Long Non-coding RNAs act as Prognostic Biomarkers in Breast and Gynecologic Cancers

Background It is well-known that long non-coding RNAs (lncRNAs) play essential roles in cancer development and progression. This study aimed to assess the potential prognostic value of specic lncRNAs in breast and gynecologic cancers. Methods systematically and identied according to eligibility criteria. A random-effects model was adopted to calculate combined hazard ratios to explore the association between specic lncRNA expression level and survival in breast and gynecologic cancers. Subgroup and publication bias analyses were also conducted. Results 111 studies encompassing nearly 20000 participants and 25 lncRNAs were included in the current study. Of the listed lncRNAs, we identied 3 lncRNAs signicantly associated with both overall survival (OS) and disease-free survival (DFS) in breast and gynecologic cancers, indicating that they might act as promising prognostic biomarkers in clinical applications. Specically, HOTAIR and PVT1 had a negative impact on survival outcome, while GAS5 was associated with better prognosis. Further subgroup analyses identied HOTAIR as a biomarker for the poor survival whether in an Asian population or in European and American populations and GAS5 as a biomarker for the relatively good prognosis of both breast and gynecologic cancers. We here highlight that abnormal expression of 3 lncRNAs, including HOTAIR, GAS5, PVT1 might signicantly affect the survival of breast and gynecologic cancer patients and serve as novel prognostic biomarkers for breast and gynecologic cancers.


Introduction
Breast and gynecologic cancers, including breast cancer, ovarian cancer, endometrial cancer and cervical cancer, are the major malignant tumors in women worldwide and seriously endanger the health of women, which account for nearly 40% of cancer risk and 30% of cancer mortality for women [1]. As estrogen associated tumors, it is widely accepted that breast cancer and gynecologic cancers share several similar risk factors and genetic characteristics [2,3]. Besides, the prognosis of breast and gynecologic cancers is still not optimistic owing to the distinctive properties of easy metastasis [4,5].
Recent genome sequencing studies have shown that the human genome consists of less than 2% proteincoding genes, and more than 90% of genome is transcribed into non-coding RNAs [6,7]. Long non-coding RNAs (lncRNAs) are de ned as transcripts greater than 200 nucleotides in length without protein encoding potential [8]. Although lncRNAs are incapable of encoding proteins, it can regulate gene expression at various levels, such as transcriptional regulation, post-transcription regulation and epigenetic mode [9,10].
Nowadays, an increasing number of lncRNAs have been identi ed due to the development of highthroughput sequencing.
Interestingly, plenty of lncRNAs were only expressed in differentiated tissues or speci c cancer types [11]. Previous studies demonstrated the participants of lncRNAs in different types of human cancers with both oncogenic and tumor repressive effects. And their enrichment, conserved sequences, and altered expression have been observed in breast and ovarian cancers [12]. It was proposed that lncRNAs could target chromatin modi cation complexes or RNA binding proteins to alter gene expressing programs which exhibited distinct gene expression patterns in primary tumors and metastases [13]. In addition, available studies have recognized the considerable role of lncRNAs in various stages of carcinogenesis and metastasis [14]. And metastasis was the most important biological behavior of tumor development and progression and closely related to cancer prognosis. Therefore, speci c lncRNAs might be explored as potential prognostic biomarkers for breast and gynecologic cancers.
Several studies have evaluated the association between tissue lncRNA expression and breast and gynecologic cancers but the results did not reach a consensus with each other. For instance, Gupta et al. demonstrated that HOTAIR was increased in expression in primary tumors and metastases in breast cancer [15]

