Background

Circular RNAs (circRNAs), consisting of a circular configuration through a typical 5′ to 3′-phosphodiester bonds, are a novel class of endogenous noncoding RNAs [1,2,3]. CircRNAs play a special role as molecular markers in many human diseases including tumors, due to their conservation, abundance and tissue specificity [4]. In addition, circRNAs can be classified into four categories: exon circRNAs, intron circRNAs, exon-intron circRNAs, and intergenic circRNAs [5]. Different types of circRNAs have distinct functions, including interacting with RNA binding proteins, regulating the stability of the mRNAs, regulating gene transcription, sponging microRNAs and participating in translation [5,6,7]. However, the underlying mechanisms and functions of circRNAs remain uncertain.

Extensive studies have indicated that circRNAs play a major role in tumorigenesis, the development of cardiovascular diseases, and the pathogenesis of neurodegenerative diseases [8]. However, the differential expression of circRNAs and their definite functions are still not totally clear in colorectal cancer (CRC). Colorectal cancer is among the most common malignancies of the digestive system and the fourth leading cause of cancer-related death worldwide [9]. Although considerable progress has been made in the diagnosis and treatment of this disease, the prognosis of CRC patients is still poor, due to the delay in early diagnosis and the high frequency of metastasis and recurrence [10]. In this study, we performed a meta-analysis and a comprehensive search of all relevant literature to summarize the diagnostic, prognostic, and clinical significance of circRNAs in CRC.

Methods

Data search strategy

The PubMed, Cochrane Library, and Web of Science online databases were searched for studies on circRNA research that were published in English before May 15, 2019. The following search strategy was applied: (1) “circRNA” or “circular RNA” and (2) “colorectal cancer” or “colorectal carcinoma” or “colorectal tumour” or “CRC”. Two researchers (JPY and DMG) assessed the title, abstract and full text to identify the appropriate articles. Other researchers (XXL), together with two researchers (JPY and DMG) were involved in the data extraction. Any disagreements were settled by a third researcher (JTC). Then, the data were extracted from the selected articles and populated it into a table.

Inclusion and exclusion criteria

This study used the following criteria when selecting articles. Studies that met the following inclusion criteria were included in the meta-analysis: (1) patients with a pathological diagnosis of CRC; (2) cohort study or case-control study; and (3) studies that detected the circRNA expression level and provided information on the clinicopathological features and prognosis of patients. Studies were excluded if the following excluded criteria were met: (1) studies irrelevant to CRC or circRNAs; (2) data similar to that in prior studies; (3) case reports, letters, animal experiments, reviews, conference reports and meta-analysis; and (4) insufficient data.

Data extraction and quality assessment

All relevant studies were independently screened by two researchers (JPY and DMG) and the following data were extracted from eligible studies: (1) first author, publication year, type of cancer and circRNA, sample size and detection method of circRNA; (2) the role of circRNAs, follow-up time; (3) diagnostic sensitivity and specificity of circRNAs; and (4) clinicopathological features with age, gender, tumour size, tumor location, differentiation grade, TNM stage, T classification, lymph node metastasis, distal metastasis [11]. The Newcastle-Ottawa Scale (NOS) [12] was adopted for the quality assessment of the studies by two independent researchers (JPY and DMG). A third investigator (XXL) discussed any differences. A study with a score ≥ 7 was considered of high quality.

Statistical analysis

Statistical analysis was conducted using STATA software (version 14). Pooled ORs and 95% CIs were used to explore the association between circRNAs expression and clinicopathological features. HRs and 95% CIs were used to assess the prognostic value of circRNAs. The number of true positive (TP), false positive (FP), false negative (FN) and true negative (TN) were calculated and finally the pooled sensitivity, specificity and AUC were obtained to assess the diagnostic value of circRNAs. The chi-square test were used to evaluate heterogeneity. When the I2 value was < 50%, no observable heterogeneity was suggested and a fixed effects model was used [13]; otherwise, a random effects model was utilized. Sensitivity analysis was performed to explore the source of heterogeneity. Qualitative analysis of publication bias was conducted using funnel plots and quantitative analysis was conducted using Begg and Egger’s tests.

Results

Search results

As shown in Fig. 1, 83 relevant studies were obtained from several databases. After abstract reviews, 46 studies were obtained for further full-text reviews. Then, 27 articles were excluded for the following reasons: 5 were not about circRNAs or CRC, 10 did not report relevant results, 3 were review articles, 1 was animal data, and 8 had insufficient data. In summary, there were 19 studies [14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32] included in this study, with a total of 1307 patients, including 11 on clinicopathological features, 8 on prognosis and 7 on diagnosis.

Fig. 1
figure 1

Flowchart of trial selection

Full size image

Study characteristics

The basic information of studies are showed in Table 1 and Table 2. All studies were published between 2015 and 2019. The follow-up time of patients ranged from 57 months to 123 months and the number of samples ranged from 40 to 204. As shown in Tables 1, 6 circRNAs were identified as tumour promoters, and 2 circRNAs were identified as tumour suppressors. As shown in Tables 2, 7 articles with AUC, sensitivity and specificity were included for the diagnosis analysis. The included studies were of high quality (See Supplementary Table 1, Additional File 1).

