Prognostic role of microRNAs in human gastrointestinal cancer: A systematic review and meta-analysis

Background Gastrointestinal cancers (GICs) mainly including esophageal, gastric and colorectal cancer, are the most common cause of cancer-related death and lead into high mortality worldwide. We performed this systematic review and meta-analysis to elucidate relationship between multiple microRNAs (miRs) expression and survival of GIC patients. Methods We searched a wide range of database. Fixed-effects and random-effects models were used to calculate the pooled hazard ratio values of overall survival and disease free survival. In addition, funnel plots were used to qualitatively analyze the publication bias and verified by Begg's test while it seems asymmetry. Results 60 studies involving a total of 6225 patients (1271 with esophageal cancer, 3467 with gastric cancer and 1517 with colorectal cancer) were included in our meta-analysis. The pooled hazard ratio values of overall survival related to different miRs expression in esophageal, gastric, colorectal and gastrointestinal cancer were 2.10 (1.78-2.49), 2.02 (1.83-2.23), 2.54 (2.14-3.02) and 2.15 (1.99-2.31), respectively. We have identified a total of 59 miRs including 23 significantly up-regulated expression miRs (miR-214, miR-17, miR-20a, miR-200c, miR-107, miR-27a, etc.) and 36 significantly down-regulated expression miRs (miR-433, let-7g, miR-125a-5p, miR-760, miR-206, miR-26a, miR-200b, miR-185, etc.) correlated with poor prognosis in GIC patients. Moreover, 35 of them revealed mechanisms. Conclusion Overall, specific miRs are significantly associated with the prognosis of GIC patients and potentially eligible for the prediction of patients survival. It also provides a potential value for clinical decision-making development and may serve as a promising miR-based target therapy waiting for further elucidation.


INTRODUCTION
Gastrointestinal cancers (GICs) mainly including esophageal cancer (EC), gastric cancer (GC) and colorectal cancer (CRC), are the most common cause of cancerrelated death leading into high mortality worldwide, and it is still among the highest threatening risk of public health for past decades [1]. Actually, GIC patients at early stage could be cured successfully by receiving proper treatment (adjuvant chemotherapy or radiotherapy after radical resection) following approximately 90% fiveyear overall survival rate. However, five-year overall survival rate will decline to merely 15% when develop into advanced stage [2,3]. Therefore, early diagnosis and prediction of individual prognosis play pivotal roles in the treatment and recovery of patients. However, there still lack of effective methods to evaluate the prognosis of GIC patients based on clinicopathology. Currently, increasing Review Oncotarget 46612 www.impactjournals.com/oncotarget studies have reported that aberrant expression of specific microRNAs (miRs) as stable molecular biomarkers was associated with the prognosis of GIC patients and related to the targeted therapy, which provides potentially novel prevention strategies and advanced therapies [4][5][6].
Recently, near 8000 human miRs are registered in miRBase (http://www.mirbase.org/), and they regulate approximately 30% of all gene expression [7]. MiR is a short (20-24 nucleotides) class of non-coding RNA that can target 3'-untranslated regions (3'-UTRs) of mRNA and regulate its expression by degrading a mRNA or suppressing its translation [8,9]. Additionally, one kind of miR can target several kinds of mRNAs at posttranscriptional level. For example, upregulated miR-377 expression promotes tumor proliferation by targeting P53, PTEN and TIMP1 [10]. Meanwhile, various miRs could target identical gene. Furthermore, miR plays a key role in the proliferation and progression of tumor cells, which not only mediates the cells growth, invision, migration and apoptosis but also induces resistance of anticancer drug [11]. For example, down-regulated miR-23b-3p induces chemo-resistence of gastric cancer cells [12]. In addition, many studies have reported that different miRs can be prognostic biomarkers in a wide range of human cancers (ovarian cancer, breast cancer, esophageal cancer, etc.) [13][14][15][16][17].
At present, accumulative evidences have demonstrated that abnormal expression of miRs as stable molecular biomarkers presented potential huge prognostic values in GIC patients [18][19][20][21][22][23]. However, these monocentric, small sample size studies and various experimental protocols from different research departments limited the ability of evaluating relationship between multiple miRs expression and prognosis of GIC patients. The aim of this paper was to elucidate relationship between multiple miRs expression and prognosis of patients and investigate the possible utility of miRs as prognostic biomarkers in GIC patients. Moreover, further understanding of prognostic value of miRs could help for clinical decision-making and develop miR-based target therapeutic treatments.