Literature search and study selection
The literature searches were performed in PubMed, EMBASE and Cochrane library (up to January, 2019) for eligible studies. The search terms were shown as follows: ("lncRNA" or "Long ncRNA" or "Long Non-Translated RNA" or "Long Non-Coding RNA" or "Long Non Coding RNA" or "Long Non-Protein-Coding RNA" or "Long Non Protein Coding RNA" or "Long Noncoding RNA" or "Long Untranslated RNA" or "Long ncRNAs" or "Long Intergenic Non-Protein Coding RNA" or "Long Intergenic Non Protein Coding RNA" or "LincRNAs" or "LINC RNA") AND ("Ovarian neoplasm" or "Ovarian cancer" or "Ovarian tumor" or "Ovarian tumour" or "Ovarian carcinoma" or "Ovarian malignancy" or "Endometrial neoplasm" or "Endometrial cancer" or "Endometrial tumor" or "Endometrial tumour" or "Endometrial carcinoma" or "Endometrial malignancy" or "Cervical neoplasm" or "Cervical cancer" or "Cervical tumor" or "Cervical tumour" or "Cervical carcinoma" or "Cervical malignancy" or "Breast neoplasm" or "Breast cancer" or "Breast tumor" or "Breast tumour" or "Breast carcinoma" or "Breast malignancy"). After excluding duplicates, titles and abstracts were scanned for potential eligible studies. The full articles of remaining studies were carefully reviewed to determine whether the inclusion criteria were met. In order to make the results more convincing, TCGA datasets were also applied to conduct the meta-analysis. This study was designed, conducted and reported according to PRISMA and MOOSE statements [20,21].
The pertinent studies were selected if they meet the following criteria: (1) Studies described the association between tissue lncRNA expression level and prognosis or clinicopathological features of breast and gynecologic cancers; (2) Clinicopathological features or Hazard ratio (HR) estimates with the corresponding 95% con dence intervals (CIs) for overall survival (OS), disease-free survival (DFS) were available or could be calculated; (3) The number of the studies which reported the association between expression level of a certain lncRNA and prognosis of breast and gynecologic cancer patients must be greater than or equal to 3 (including TCGA datasets). (4) Articles were eligible for evaluation as full English papers. Reviews, letters, meeting abstracts, notes or comments were excluded. The studies concentrating on circulating lncRNAs were also excluded in our studies. Besides, to avoid overlapping data, we excluded the studies reporting the survival outcome with data extracted from TCGA database.

Data extraction and quality assessment
The data from each study were independently evaluated and extracted by 2 investigators (Fan Zhao and Huiqi Chen). We collected the available information from each study as follows: author, year of publication, country of origin, total number of participants, cancer types, date of inception, follow-up period, type of specimens, detection method, HR, and corresponding 95% CI. And we also extracted the data of clinicopathological parameters if available. HRs from multivariable analysis were rst considered in our study due to their adjustment for confounding factors. If a study reported only Kaplan-Meier curves, the survival data were extracted with Engauge Digitizer version 10.6.
The quality assessment of each study was conducted according to the Reporting Recommendations for Tumor Marker Prognostic Studies (REMARK) guideline [22]. The total score ranges from 0 to 20, and a higher score represents higher quality.

Data synthesis and statistical analysis
The primary meta-analyses were conducted to identify speci c lncRNAs signi cantly associated with survival of patients with breast and gynecologic cancers. Chi-square and I 2 test were applied to assess the heterogeneity among studies. P≤0.10 and/or I 2 >50% suggests signi cant heterogeneity [23]. Combined HRs or ORs and 95% CIs were calculated using the DerSimonian-Laird random-effects methods [24]. The signi cance of the pooled HRs or ORs were determined by Z test (p<0.05 was considered signi cant).
Subgroup analyses were adopted to evaluate potential modifying effect of variables and explore the source of heterogeneity in term of ethnic populations and cancer types. Funnel plots were constructed to assess the potential publication bias. All analyses were conducted using Stata software (version 12.0; StatCorp, College Station, TX, USA).

