Diagnostic and Prognostic Value of microRNA-21 in Colorectal Cancer: An Original Study and Individual Participant Data Meta-analysis Free

Colorectal cancer is one of the most common type of cancers and a leading cause of death worldwide (1). Screening tests for colorectal cancer, including colonoscopy and fecal occult-blood testing, have been frequently used in recent years (2). However, these screening tests are far from adequate because of their invasiveness, high cost, or low sensitivity (3). New biomarkers, especially noninvasive biomarkers are in urgent need for colorectal cancer early detection. In addition, new prognostic biomarkers for colorectal cancer are also in need to improve treatment strategies.

microRNAs (miRNA) are small noncoding RNAs that play important roles in the regulation of cell differentiation, cell-cycle progression, apoptosis, and tumorigenesis (4, 5). They regulate protein-coding mRNAs at the posttranscriptional level by binding to target mRNA, preventing its translation or targeting it for destruction, or by transcriptional gene silencing at the chromatin level (6, 7). About half of the human miRNAs are located at cancer-related regions of the genome, and miRNAs are reported to act as oncogenes or tumor-suppressor genes (8, 9), implying their important roles in the progression and prognosis of cancers, including colorectal cancer (10, 11). miR21 is an oncogenic miRNA that can modulate the expression of cancer-related genes, including PTEN, TPM1, and PDCD (12–14). Interestingly, miR21 expression is elevated during tumor progression, for example, its level is upregulated in colorectal cancer tissues. Furthermore, it can influence patients survival and response to chemotherapy (10, 15, 16). Many studies have evaluated the application of miR21 in the diagnosis and prognosis of colorectal cancer; however, the results were contradictive (10, 17–19). As a consequence, the aim of this study was to comprehensively explore the potential value of miR21 in colorectal cancer diagnosis and prognosis.

We conducted an original study to explore the diagnostic and prognostic value of miR21 expression in colorectal cancer. Briefly, colorectal cancer tissues and corresponding normal tissues were obtained from 79 patients by surgical resection and blood samples were collected in EDTA tubes from 41 patients with colorectal cancer before surgical operation and from 30 healthy controls. Total RNA was extracted from tissue and plasma samples, followed with DNase I digestion to exclude genomic DNA contamination. Mature miR21 and internal control U6 were detected by stem-loop real-time RT-PCR methods. ROC analysis was performed and the AUC value was calculated to determine the diagnostic performance of serum miR21 level in colorectal cancer. Survival analyses were conducted using the Kaplan–Meier method. Univariate and multivariate Cox's proportional hazard regression analyses were applied to estimate HRs of death according to tissue miR21 expression levels. The detailed methods were described in the Supplementary File (Supplementary Methods for Original Study). The diagnostic and prognostic data calculated from the original study were pooled with studies identified from literature search in the meta-analysis process.

This meta-analysis was designed, conducted, and reported according to the PRISMA statement (20). The meta-analyses process, including the individual participant data (IPD) meta-analysis, was carried out in accordance with the Cochrane Handbook for Systematic Reviews of Intervention (21). The review has been registered in an international registry of systematic reviews PROSPERO (CRD42013005119).

Comprehensive literature searches were conducted (to August, 2013) in PUBMED and EMBASE to identify eligible studies. The detailed selection process was presented in Supplementary File (Supplementary Methods for Literature search and Study selection).

Two reviewers (Drs. P. Li and H. Zhang), independently collected data using standardized forms and discrepancies were resolved by a third investigator. The following information from each study was extracted: first author, year of publication, origin of the study population, patient characteristics (age, sex, cancer type, and stage), source of samples, number of participants, miR21 assay method, follow-up time, and variables adjusted for in the analysis. For diagnostic studies, the numbers of true-positive (TP), false-positive (FP), true-negative (TN), and false-negative (FN) results were extracted. For prognostic studies, HR estimates with 95% confidence intervals (CI) for overall survival (OS), disease-free survival (DFS), recurrence-free survival, recurrence-free cancer–specific survival (RF-CSS), or progression-free survival were extracted. If the HRs and their 95% CIs were not provided, the numbers of deaths or recurrences and total samples in each study were extracted to calculate these numbers (22). An IPD meta-analysis approach was conducted to assess the prognostic performance of tissue miR21 expression in patients with colorectal cancer. The principal investigators of relevant studies were contacted and asked to provide the raw data. The quality of the diagnostic study was assessed using the quality assessment of diagnostic accuracy studies 2 (QUADAS-2; ref. 23). For prognostic studies, a modified version of the QUADAS-2 assessment tool (Supplementary Table S1) was applied to assess the risk of bias and the criteria for reporting observational studies proposed in the STROBE statement was used to complete the methodologic evaluation (24).