Study identification and characteristics
60 studies (12 EC, 35 GC and 13 CRC) involving a total of 6255 patients (1271 with EC, 3467 with GC and 1517 with CRC) were included in our meta-analysis based on selection criteria and specific steps were presented in Figure 1 [1, 2, 10-12, 16, 18-71]. More than   Table 2). Moreover, 35 of them revealed mechanisms (Table 3).

DISCUSSION
Gastrointestinal cancer is still a deadly threat in human health due to tumor metastasis and relapse inducing refractory advanced tumor stage and poor prognosis. Yan et al. [72] have demonstrated that there was 40%-65% recurrence rate due to distant metastases and regional relapse in GC patients. Recently, numerous studies focused on the miRs as prognostic molecular biomarkers in GIC patients for precise prediction. For example, Kang et al. [73] reported that miR-21 can be an independent predictor for tumor relapse in CRC patients, and Xu et al. [74] demonstrated that miR-21 as a promising biomarker can predict the lymph node metastases of tumor in GC patients.
The pooled HR value of OS correlated with different miRs expression in GIC patients was 2.14 (1.98-2.30), which implied specific miRs as independent risks inducing poor prognosis and could be considered as prognostic indicators for clinical decision-making. OS was defined as the time interval between GIC confirmed and end of follow up [75]. Moreover, elevated miR-21 expression promoted the tumor cell growth, invasion and migration, and inhibited its apoptosis by targeted PTEN and TIMP1, which was associated with low overall survival. Therefore, miR-21 as a stable molecular biomarker can be used to predict the prognosis of GIC patients. Additionally, miR-21 can also play a diagnostic role in GIC patients [76]. The pooled HR value of DFS associated with different miRs was 2.12 (1.72-2.61), which demonstrated different miRs leading to poor DFS and can be applied to monitor the therapeutic effects after receiving radical resection or chemotherapy. DFS was described as the time interval from GIC confirmed to relapse or end of follow up [68]. All included miRs were statistically significant associated with poor prognosis in GIC patients. Generally, the expression level of identical miR in GIC patients was consistent. For example, Yang et al. [2] reported that decreased miR-206 expression correlated with worse OS in GC patient and the finding was confirmed by Shi et al. [58]. While there were inversely results from different research institutions for identical miR associated prognosis of GIC patients. For instance, Ayerbes et al. [18] revealed   [70] demonstrated that low expression of miR-200c leaded to worse DFS in GC patients. Usually, evaluating prognosis of patients is inextricably bound to clinical decision-making. And researching signal pathways and target genes of miRs may promote the development of novel drug target therapies. Therefore, we summarized the miRs mechanism research associated with prognosis of GIC patients. We found 35 miRs associated with prognosis of GIC patients had explicit targets and some of them have established animal models but further study on clinical trials is required.
Based on this meta-analysis, we can preliminarily draw the clinical value of multiple miRs correlated with prognosis of GIC patients. (1) Aberrant expression of different miRs was associated with the survival of patients and miR-21 as a stable molecular biomarker can predict the individual prognosis through detecting its expression levels in GIC patients. (2) MiRs can offer more precise information for clinical decision-making comparing with the clinicopathological characteristics (such as tumor grade and size) of GIC patients. (3) Expression levels of specific miRs can be detected in tumor tissues or blood samples, which can be used to monitor the therapeutic effects of GIC patients after receiving chemotherapy treatment. (4) Abnormal miRs expression may provide a clinically valuable application for identifying patients with high risk at early stage avoiding advanced cancer progression. (5) It also provides a potential value for clinical decision-making development and may serve as a promising miR-based target therapy waiting for further elucidation.
However, several limitations deserved focused. First, both detection methods (RT-PCR, ISH and microarray) and cut-off values (mean, median, etc.) were applied to evaluate the different miRs expression that may be the source of heterogeneity due to different algorithms. Second, several sample types (tissue, blood, serum, plasma Oncotarget 46618 www.impactjournals.com/oncotarget and bone marrow) were researched by all included studies can also induce the heterogeneity. Quantifiable miRs can be obtained from tissue samples because of its endogenous expression and mostly used to predict the patient survival after receiving resection treatment. Circulatory miRs as noninvasive biomarkers were more likely to predict the prognosis of GIC patients at unresectable stage and surveille the treatment effects of receiving chemotherapy for long term follow up study when compared with tissue samples. Third, clinicopathology characteristics (American Joint Committee on Cancer stage, AJCC stage) associated with prognosis of GIC patients could be the confounding factors inducing high heterogeneity. Therefore, we merely included studies that were focusing on the full sages (I-IV) rather than one certain stage GIC research. Fourth, we extracted HR and 95% CI values from Kaplan-Meier curve according to Tierney's methodology because there were 21 studies lack of survival data, which may cause potential heterogeneity [77]. Fifth, more than half included studies that did not report the adjusted HR values were prone to high heterogeneity. As for publication bias, failure to publish negative results of articles leading to overestimate the pooled effect value, which have reached a consensus. Besides, language bias was existed because only English publications were enrolled in this study. Thus, we systematically searched a wide range of database and found there was no publication bias in all analysis except miR-21 related meta-analysis. After excluding one study in miR-21 related meta-analysis for sensitivity analysis, the pooled effect value did not substantially change implying high stability.