Table 1 Basic features of studies for prognosis analysis
Full size table
Table 2 Basic features of studies for diagnosis analysis
Full size table

Clinicopathological parameters

The associations between circRNAs and the clinical parameters are shown in Table 3. Up-regulation of oncogenic circRNAs was closely associated with unfavorable clinical features (tumor size: OR = 1.769, 95% CI: 1.097–2.852; differentiation grade: OR = 1.743, 95% CI: 1.032–2.946; TNM stage: OR = 3.320, 95% CI: 1.529–7.207; T classification: OR = 3.410, 95% CI: 2.088–5.567; lymph node metastasis: OR = 3.357, 95% CI: 2.160–5.215; distal metastasis: OR = 4.338, 95% CI: 2.503–7.520). Additionally, down-regulation of tumor-suppressor circRNAs was closely associated with favorable clinical parameters (differentiation grade: OR = 0.453, 95% CI: 0.261–0.787; T classification: OR = 0.553, 95% CI: 0.328–0.934; distal metastasis: OR = 0.196, 95% CI: 0.077–0.498). However, there was no difference between oncogenic circRNAs expression and other clinical parameters such as age, gender, and tumor location.

Table 3 Clinical Parameters of circRNAs in CRC
Full size table

Overall survival

Up-regulation of oncogenic circRNAs was notably associated with worse prognosis (HR = 2.29, 95% Cl: 1.50–3.52, p < 0.001, Fig. 2 a), and a fixed-effects model was utilized as no heterogeneity was found (I2 = 0.0%, p = 0.937). In addition, down-regulation of tumour-suppressor circRNAs was associated with better prognosis (HR = 0.37, 95% Cl: 0.22–0.64, p < 0.001, Fig. 2 b), and a fixed-effects model was applied because of no heterogeneity between studies (I2 = 0.0%, p = 0.525).

Fig. 2
figure 2

Forest plots for the association between circRNAs and overall survival (OS) in colorectal cancer (CRC). a oncogenic circRNAs; b tumor suppressor circRNAs

Full size image

Diagnosis analysis

To further evaluate the diagnostic value of circRNAs, the pooled sensitivity and specificity were calculated, and the results were shown in Fig. 3. And a random-effects model was utilized because of high heterogeneity between studies (I2 = 76.15% and I2 = 48.29%). The pooled results showed a sensitivity of 0.83 (95% CI: 0.75–0.88) and a specificity of 0.72 (95% CI: 0.66–0.78). In addition, the summary receiver operator characteristic (SROC) curve analysis indicated AUC of 0.82 (95% CI 0.78–0.85, Fig. 4). Taken together, these results suggested that circRNAs have a good diagnostic accuracy for CRC.

Fig. 3
figure 3

Forest plots for the pooled sensitivity and specificity of circRNAs

Full size image
Fig. 4
figure 4

SROC curve in the diagnostic analysis

Full size image

Publication bias and sensitivity analysis

No evidence of publication bias were identified from the funnel plot by qualitative analysis (See Supplementary Fig. 1, Additional File 2). In quantitative analysis, there was no obvious publication bias by Begg’s (p = 0.213, See Supplementary Fig. 2, Additional File 2) and Egger’s test (p = 0.722, See Supplementary Fig. 3, Additional File 2). Furthermore, Deek’s funnel plot asymmetry test [33] was performed to assess the publication bias among studies for diagnosis analysis, and the result showed no obvious publication bias was found (p = 0.07, See Supplementary Fig. 4, Additional File 2). Sensitivity analysis indicated the pooled results were stable in our studies (See Supplementary Fig. 5, Additional File 2).

Discussion

Recently, many studies have focused on the significant role of circRNAs, whereas no relevant meta-analyses on circRNA expression in CRC have been performed. A total of 1307 cancer patients from 19 eligible studies were collected and analyzed in this study, including 7 on diagnosis, 8 on prognosis, and 11 on clinicopathological features. For diagnostic value, the summarized results revealed AUC of 0.82, with a sensitivity of 83% and a specificity of 72%. For clinical and prognostic value, abnormal expression of circRNAs were closely associated with clinical parameters and prognosis.

Our current study observed a significant relationship between abnormal circRNA expression and its diagnostic value in CRC patients. As aberrant expression of circRNAs in different tumor tissue can be easily detected, measurements can be performed conveniently and economically. Coupled with the structural stability of circRNAs, circRNAs are considered as potential biomarkers for the diagnosis of CRC patients. Although sensitivity analysis showed no significant heterogeneity, more pertinent investigations are warranted to corroborate our findings.

In previous meta-analyses, only five meta-analyses [34,35,36,37,38] detected an association between the circRNAs and carcinoma. However, in the studies of Wang et al. [34], Chen et al. [35] and Li et al. [36], only one study was included to investigate the relationship between the circRNAs and CRC. Li et al. [37] and Ding et al. [38] assessed the diagnostic value of circRNAs for human cancers, in which five articles were included to investigate the diagnostic value of circRNAs in CRC, whereas they failed to discuss the role of circRNAs in CRC patients. In the present study, we collected all the relevant articles published to date and performed a meta-analysis including 19 articles with 1307 CRC patients. Furthermore, we evaluated the prognostic and diagnostic value of circRNA expression in CRC patients. Nonetheless, further large-scale studies are needed to confirm these results.

However, several limitations must be considered when interpreting the conclusions of this meta-analysis. First, since all patients included in the article were from China, this reduced the applicability of the results across different ethnicities and regions. Moreover, there was a limited number of articles for a subgroup analysis. Furthermore, a relatively small number of patients was included in this meta-analysis, so larger-scale studies would be necessary to verify the obtained results. Finally, several studies did not provide HRs with their 95% CIs in the article, so we needed to extract them from the Kaplan-Meier survival curve.

Conclusions

In summary, our study demonstrated a crucial relationship between the aberrant expression of circRNAs and clinicopathological, prognostic, and diagnostic value in CRC patients. Furthermore, circRNAs may be promising biomarkers and treatment targets for colorectal cancer.