For the diagnostic meta-analysis, numbers of TP, FP, TN, and FN were analyzed and a bivariate model was constructed to summarize the sensitivity, specificity, positive and negative likelihood ratios (LR), diagnostic odds ratio (DOR), and generate the summary receiver operator characteristic curve (25). Confidence intervals were calculated, assuming asymptotic normality after a log transformation for variance parameters and for LRs and a logit transformation for proportions. The formula for a positive LR is sensitivity/(1-specificity), and the formula for a negative LR is (1-sensitivity)/specificity. The combined LRs provide the DOR (positive LR/negative LR). A clinically useful test was defined as having a positive LR greater than 5.0 and a negative LR less than 0.2 (26). Heterogeneity was assessed using the I² test (27), and the Deek funnel plot method was applied to test publication bias (28).

HR was adopted for prognostic evaluation in the current meta-analysis, because all the included studies used HR to measure the prognostic performance of miR21. Study-specific HR estimates were pooled using a fixed-effects model, if there was no significant heterogeneity. Otherwise, a random-effects model was applied. The extent of heterogeneity across studies was checked using the χ² and I² tests; P ≤ 0.10 and/or I² > 50% indicates significant heterogeneity. Subgroup analyses and sensitivity analysis were performed to dissect the heterogeneity. Begg funnel plots and Egger linear regression test were used to assess publication bias. In the Hong Kong Validation Cohort of Schetter and colleagues (10), tissue miR21 expression values were dichotomized with the highest tertile classified as high and the lower two tertiles defined as low, and this high–low cutoff was applied universally in the IPD analysis. Survival analyses were conducted using the Kaplan–Meier method. Then univariate Cox's proportional hazard regression analyses were applied to estimate HR of death according to tissue miR21 expression levels. And multivariate models were used to adjust potential confounding factors for death, including age, sex, TNM stage, pathologic differentiation, and side of the tumor (left or right).

All analyses were conducted using the Stata software (version 11.0; StataCorp.). A P value of <0.05 was considered statistically significant.

A total of 71 individuals were included (41 patients with colorectal cancer and 30 normal controls) to evaluate the diagnostic value of miR21. Serum miR21 level was suggested to be a potential biomarker in discriminating patients with colorectal cancer from control subjects, with an AUC value of 0.657 (95% CI, 0.530–0.783; Supplementary Fig. S1), a sensitivity of 51.2% and specificity of 79.0%, respectively. To assess the prognostic value of tissue miR21 for patients with colorectal cancer, 79 patients with colorectal cancer were included. The mean follow-up time was 65.9 (95% CI, 61.5–70.3) months, and 22 patients died of colorectal cancer during the follow-up period. As Kaplan–Meier survival analysis indicated, patients with higher levels of miR21 in the tumor tissues had a nonsignificant worse OS (61.5 months vs. 68.4 months; P = 0.280; log-rank test; Supplementary Fig. S2). Furthermore, univariate Cox proportional hazard regression analysis revealed an HR of 1.58 (95% CI, 0.68–3.64; P = 0.285) for tissue miR21 in colorectal cancer prognosis. In the multivariable analysis, which included miR21 level, age, gender, side of the tumor, TNM stage, and differentiation, the HR for tissue miR21 in colorectal cancer prognosis was 1.92 (95% CI, 0.74–4.97; P = 0.177).

Searching PUBMED and EMBASE resulted in the inclusion of 1,307 articles. Finally, a total of 13 articles were identified as eligible studies (10, 16–19, 29–36). With our original study, 14 studies were finally included in the meta-analyses. The selection process was shown in Fig. 1, and the characteristics of the included studies were presented in Supplementary Tables S2 and S3. Among the included articles, 11 articles reported the prognostic value of miR21 (including our study; refs. 10, 16–19, 29–32, 36), whereas 6 examined diagnostic value of miR21 (including our study; refs. 17, 18, 33–35; 3 articles reported both prognostic and diagnostic value).

