resulted in thrombosis in 3–5% of patients.

#### Anti-estrogens in Combined Therapy

As shown in **Table 3**, the seven studies included in our analysis investigated either a combination of progestin and tamoxifen (four studies), a combination of chemotherapy with progestin and tamoxifen (one study), or a combination of anti-estrogen treatment with mammalian target of rapamycin (mTOR) inhibitors (two studies). Two studies on progestin/tamoxifen combined daily progestin with tamoxifen while the other two studies alternated between progestin and tamoxifen or added progestin to daily tamoxifen only in even weeks.

Among the seven studies, median age ranged from 61 to 70 years, and the proportion of low grade EEC tumors ranged from 38 to 80%. The overall RR of combined therapy ranged from 14% (95% CI: 3–36) to 43% (95% CI: 23–64). For combined progestin/tamoxifen treatment the RR varied between 19% (95%


*response; SD, stable disease.*

CI: 7–31) and 37% (95% CI: 27–47), for chemotherapy with progestin/tamoxifen the RR was 43% (95% CI: 23–64) and for combination therapy of hormonal treatment and mTOR inhibitor the RR ranged from 14% (95% CI: 3–36) to 31% (95% CI: 17–49) (**Figure 3**).

Toxicity was scored according to a standardized scale in five out of seven studies. Thrombosis occurred in 2% of patients with daily tamoxifen and progestin in even weeks only and in 9% of patients who alternated tamoxifen with progestins (51, 53). Chemotherapy and progestin/tamoxifen resulted in moderate to severe hematologic or gastro-intestinal toxicity in 14 and 12% of the patients (45). Seventy-five percent of patients received the optimal treatment dose. The combination of the mTOR inhibitor temsirolimus with progestin and tamoxifen resulted in serious thrombotic events in 43% of the patients, causing a premature stop to accrual in this study (56). The study that combined everolimus and letrozole reported grade 3–4 fatigue in 11% and nausea or vomiting in 6% of the patients (58). No thrombosis was reported. Thirty-two percent of patients required a dose reduction because of side effects, but no patient had to stop treatment due to toxicity.

#### Effect According to ER Status

Among all included studies, six investigated RR and/or CBR for patients with ER positive and ER negative tumors separately (**Table 4**). Tumor tissue used for ER analysis was taken either before start of hormonal therapy, from the primary tumor or from the recurrence. Immunohistochemical analysis for ER was performed using a staining-intensity index in most studies with different cutoff values, although two studies dichotomized ER status based on percentage of positive tumor cells. RR in ER positive patients ranged from 10% (95% CI: 1–19) to 47% (95% CI: 25–70) and RR in ER negative patients was 0% in all but one study. The highest RRs were found in studies that based ER positivity on tumor samples taken from the metastatic site before start of hormonal therapy. CBR ranged from 35% (95% CI: 20– 50) to 59% (95% CI: 39–80) in ER positive to 0 to 18% (95% CI: 2–34) in ER negative disease.

#### DISCUSSION

In this systematic review, we have outlined the effect of selective estrogen receptor modulators (SERM), down-regulators (SERD) and aromatase inhibitors in patients with advanced stage and recurrent endometrial cancer (EC). None of the included studies investigated patients with a wish to preserve fertility indicating that there is a lack of evidence for anti-estrogenic treatment in this population. Among studies on advanced stage and recurrent EC, comparison between different types of anti-estrogenic drugs


Frontiers in Oncology | www.frontiersin.org

 *response;* 

 *cancer;* 

 *non* 

 *cancer.*

was challenging because of the lack of randomized studies and differences in patient and tumor characteristics. The investigated treatments reported similar response rates for all treatments except for aromatase inhibitors, which had a limited effect in the investigated populations. Serious side-effects were rare for antiestrogens, but occurred more frequently when anti-estrogenic drugs were combined with chemotherapy or mTOR inhibitors. The expression of estrogen receptor in the tumor taken prior to start of treatment was associated with improved response to anti-estrogens.