CONCLUSIONS
Overall, specific miRs are significantly associated with the prognosis of GIC patients and potentially eligible for the prediction of patients survival. It also provides a potential value for clinical decision-making development Oncotarget 46619 www.impactjournals.com/oncotarget and may serve as a promising miR-based target therapy waiting for further elucidation.

Selection criteria
Two reviewers read the studies intensively and evaluated the eligibility of studies independently based on selection criteria involving inclusion criteria: (1) Patients were diagnosed with gastrointestinal cancer by histopathology; (2) MiRs as prognostic markers were used to predict the prognosis for full stage (I-IV) patients. (3) Control group (healthy people or patients without GIC) was contained; (4) The effective outcomes were OS, DFS, HR and 95% CI; (5) Observational studies that we can extract the survival data from the articles or Kaplan-Meier survival curve were included; and exclusion criteria: (1) Non-English and non-human subject studies were excluded; (2) Studies were letters, reviews and reports lack of survival data; (3) Studies focused on genetic alterations about the polymorphisms or modification of miRs. We would get to consensus finally through discussion when disagreements came out.

Data extraction and quality assessment
We collected specific information (the first author, year of publication, nation, number of patients, OS/DFS HR and 95% CI, cut-off value, detection method, sample type and follow up) from each included study. The quality of included studies was assessed according to the checklist of meta-analysis of observational studies in epidemiology (MOOSE) [78]: Explicit definition of study population exposure. Oncotarget 46620 www.impactjournals.com/oncotarget Explicit definition of measurement of miRs expression such as qRT-PCR, ISH and microarray.
Explicit definition of outcomes (OS and DFS). Explicit definition of cut-off value and follow-up. Explicit definition of study design.

Statistical analysis
Analysis was implemented by Review Manager 5.3 (The Nordic Cochrane Centre, The Cochrane Collaboration, London, UK) and Stata 12.0 (Stata Corporation, College Station, Texas, USA) software. We applied both fixed-effects and random-effects models to evaluate the pooled value of HR by calculating Cochran Q test and I 2 Index values. If P >0.10 and I 2 <50% implied that low heterogeneity of pooled HR value is statistically significant difference, fixed-effects model should be used finally. Otherwise, random-effects model would be performed. In addition, forest plots of pooled HR values were presented. Funnel plots were used to qualitatively analyze the publication bias and verified by Begg's test while it seems asymmetry. Moreover, we also conducted sensitivity analysis for this meta-analysis.