Six studies with 1,071 patients assessed the diagnostic value of blood miR21 level for colorectal cancer. The included studies were conducted in Europe (n = 1), East Asian (n = 4), and the United States (n = 1). Sample size of each study ranged from 40 to 374. The types of specimen contain serum (n = 3) and plasma (n = 3). All the studies adopted the quantitative reverse transcription PCR (qRT-PCR) method to measure the expression of miR21. The quality assessments were shown in Supplementary Table S4.

The pooled sensitivity and specificity were 0.76 (95% CI, 0.59–0.88) and 0.81 (95% CI, 0.76–0.85), respectively (Fig. 2). And the area under the ROC curve was 0.81 (95% CI, 0.78–0.85; Fig. 3), indicating miR21 has a relatively high diagnostic performance in colorectal cancer. There was significant heterogeneity in sensitivity (Q = 82.58, df = 5, I² = 93.94, P < 0.005) but not in specificity (Q = 6.57, df = 5, I² = 23.86, P = 0.25). Sensitivity analyses indicated that country of origin, type of specimen, sample size, and study quality were not the source of heterogeneity. The pooled positive LR, negative LR, and DOR were 3.80 (95% CI, 3.00–4.81), 0.29 (95% CI, 0.17–0.50), and 13.68 (95% CI, 5.35–35.02), respectively.

A total of 11 studies were included in the prognostic analyses. All the studies were published in English and conducted in Europe (n = 4), East Asian (n = 6), and the United States (n = 2; one study included both American and Chinese population). Each study assessed 46 to 520 patients with colorectal cancer. The types of specimen contain tissue (n = 9), serum (n = 2), and tissue stroma (n = 1; one study evaluated both tissue and serum sample). Nine studies adopted the qRT-PCR method to measure the expression of miR21, whereas two studies applied the in situ hybridization (ISH) assay method. Three studies (including our study) presented IPD (10, 36). The quality assessments were shown in Supplementary Tables S5 and S6.

A total of nine studies (including our study) with 1,598 patients assessed the impact of tissue miR21 expression on colorectal cancer OS. The pooled HR was 1.56 (95% CI, 1.16–2.11) for all the studies, indicating that higher tissue miR21 expression level predicate poorer OS for patients with colorectal cancer (Fig. 4A). Significant heterogeneity across studies was observed (I² = 85.1%, P < 0.001; Fig. 4A). Subgroup and sensitivity analyses suggested that methods to measure miR21 expression, country of origin might contribute to heterogeneity across studies (Table 1).

We then applied a meta-analysis of IPD to further explore the potential value of miR21 in colorectal cancer prognosis. Three studies (including our study) presented IPD data (10, 36), and Schetter and colleagues (10) only provided IPD of the Hong Kong Validation Cohort because the Maryland Test Cohort was not tested by the qRT-PCR method. A total of 236 patients with colorectal cancer were included. There was no significant association between miR21 expression and age, gender, tumor location, colorectal cancer stage, and differentiation (all P > 0.05). Kaplan–Meier survival analysis demonstrated that patients with higher miR21 levels in the tumor tissues had statistically worse OS (84.1 months vs. 110.6 months; P = 0.001; log-rank test; Fig. 5). As univariate analysis indicated, higher tissue miR21 level was associated with shorter OS (HR, 2.06; 95% CI, 1.35–3.15; P = 0.001). In the multivariable analysis, which included miR21 level, age, gender, side of the tumor, TNM stage, and differentiation, higher miR21 level served as an independent prognostic marker for indicating poorer OS in patients with colorectal cancer (HR, 1.69; 95% CI, 1.07–2.67; P = 0.023). Interestingly, the association between tissue miR21 expression and colorectal cancer survival appeared to be confined to male patients (multivariable HR, 2.47; 95% CI, 1.28–4.77; P = 0.007) but not for females (multivariable HR, 1.07; 95% CI, 0.53–2.16; P = 0.845).

Five studies comprising 953 patients evaluated colorectal cancer DFS for miR21. We found a significant association between higher miR21 expression level and poorer colorectal cancer DFS (pooled HR, 1.35; 95% CI, 1.08–1.69; Fig. 4B). There was significant heterogeneity in the analysis (I² = 66.4%, P = 0.018; Fig. 4B). Through subgroup and sensitivity analyses, we found that methods to measure miR21 expression, country of origin, and sample size of the study might be the source of heterogeneity (Table 1).

Two studies explored the performance of circulating miR21 levels in the prognosis of colorectal cancer (17, 18). The HR of these two studies for colorectal cancer OS was 4.12 (95% CI, 1.10–15.4) and 1.58 (95% CI, 0.77–3.21), respectively.