Study characteristics and data quality
After searching PubMed, EMBASE and Cochrane library, a total of 3748 articles were retrieved. And 2632 articles were assessed after removing 839 duplicated records. Then, titles and abstracts were scanned, after which, 277 remaining articles entered the process of full-text reading. 165 articles were excluded for following reasons: insu cient data (n=32), foreign language (n=4), review article (n=18), meeting abstract (n=5), limited number of studies which focused on a certain lncRNA (n=108). After the above screening process, we identi ed 110 studies with 25 speci c lncRNAs which might play an important part in the development of breast and gynecologic cancers. Afterwards, we extracted data of breast and gynecologic cancers form TCGA datasets according to the type of lncRNAs. Association between speci c tissue lncRNA expression and prognostic outcome in breast and gynecologic cancers Based on literature screening, we identi ed 25 speci c lncRNAs which might be crucial in the development of breast and gynecologic cancers. And we conducted meta-analyses which covered nearly 20000 patients to systematically evaluate the association between speci c lncRNA expression and prognostic outcome of breast and gynecologic cancers, and the results were shown in Fig. 1a and b and Table 1 and   Supplementary Table S3 in Additional le 3. Our analyses indicated that 3 lncRNAs (HOTAIR, GAS5, PVT1) might act as promising prognosis biomarkers for breast and gynecologic cancers, because they signi cantly associated with both OS and DFS for breast and gynecologic cancer patients.
Interestingly, 2983 patients were included to evaluate the association between tissue GAS5 expression and prognostic outcome of breast and gynecologic cancer patients, and the pooled HR was 0.51 (95%CI: 0.34-0.77) for OS and 0.40 (95%CI: 0.25-0.63) for DFS, indicating that higher tissue GAS5 expression level predicted better prognostic outcome in breast and gynecologic cancers ( Fig. S4a and S4b).
In addition, there was a signi cant association between tissue MALAT1 (pooled HR=1.51, 95%CI: 1.09-2.08), NEAT1 (pooled HR=1.80, 95%CI: 1.25-2.58), CCAT2 (pooled HR=1.53, 95%CI: 1.09-2.14) expression and OS rather than DFS in breast and gynecologic cancers, as shown in Table 1 Of note, no signi cant association was found between tissue ANRIL, CCAT1, FEZF1-AS1 and HOTTIP expression and prognostic outcome of breast and gynecologic cancers, indicating that they might not be effective biomarkers for breast and gynecologic cancers in clinical applications (Table1). And another included lncRNAs whose study number was less than 3 were shown in Additional le 1.

Subgroup analyses for the association between speci c lncRNA expression and OS in breast and gynecologic cancers
In order to evaluate potential modifying effect of variables and explore the source of heterogeneity, subgroup analyses were conducted according to population and cancer types, which were shown in Fig.   1a, Table 2  indicating it might act as potential prognostic biomarkers for breast cancer with protective effects. And the subgroup analyses of another included lncRNAs were shown in Supplementary Table S4 in Additional le 3.
In addition, subgroup analyses suggested that heterogeneity was mainly in uenced by different populations.
Association between speci c lncRNA expression and clinicopathological features in breast and gynecologic cancers As shown in Table 3, we also explored the association between speci c lncRNA expression and clinicopathological features in breast and gynecologic cancers, such as age, FIGO stage, histology grade, lymph node metastasis (LNM), lymphovascular space invasion (LVSI) and tumor size, which might provide clues why some lncRNAs associated with survival outcome of breast and gynecologic cancers.

Publication bias
Funnel plots were shown in Fig. S5. No publication bias was found for the included studies except for the assessment of association between tissue MALAT1 expression and DFS in breast and gynecologic cancers. Nevertheless, it was still di cult to con rm whether the publication bias really existed due to the limited number of studies.

