Circular RNAs Serve as Prognostic, Diagnostic and Clinicopathological Markers in Lung Cancer: An Updated Systematic Review and Meta-Analysis

Background: In recent years, the roles of circular RNAs (circRNAs) in the biogenesis and clinical application have gradually garnered much interest in the eld of cancer research. However, high-quality studies evaluating the roles of circRNAs in clinicopathological features and clinical application of lung cancer are still unrevealed. Herein, we aimed to elucidate the functions of circRNAs in the association with clinicopathology, diagnosis and prognosis in lung cancer. Methods: Comprehensive and reasonable search strategies were used in four databases up to Nov. 24 th , 2019. The odds ratio (OR) was used to analyze the risks of circRNAs in lung cancer. Sensitivity (Sen), specicity (Spe) and area under curve (AUC) were used to assess the diagnostic value. What’s more, the hazard ratio (HR) was used for the analysis of survival outcomes. A xed-effect model was rstly used in data analysis. If I 2 >50% in heterogeneity, the random-effect model was chosen. Results: A total of 50 studies with 3815 samples were incorporated into our meta-analysis, in which 40 focused on clinical characteristics, 10 related to diagnosis and 31 were aimed at prognosis. In terms of clinical characteristics, both OR and 95% condence interval (95%CI) were shown that circRNAs were signicantly associated with TNM, tumor differentiation, lymph node metastasis, distal metastasis and tumor size in lung cancer. For diagnosis, both upregulated and downregulated circRNAs distinguished patients with lung cancer from healthy people with pooled Sen, Spe and AUC, of which the values were 0.78, 0.76 and 0.81 respectively. In the section of prognosis, circRNAs can better predict the survival time of patients with lung cancer (overall survival, OS: HR=0.46, 95%Cl: 0.33–0.65 in downregulated circRNAs; HR=2.23, 95%CI: 1.97-2.53 in upregulated circRNAs). Conclusions: The results presented here suggested that circRNAs were the potential cancer biomarkers


Background
According to global cancer statistics in 2018, lung cancer has become the leading cause of cancer morbidity and mortality, with 11.6% of all new diagnosis cancer cases and 18.4% of total cancer death cases [1]. Lung cancer is composed of non-small cell lung cancer (NSCLC) and small cell lung cancer (SCLC), of which NSCLC accounts for nearly 85%. Complete surgical resection is the most effective and available treatment in the early-stage of lung cancer. Notably, 5-year survival rates of lung cancer will decrease by 45% in stage III, compared with stage I of lung cancer [2]. However, due to a lack of accurate and sensitive screening methods to detect early-stage lung cancer, it was estimated that 7 out of 10 patients died after the diagnosis of lung cancer in the United States [3]. Hence, it is urgent to seek and develop brand-new biomarkers to improve the status. Circular RNA (circRNA), a loop structure with a 3'-5' covalently closed link, is a novel molecule for tumorigenesis and tumor development [3]. Since the absence of 5'-cap structure and 3'-poly(A) tail, circRNAs are stable and kept from exonucleases [4]. Characterized by master regulatory function, circRNAs has become the focus of attention in the research of lung cancer. In recent years, researches on circRNAs on oncogenesis, metastasis, diagnosis and prognosis have been extensively performed in various cancer types, especially in lung cancer. Xin Huang et al. pooled a total of 23 studies to assess the relationship between circRNAs and lung cancer in terms of clinicopathological features, prognosis and diagnosis [5]. However, the results showed in their study were lack of convincing due to limited included researches.
Although a growing body of studies suggested that circRNAs play a key role in lung cancer in the elds of diagnosis, prognosis and treatment, a high-quality study with reliable pooled results is still absent. Therefore, our study shed highlight on investigating and assessing the diagnostic and prognostic roles of circRNAs in lung cancer. What's more, the relationship between clinicopathological features in lung cancer and circRNAs was also thoroughly evaluated in this study.

Search strategy
The searching work was performed in four main databases, including PubMed, PubMed Central (PMC), Embase and Web of Science, up to Nov. 24 th , 2019, with a search strategy combining terms: ("circRNA" or "circular RNA") and ("lung cancer" or "lung neoplasm" or "lung carcinoma" or "lung tumor") in Title/Abstract and Subject heading. The references of the nal searched studies were initially processed examination by title and abstract, and then potential eligible studies were further looked for full-text articles when necessary. It is indispensable for two researchers (Liang and Guo) to independently carry out this part.