The observed RR and CBR differed according to the selected population, with higher response rates in endometrioid tumors with positive ER status. Among studies investigating tamoxifen, Rendina et al reported a RR as high as 36% (95% CI: 22– 50) in patients with predominantly grade 1–2 tumors, whereas Thigpen found a RR of 10% (95% CI: 6–18) in patients with NEEC histology in 44% of cases (44, 49). The limited therapeutic response in NEEC reflects the low impact of estrogen in the carcinogenesis of these tumors (60). The reported tamoxifen related toxicity was limited to nausea.

As expected, the therapeutic response to anti-estrogens was higher among EECs, as illustrated by the study of McMeekin in which a RR of 31% (95% CI: 15–51) to arzoxifene was reported in a cohort that included only EEC. Despite these data, arzoxifene was never introduced into clinical practice. Two studies explored the use of fulvestrant, reporting limited responses ranging from 9 to 11%. Furthermore, fulvestrant can only be administered through intramuscular injection because of low oral bioavailability, which complicates the clinical implementation of this drug in a palliative setting. Aromatase inhibitors were shown to have only limited response rate in the investigated populations. Thus, aromatase inhibitors should not be a first choice when selecting anti-estrogenic therapy for EC. As aromatization of androgens into estrogen occurs predominantly in fat tissue, patients with obesity might represent a subgroup of EC patients in which aromatase inhibitors can be more effective. However, this hypothesis has not been tested in EC patients and studies in breast cancer do not show superior results of aromatase inhibitors compared to other hormonal treatments in obese patients (61).

Out of the four studies investigating combined treatment of tamoxifen and progestin, three studies enrolled a comparable patient population and reported a RR ranging from 19 to 27%. Also considering that serious toxicity occurred in just 2–5% of the included patients, the use of tamoxifen combined with progestins is an attractive treatment regimen. The addition of progestin and tamoxifen to chemotherapy was evaluated by one study, which reported a higher RR for the combination compared to chemotherapy alone. However, the applied chemotherapy regimen in this study is no longer in use in EC and no studies that combined anti-estrogenic therapy with currently used chemotherapeutic drugs have been performed (62). The combination of hormonal therapy with an mTOR inhibitor did not result in superior RRs compared with other anti-estrogenic treatments. Toxicity remains an important concern, especially for the combination of temsirolimus with alternating treatment with progestin and tamoxifen. Interestingly, the combination of letrozole with



*MPA, medroxyprogesterone-acetate; TMX, tamoxifen; MA, megestrol acetate.*

everolimus was less toxic. A recent GOG study presented at the SGO meeting 2018 showed similar RRs and adverse events for letrozole/everolimus and progestin/tamoxifen (63). Upon validation, this regimen could be an alternative to progestin/tamoxifen. Further investigation into molecular alterations that lead to resistance to hormonal therapy might also provide us with improved individualized combination treatment for these patients (64).

In summary, treatment with tamoxifen or combined treatment of tamoxifen and progestin are currently the best options in anti-estrogen therapy, because of similar or higher RR when compared to other treatments and limited toxicity. Preferably, patients with ER positive tumor and endometrioid histology should be selected for anti-estrogen therapy in order to optimize the chance of response.

Whether combined tamoxifen/progestin results in improved response when compared to progestins, has unfortunately not been studied in a randomized trial. The only study that randomized between progestin and progestin with tamoxifen was a low quality study that stopped the progestin arm prematurely due to poor accrual (48). However, several good quality studies reported an average response rate of 25% to progestin in an unselected population, which is similar to the responses to tamoxifen and progestin/tamoxifen found in this review (65, 66). The rationale for adding tamoxifen to progestin is to counteract the down regulation of the progesterone receptor that is induced by progestin treatment in order to prolong the duration of response (67, 68). Different combinations of progestin and tamoxifen have been explored. One option is to start progestin monotherapy and add or replace progestin by tamoxifen upon progression, as shown by two studies among progestin unresponsive patients (43, 44). Alternatively, combined treatment of tamoxifen with progestin or alternating treatment can be applied. From the reported RR in our study, it is not possible to define which regimen is superior.