Discussion
Recently, accumulating evidence has demonstrated that lncRNAs performed their vital function in cancer progression and development [25]. Indeed, it was well established that lncRNA expression pro le was closely related to clinically relevant cancer subtypes, suggesting its potential ability to predict tumor behavior and disease prognosis [26]. Besides, several cancer-associated lncRNAs were proved to regulate cancer invasion and metastases [27]. However, it was still controversial whether some speci c lncRNAs could act as prognostic biomarkers in breast and gynecologic cancers.
In our present study, based on 111 studies and nearly 20000 participants, we performed a meta-analysis to systematically assess the role of some speci c lncRNAs in the progression of breast and gynecologic cancers and the feasibility of clinical applications of lncRNA pro le. According to our study, we identi Furthermore, our studies proposed the potential value for SPRY4-IT1, HOXA11-AS, SNHG15, TP73-AS1 to predict poorer survival outcome, but the reliability of the conclusion was limited due to the numbers of included studies. For example, the exact mechanism for SPRY4-IT1 and HOXA11-AS to play oncogenic roles was still unclear. SPRY4-IT1 was demonstrated to be upregulated in breast cancer tissues and suppress proliferation and increase apoptosis of breast cancer cell through targeting ZNF703 [35]. SPRY4-IT1 knockdown might inhibit the proliferation and arrest cell cycle at G0/G1 stage in ovarian cancer cells [36]. And it was shown that HOXA11-AS could increase cell proliferation, invasion and metastasis of breast cancer in vivo and in vitro experiments through regulating EMT program [37]. In general, more studies of high quality were needed to con rm their prognostic value.
Unlike the lncRNAs mentioned above, our study showed higher tissue GAS5 expression might predict better prognosis in breast and gynecologic cancers. Overexpression of GAS5 might enhance the sensitivity of cervical cells to cisplatin through miR-21 by regulating the level of PTEN and the phosphorylation of Akt, thus inhibiting cancer cell migration and invasion [38]. Another study demonstrated that GAS5 could act as a ceRNA for miR-196a-5p, which regulated FOXO1 expression and downstream PI3K/Akt phosphorylation and then promoted the progression of triple-negative breast cancer [39]. Consistently, our study noted that higher tissue GAS5 expression inversely associated with lymph node metastasis in breast and gynecologic cancers, which partly explained its capacity as a prognostic biomarker.
Our studies had some strengths. To our knowledge, it was the rst study to systematically evaluate the prognostic value of lncRNA pro le in breast and gynecologic cancers. Of note, the study was performed through literature search and data extracted from TCGA database, which contributed to the reliability of the results. In addition, we offered some information about the association between speci c lncRNA expression and clinicopathological features in breast and gynecologic cancers, which explain partly the prognostic value of some speci c lncRNAs. Finally, the methods of this studies were rigorous and in accordance with guidelines for conducting a meta-analysis.
However, there were also some limitations in the current study. Firstly, though subgroup analyses were conducted, the heterogeneity of the studies for some lncRNA analyses could not be fully explained.
Besides, due to the limited number of studies, we could not conduct subgroup analyses for some lncRNAs, thus it was di cult for us to explore the potential modifying effect of ethnic populations and cancer types for some lncRNAs, which increased restrictions on our conclusions. What's more, the criterion of high expression of lncRNAs was not consistent among the included studies. And limited data made it impossible for us to analyze the impact of co-expression of several lncRNAs on the prognostic outcome in breast and gynecologic cancers. Therefore, further high-quality and well-designed studies are warranted to con rm our current ndings.

Conclusions
The current study demonstrated that tissue lncRNA expression might associated with survival outcome of breast and gynecologic cancers patients and speci c lncRNAs could act as prognostic biomarkers of breast and gynecologic cancers, especially for HOTAIR, PVT1 as poorer prognostic biomarkers and GAS5 as a positive prognostic biomarker.  Tables Table 1 Meta-analysis of the association between tissue lncRNA expression and survival outcome in breast and gynecologic cancers.   Table 3 The association between tissue lncRNA expression and clinicopathological features in breast and gynecologic cancers.

Supplemental
Supplementary tables Table S1. Characteristics of the included studies. Table S2. Quality assessment performed according to the REMARK guideline. Table S3. Meta-analysis of the association between tissue lncRNA expression and survival outcome in breast and gynecologic cancers (The number of included studies was less than or equal to 3). Table S4. Subgroup analyses for the association between tissue lncRNA expression and OS in breast and gynecologic cancers (The number of included studies was less than or equal to 3). Table S5. The association between tissue lncRNA expression and clinicopathological features in breast and gynecologic cancers (The number of included studies was less than or equal to 3).
Supplementary gure legends