Prognostic value of microRNAs in gastric cancer: a meta-analysis

Background Previous articles have reported that expression levels of microRNAs (miRNAs) are associated with survival time of patients with gastric cancer (GC). A systematic review and meta-analysis was performed to study the outcome of it. Design Meta-analysis. Methods English studies estimating expression levels of miRNAs with any of survival curves in GC were identified up till March 19, 2017 through performing online searches in PubMed, EMBASE, Web of Science and Cochrane Database of Systematic Reviews by two authors independently. The pooled hazard ratios (HR) with 95% confidence intervals (CI) were used to estimate the correlation between miRNA expression and overall survival (OS). Results Sixty-nine relevant articles about 26 miRNAs with 6148 patients were ultimately included. GC patients with high expression of miR-20b (HR=2.38, 95%CI=1.16-4.87), 21 (HR=1.77, 95%CI=1.01-3.08), 106b (HR=1.84, 95%CI=1.15-2.94), 196a (HR=2.66, 95%CI=1.94-3.63), 196b (HR=1.67, 95%CI=1.38-2.02), 214 (HR=1.84, 95%CI=1.27-2.67) or low expression of miR-125a (HR=2.06, 95%CI=1.26-3.37), 137 (HR=3.21, 95%CI=1.68-6.13), 141 (HR=2.47, 95%CI=1.34-4.56), 145 (HR=1.62, 95%CI=1.07-2.46), 146a (HR=2.60, 95%CI=1.63-4.13), 206 (HR=2.85, 95%CI=1.73-4.70), 218 (HR=2.61, 95%CI=1.74-3.92), 451 (HR=1.73, 95%CI=1.19-2.52), 486-5p (HR=2.45, 95%CI=1.65-3.65), 506 (HR=2.07, 95%CI=1.33-3.23) have significantly poor OS (P<0.05). Conclusions In summary, miR-20b, 21, 106b, 125a, 137, 141, 145, 146a, 196a, 196b, 206, 214, 218, 451, 486-5p and 506 demonstrate significantly prognostic value. Among them, miR-20b, 125a, 137, 141, 146a, 196a, 206, 218, 486-5p and 506 are strong biomarkers of prognosis in GC.


INTRODUCTION
Great quantities of previous articles have reported that expression levels of microRNAs (miRNAs) are associated with survival time of gastric cancer (GC) patients . GC is still the fourth most common cancer all over the world and the second most universal cause of cancer death globally, although there has been a constant descent in morbidity and mortality in the past few decades [168,169]. The early clinical inspection of GC www.impactjournals.com/oncotarget/ Oncotarget, 2017, Vol. 8, (No. 33), pp: 55489-55510

Meta-Analysis
was under 15%, and cases of advanced GC accounted for 85% [170]. At present, the primary treatment choices are surgical intervention, chemotherapy, immunogene therapy, and target therapy. The clinical result of GC mainly depends on the stage of tumor. Unfortunately, GC patients' median survival time is no more than 6-9 months [171]. It is unlimited proliferation of cancer cells and ability of intense invasive and metastasis that mainly causes high malignancy degree and poorer survival time. As a result, a novel diagnostic means and improved prognosis of GC might be created through identification of molecular aberrations, which can predict cancer progression and survival rate.
During the past decade, the associations between non-coding RNAs (ncRNAs) and GC have been widely researched. Generally speaking, ncRNAs have been classified as small ncRNAs, consisting of miRNAs and long non-coding RNAs (lncRNAs).
MiRNAs, a novel class of small (20-24 nucleotides [nt]) non-coding regulatory RNAs, play a significant role in multiple biological processes, such as cell division, differentiation, senescence and apoptosis [172,173]. An increasing number of evidence shows that various miRNAs are unconventionally expressed in diverse types of human cancers, and a few miRNAs have been shown to be related with tumor formation, development, progression, and response to treatment by miRNA expression profiling [174].
Moreover, a series of studies have already demonstrated that lncRNAs also play crucial roles in GC progression. A previous investigation reported that, compared with non-tumor tissues, H19 was one of the  [175]. In addition, Li et al. [176] recognized certain potential lncRNAs that abnormally expressed between GC and normal tissues by screening a cohort of 74 GC patients as well, among which, H19 was chosen as a result of a significant overexpression. Furthermore, expression levels of the lncRNAs H19, ANRIL, GHET1, HOTAIR, GAS5, LET, GAPLINC and FENDRR are also significantly associated with the 5-year survival rate of GC patients [176][177][178][179][180][181][182][183].
In GC research area, quite a number of investigations have demonstrated that miRNAs are associated with survival time of patients . However, the number of patients during the articles mentioned above is generally not big enough. Therefore, a systematic review and meta-analysis was performed for the sake of better understanding accurate prognostic value between expression levels of numerous miRNAs and HR of GC patients.