Immunohistochemical expression of ER was evaluated by the studies included in this review using different methods and cutoffs for positivity. One study defined an optimal cutoff based on a staining intensity index, but even among ER negative patients, a high response rate of 26% was observed suggesting that differentiation between ER positive and ER negative can still be optimized (59). Future studies on this topic would ideally result in a test that can be used for all types of stored and fresh EC tissues and will be adopted worldwide. Most studies used primary tumor tissue for ER analysis. Yet, primary tumor and metastases are not comparable due to changes in the tumor caused by intercurrent therapy and the metastatic process itself (69–71). Therefore, it is essential that tumor tissue is obtained directly before start of hormonal therapy to relate receptor status to response. In case tumor tissue cannot be procured, non-invasive visualization of estrogen receptor status on a PET scan with an estrogen tracer might be an alternative approach (72, 73).

While the strengths of this review include the systematic approach and the quality assessment for eligible studies, there are some limitations to be addressed. First, systematic reviews are based on published data, and may therefore be biased toward selective reporting of positive results. Although we have tried to improve the quality by excluding case series with <10 patients, still this limitation should be taken into account. Second, criteria for response duration were not consistently used among all studies hampering proper comparison of outcome. Finally, most of the included studies evaluated patients with advanced and recurrent EC. However, both patient groups might differ in patient and tumor characteristics. Unfortunately, we could not discriminate in this review between advanced stage and recurrent EC, since most studies did not report outcome separately for both groups.

The effect of anti-estrogens in advanced and recurrent EC needs further improvement. In our review, the average response or clinical benefit rates were (far) below 50% and the effect of anti-estrogen therapy on progression free survival and overall survival was limited. Therefore, there is a need for additional biomarkers to improve selection of patients that benefit most from anti-estrogen hormonal therapy. Currently, selection for hormonal treatment is mainly based on estrogen and progesterone receptor status. However, several studies observe a benefit for patients even in ER negative disease, highlighting the need for in depth analysis of the intracellular pathway that is activated upon binding of estrogen to the estrogen receptor (54, 58, 59). An initial study on this topic has reported promising results in breast cancer, but so far no research on this topic has been performed in endometrial cancer (74). Furthermore, recent studies have shown that proteins involved in intracellular conversion of inactive estrogens to active estrogens have a prognostic role in EC (75, 76). These proteins can theoretically also oppose the effects of hormonal therapy warranting further research on this topic. Also, combining hormonal therapy with targeted therapies is an attractive strategy to overcome resistance to hormonal treatment and is the subject of many of the ongoing studies (34–36, 38, 39, 64). Finally, new studies should focus on patients with stable disease instead of complete or partial response only. Stable disease can be considered of clinical benefit for patients in a palliative setting especially if the disease remains stable for several months. Ideally, future studies would incorporate a predefined period of stable disease as outcome measure and would report on clinical benefit as primary outcome, as some of the included studies already have (57, 58).

#### REFERENCES


#### CONCLUSION

Treatment with tamoxifen or the combination of tamoxifen and progestin should be first choice in anti-estrogen therapy for patients with advanced and recurrent endometrial cancer because response rates are comparable to first line hormonal treatment with progestins and toxicity is limited. Therefore, these therapies are a good second-line hormonal treatment option in endometrial cancer. Responses to anti-estrogen therapy can be improved by selecting patients with endometrioid tumors and positive estrogen receptor status, which should be based on a pretreatment biopsy.

#### AUTHOR CONTRIBUTIONS

AR and JP came up with the concept. WvW and AR performed the search and selected relevant articles. WvW, AR, and JP performed bias assessment. WvW, LM, JP, and AR interpreted the data. WvW, AR, and JP wrote the first draft. All authors contributed to manuscript revision, read and approved the submitted version.

#### FUNDING

The study was partly supported by the Dutch Cancer Society (KWF Kankerbestrijding, www.kwf.nl), contract number UM-13-5782 granted to AR.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fonc. 2019.00359/full#supplementary-material


concentrations, and uterine endometrium in postmenopausal women. N Engl J Med. (1997) 337:1641–7. doi: 10.1056/NEJM199712043372301


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 van Weelden, Massuger, ENITEC, Pijnenborg and Romano. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.