The funnel plots of the diagnostic and prognostic meta-analyses were shown in Supplementary Fig. S3. Funnel plot tests (Begg and Egger tests) indicated no significant publication bias in this study. However, because of the limited number of the included studies, it is still difficult to ascertain whether publication bias exists or not.

Although significant progress has been achieved in the diagnosis and prognosis of colorectal cancer over the years, development of better biomarkers is still important for colorectal cancer early detection and for predicting patients' outcome. The application of miRNAs as biomarkers for cancer diagnosis and prognosis has gained much attention in recent years. miR21 is one of the most studied miRNAs as a potential biomarker of colorectal cancer diagnosis and prognosis. To examine the reported diagnostic and prognostic accuracies and evaluate whether miR21 can be a useful biomarker for colorectal cancer, we performed this systematic review on 14 diagnostic or prognostic studies.

As the present meta-analysis showed, circulating miR21 achieved a pooled sensitivity of 0.76, specificity of 0.81, and AUC of 0.81. These results suggested that measuring blood miR21 level is a promising noninvasive method for colorectal cancer diagnosis. The DOR combines the strengths of both sensitivity and specificity, and was reported to be a useful indicator for evaluation of the diagnostic method (37). The DOR value of miR21 was 13.68, indicating a moderate diagnostic accuracy. However, the positive LR (3.8) and negative LR (0.29) suggested that miR21 may be not adequate enough to discriminate and distinguish patients with colorectal cancer. We found a significant heterogeneity in sensitivity, whereas sensitivity analyses indicated that country of origin, type of specimen, sample size, and study quality were not the source. Different cutoff values of miR21 expression across studies may be one source of heterogeneity.

Measuring circulating miR21 might also be a useful screening method for colorectal advanced adenoma. In a study conducted by Toiyama and colleagues (17), in which 43 patients with Japanese advanced adenoma and 53 control subjects were enrolled, miR21 had a relatively high diagnostic performance for advanced adenoma (AUC value of 0.813, sensitivity of 0.811, and a specificity of 0.767). In another study, which included 50 Chinese advanced adenoma patients and 80 healthy controls, miR21 yielded an AUC value of 0.709 for discriminating advanced adenomas from controls (18). However, it should be noted that these two studies were both conducted in East Asia and the sample size was not large; thus, more studies are warranted to clarify this issue.

The results of the meta-analyses indicated that tissue miR21 expression level was a promising biomarker to predict survival in patients with colorectal cancer. Compared with patients with low tissue miR21 expression level, patients with an increased level of miR21 expression had a 1.56-fold higher risk of poor OS and 1.35-fold higher risk of poor DFS. There was significant heterogeneity in the meta-analyses of the data for OS and DFS. To decipher the reason for the heterogeneity, we applied subgroup and sensitivity analyses and found that methods used to measure miR21 expression, country of origin and sample size of the study could partially explain the heterogeneity. Besides, different cutoff values among the included studies may also be a potential source of heterogeneity. An IPD meta-analysis approach was conducted to further explore the prognostic potential of miR21 in patients with colorectal cancer. As the results demonstrated, higher miR21 level was an independent marker for predicting OS in patients with colorectal cancer. And our results showed that the effect of miR21 in predicting colorectal cancer survival was observed only in male participants, suggesting that gender may modify the observed effect. More well-designed studies with large sample size are warranted to clarify this issue and explore the relevant mechanisms. Besides, whether the prognostic value of other miRNAs is differed by gender may also need further study. Circulating miR21 was also developed as a noninvasive prognostic biomarker for colorectal cancer, and studies indicated that higher circulating miR21 level might be associated with poor OS for colorectal cancer (17, 18).

Though sensitivity and subgroup analyses were applied, heterogeneity in both diagnostic and prognostic meta-analyses was not fully explained. The heterogeneity across studies was probably due to the different methodology of evaluating miR21 expression. Different kinds of samples are used in assessing miR21 expression, including frozen tissues, formalin-fixed paraffin-embedded (FFPE) tissues, serum, and plasma. One previous study found that the results of frozen tissues differed from FFPE tissues in evaluating the prognostic performance of miR21 expression for non–small cell lung cancer, suggesting that type of samples may influence the outcomes (38). Normalization is another problem for quantitative estimation of miRNAs. For the included studies, RNU6B, miR16, and total RNA were used by different studies. However, there is no conclusion about the performance of these normalization controls in the estimation of miRNAs and no optimal approach is generally recommended. To draw a convincing conclusion on the value of miR21 for the diagnosis and prognosis of colorectal cancer, an appropriate and unified method should be established and applied.