Data inclusion and exclusion
Studies included in our work must meet the following criteria: Studies focused on the relationship between certain circRNA and lung cancer and data included clinicopathology, diagnosis or prognosis; The total number of the sample (including lung tumor tissue and paired adjacent non-tumorous tissue), Sen, Spe, AUC, HR and follow-up time were available; Case-control or cohort.
Researches on circRNA in all kinds of specimens in patients suffering from lung cancer using qRT-PCR were included.
Exclusion can be performed at the following criteria: Meta-analysis, review, case report or only abstract; The article was not written in English; No available data can be extracted in clinical parameters, diagnosis or prognosis; Animal researches or cancer cell experiments.

Data extraction
The extracted information was obtained in each eligible study by two independent researchers. To rule out discrepancy and obtain consistent data, extracted information was then cross-checked between the two researchers. Disagreement was solved by consulting with a third researcher (Chen). All collected studies were extracted the following data: rst author, publication year, circRNAs, the expression level of circRNA (upregulated or downregulated) and cancer types. For clinicopathological features, the extraction of age, gender, smoking, histological type, tumor size, distal metastasis, lymph node metastasis, differentiation and TNM stage were available. For the part of the diagnosis, the samples of case and control, sample types and diagnostic value, including sensitivity, speci city and AUC were availably collected. As for the assessment of prognosis, case number classi ed by expression level, HR and 95%CI, following-up period, and overall survival (OS) were availably obtained. Notably, some of the eligible articles missed the important values (HR and 95%CI), but these values could be calculated by Engauge Digitizer Software (version 11.1). The quality of the articles was examined by the Newcastle-Ottawa scale (NOS, SupplementaryTable 1). The articles with scores≥7 were identi ed as high quality.

Statically analysis
Stata, V14.0 software was utilized to analyze in our meta-analysis to combine Sen, Spe, AUC, HR and OR value. OR and 95%CI were used to assess clinicopathological features, and sensitivity, speci city and AUC were used to assess the effect of diagnosis. What's more, HR and 95%CI were needed to determine the prediction of prognosis. Q-test and I 2 -test were used to determine the heterogeneity between studies. A xed-effects model was applied for the analysis if I 2 <50%. The random-effects model would be used otherwise. If P<0.1 in the Q-test or I 2 >50%, sensitivity analysis and subgroup analysis would be performed to further analyze the potential source of heterogeneity. Both Begg's test and Egger's test were used to quantitatively con rm whether publication bias existed. P<0.05 in both results indicated that there was no publication bias in the analysis.

Search results
Referencing comprehensive and appropriate search strategies, the ow chart showed the steps related to screening the studies for the meta-analysis ( Figure 1). A total of 479 eligible articles were selected from four main databases up to Nov. 24 th , 2019. In these articles, 429 records were eliminated, including 232 records for duplicates, 161 records excluded after screening for title and abstract, and 36 studies were got rid of our meta-analysis because of the following reasons: no full-text articles, review or meta-analysis, insu cient data and not related to circRNAs or lung cancer. Finally, 50 full-text articles with a total of

Diagnostic analysis of circRNAs
Our meta-analysis presented the pooled results of 10 upregulated and 6 downregulated circRNAs in 10 studies for diagnosis in lung cancer ( Table 2). This study estimated some indexes as follows: sensitivity Fagan's diagram presented that the positive probability of precisely diagnosing lung cancer was 76% at the high expression of oncogenic circRNAs or low expression of anti-tumor circRNAs, while the negative probability was only 23% at the normal expression of upregulated and downregulated circRNAs( Figure  2B). As we can see in Figure. 2C, nearly all circRNAs located in lower right quadrants, which means no exclusion or con rmation value of circRNAs in lung cancer. To nd out the between-study heterogeneity, subgroup analysis and sensitivity analysis were performed to uncover the source of heterogeneity. The subgroup analysis revealed that only the speci city of upregulated circRNAs had remarkable heterogeneity ( Figure 3). The sensitivity analysis illustrated that the stability of results by omitting studies one by one and construed the source of heterogeneity (Supplementary Figure 2). The results of sensitivity analysis revealed that the heterogeneity was mainly derived from the study of Xiaoli Zhu (hsa_circ_0013958 in plasma) and the study of Liang Zong (hsa_circ_102231 in tissue). Although the heterogeneity of sensitivity and especially speci city was overwhelmingly decreased after omitting the above studies one by one, the diagnostic value almost stayed unchanged (Supplementary Figure 3). Furthermore, the funnel plot revealed that publication bias might exist in studies (Supplementary Figure   4) and the results were further quanti cationally con rmed by Begg's test (Z=3.38, P=0.001) and Egger's test (t=3.85, P=0.002).
Prognostic analysis of circRNAs 31 included studies with 26 upregulated circRNAs and 5 downregulated circRNAs were summarized in Table 3. The gure showed that high expression of oncogenic circRNAs signi cantly related to poor