It is hypothesized that miRNAs enter the circulation directly secreted by cells, released by cells via exosomes, and via shedding of microvesicles (39). miRNAs have unusually high stability in tissues, and previous study has indicated that tumor-derived miRNAs are also present in plasma and serum in a remarkably stable form (40). Mostly, miRNAs demonstrate the same change in expression in blood (plasma or serum) and tumor tissues of patients with various types of cancer (39, 40). And thus, circulating miRNAs may serve as ideal biomarkers for cancer detection and prognosis. miR21 was reported to be one of the most relevant oncogene-like factors among various miRNA. Colon cancer cell lines with higher miR21 expression levels showed an enhanced ability of motility and invasion (41). Furthermore, suppression of miR21 inhibits cell growth in vitro and inhibits tumor growth in animal models by indirectly downregulating the antiapoptotic factor, B-cell lymphoma 2 (42). Studies in human cell lines further investigated the physiologic targets of miR21, and showed that miR21 could target tumor-suppressor genes, such as phosphatase, tensin homolog (PTEN), tropomyosin 1 (TPM1), and programmed cell death 4 (PDCD4; refs. 13, 43, 44). Besides, miR21 has been reported to play an important role in suppressing proapoptotic genes and modulating the vital components of the Ras/MEK/ERK pathway (45). Therefore, miR21 may be involved in the critical steps in carcinogenesis the genesis and progression of human cancer by promoting tumor growth, proliferation, antiapotosis, and migration (42, 43, 46, 47). It has been further demonstrated that tissue miR21 expression is associated with lymph node positivity and the development of distant metastases for colorectal cancer, and therefore miR21 expression serves as a marker clinicopathologic feature of the disease (15). These findings support a vital role for altered miR21 expression in tumorigenesis.

This systematic review had several important strengths. First, we conducted a relatively thorough systematic search and applied a comprehensive analytic approach to evaluate the diagnostic and prognostic value of miR21 in patients with colorectal cancer. Second, an original study was also conducted to explore the diagnostic and prognostic potential of miR21 in colorectal cancer, and an IPD meta-analysis was then used, which further supported the conclusions of the study. The methods of this study were rigorous and followed the guidelines for conducting and reporting systematic reviews.

There were also some limitations in our analysis. First, most of the diagnostic studies enrolled healthy people as controls and were not blind designed. This may affect the diagnostic performance. Second, there was considerable heterogeneity for both the diagnostic and prognostic meta-analyses. Subgroup and sensitivity analyses were applied, whereas the results could not fully explain the observed heterogeneity. Third, the different chemical assays used in the included studies might result in systematic errors among studies. Finally, only Asians and Caucasians were enrolled in this study, and thus the conclusions should be taken cautiously for other ethnic populations.

Taken together, in this study, it is concluded that circulating miR21 level is a useful biomarker for colorectal cancer detection, and tissue miR21 is a promising marker for colorectal cancer prognosis. Further research is needed to explore the combination of other variables associated with colorectal cancer diagnosis and prognosis, in an effort to develop better diagnostic and prognostic models with higher discriminative capacity.

No potential conflicts of interest were disclosed.

The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the article.

Conception and design: P. Li, D. Xia, J. Yang, Y. Wu

Development of methodology: H. Zhang, M. Pesta

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): H. Zhang, H. Ju, M. Pesta, V. Kulda, M. Cai, Y. Wu

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): W. Jin, C. Liu, H. Wu, J. Xu, Y. Ye, G. Zhang, J. Cai, Y. Wu

Writing, review, and/or revision of the manuscript: H. Zhang, P. Li, Y. Wu

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H. Ju, M. Pesta, V. Kulda, G. Zhang, E. Xu, M. Lai

Study supervision: D. Xia

The authors thank Drs. Curtis C. Harris and Aaron J. Schetter from the Laboratory of Human Carcinogenesis, Center for Cancer Research, National Cancer Institute, for their kind help with providing raw data adopted in the IPD meta-analysis process and revising the article.

The work was funded by the National Natural Science Foundation of China (grant numbers 81302455, 81201557, and 81373036).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.