Discussion
The present study mainly focused on investigating the potential clinical role of circRNAs in lung cancer in terms of clinical characteristics, diagnosis and prognosis. Our results could be obtained from a total of 50 articles involving 3815 samples (lung tumor tissues and paired adjacent non-tumor tissues). The previous meta-analysis reported by Xin Huang et al. revealed that the diagnostic value of circRNAs in lung cancer was 0.86 AUC, with 77% sensitivity and 81% speci city. Moreover, they also reported that patients with increased expression of oncogenic circRNAs had poor survival (OS: HR=3.24), while patients with elevated anti-tumor circRNAs had longer survival (OS: HR=0.57) [5]. However, the results of the present work were different from the previous study. For diagnosis, although the sensitivity of 78% had a little increase, both AUC and speci city were decreased by 5%. A conclusion that circRNAs might not accurately con rm or exclude lung cancer could be drawn from Fagan's diagram and likelihood ratio diagram. To sum up, the current circRNAs had limited clinical diagnostic value. It turned out that the combination with two or more circRNAs rather than a single might improve overall diagnostic value.
As for prognosis, increased oncogenic circRNAs had more than a two-fold increased risk with poor survival. At the same time, elevated anti-tumor circRNAs had a two-fold decreased risk with unfavorable survival. Our results showed that both upregulated and downregulated circRNAs could be identi ed as good predictors of prognosis. When it comes to clinical features, something interesting was found in subgroup analysis. Although TNM stages, differentiation of tumor and tumor size were remarkably related to upregulated circRNAs, when compared stage II-IV with stage I in TNM, there was no signi cant relationship between circRNAs (both tumor promoter and suppressor) and TNM stages. These might be explained that circRNAs were not sensitively expressed in the early-stage of lung cancer, including stage I and stage II. On the whole, compared with the study of Xin Huang, we newly discovered there were signi cant associations between downregulated circRNAs and differentiation, lymph node metastasis and tumor size. Moreover, we included patients with distal metastasis in our research and concluded that the expression of upregulated circRNAs was positively correlated with more distal metastasis.
The current study still had some de ciencies. The sample limitation of diagnosis analysis was existing. Only 10 studies with 16 circRNAs were pooled in our study, which would lead to unstable results of the diagnosis. In addition, due to insu cient studies of investigating plasma and serum samples, subgroup analysis could not be performed to analyze the diagnostic value of different samples.
In short, although the investigation of circRNAs has made a great stride in recent years, the diagnostic value of circRNAs is still limited. To summarize, circRNAs were good predictors of prognosis in lung cancer, and effective indicators of many clinical parameters, such as TNM stages, differentiation, lymph node metastasis, tumor size and distal metastasis.

Conclusion
It is possible that the circRNAs can be served as prognostic biomarkers and great predictors of clinical pathology for lung cancer. However, it seems that a single circRNA is not suitable as a good diagnostic biomarker for lung cancer. Therefore, future investigation should be required to develop diagnostic value of the combination of two or more circRNAs.

Declarations
Ethics approval and consent to participate Not applicable.

Consent for publication
Not applicable.

Availability of data and materials
Not applicable.

Competing interests
The authors declare that they have no competing interests.  Note: LSCC, lung squamous cell carcinoma; LAC, lung adenocarcinoma; OR, odds ratio; CI, con dence interval.

Figure 1
The owchart showed the algorithm on how to perform the exclusion and inclusion of the eligible studies in our meta-analysis.

Figure 2
Page 19/20 The clinical application of circRNAs in lung cancer. (A) The ability of circRNAs to differentiate patients with lung cancer from normal people was calculated from the Summary receiver operating characteristic (SROC) curve. (B) Fagan's nomogram was shown to estimate the clinical utility of overall circRNAs to differentiate patients with lung cancer from normal people. (C) Likelihood ratio diagram showed the clinical diagnosis value of circRNAs in lung cancer.   Supplementary Files