Study selection
A flow diagram with details of the study selection process was presented in Figure 1.

Study frequency
Frequency of studies estimating prognostic value of miRNAs in GC were shown in Table 1 (highlighted studies were included in the present meta-analysis), including miRNA name, number of studies estimating prognostic value, and reference.

Meta-analysis
A summary of the HR evaluated from the whole combined analysis for all the miRNAs was shown in Table 3.

Publication bias
In order to evaluate publication bias for OS of GC patients with high miR-21 expression, the Begg's funnel plot was used by us ( Figure 2B). And the P value was 0.62, indicating absence of publication bias.

Sensitivity analysis
During the study about OS of GC patients with high miR-21 expression, our sensitivity analysis did not indicate alterations in the results according to the exclusion of any individual study ( Figure 2C), suggesting that no single research significantly influenced the pooled HR and the 95%CI.

Present situation
Increasing evidence has shown that various miRNAs are associated with survival outcome in GC www.impactjournals.com/oncotarget patients . However, inconsistent results have emerged. For example, expression levels of miR-200c are up-regulated in blood [98,99] but down-regulated [66,96] in tissue compared with normal samples. Furthermore, expression levels of miR-214 [1,34,110,111] and miR-451 [4,[141][142][143] are unsteadily expressed (up or down). Surprisingly, there are significant associations between aberrant expression levels of them and OS (P<0.05, Table  3, Figures 7 and 8). Therefore, it is essential to conduct a meta-analysis to better understand associations between expression levels of miRNAs and prognosis of GC patients.

Main findings
We performed the meta-analyses about 26 miRNAs and OS. As the most studied miRNA, GC patients with high miR-21 expression have a significantly poorer OS than those with low miR-21 expression (P<0.05). But there is no significant association between high miR-21 expression and RFS/CSS (P=0.17). According to our reference standard, miR-21 is still considered to be a significantly prognostic biomarker. There are some other miRNAs with significantly prognostic value in GC, including miR-20b, 106b, 125a, 137, 141, 145, 146a, 196a, 196b, 206
In a word, these relationships may be involved in the progression of GC.

Strengths of the meta-analysis
This meta-analysis has several strengths which are as follows: (1) we searched almost all articles with survival outcomes in GC patients with diverse miRNAs. Moreover, the present expression profile of miRNAs was clearly listed in Table 1 in terms of names of miRNAs; (2) articles measuring at least one of survival curves about OS, CSS, DFS, RFS, PFS and MFS were finally included and articles only reporting HR or 95%CI without any of survival curves were excluded by us; (3) miRNAs investigated more than or equal to 3 times were conducted meta-analyses; (4) almost all sample sizes of included studies are more than or equal to 30 (except 1 study [64]), enhancing the power and broadening the applicability of the outcomes to GC patients.

Limitations
However, one should keep in mind the following limitations: (1) 1 miRNA considered as significant biomarker of prognosis contained a high heterogeneity (miR-21); (2) there are many variables among the present meta-analysis, such as different types of samples (tissue, plasma and serum), disease stages, cut-off values and miRNA methods; (3) our meta-analysis only included English articles, which might exclude certain relevant articles with other languages; (4) articles only reporting HR or 95%CI without survival curves were excluded by us, reducing the sample sizes of included articles; (5) as a result of substantial relevant articles and data about GC, we subjectively and selectively included some researches according to the criteria of inclusion and exclusion (Table 5), leading to ignore a few potential miRNAs with prognostic value.

Implications for future clinical and scientific research
It is worth mentioning that this meta-analysis is the first systematic estimation of the relevance between miRNA expression and prognosis of GC patients. There are some implications for future clinical and scientific      research in the present meta-analysis: (1) for clinical doctors and other healthcare providers, combined detection of miRNA expression can greatly enhance the estimation about survival time of GC patients and timely treatment can be offered; (2) for scientific researchers, the present study trend on associations between miRNAs and prognosis of GC patients can be conveniently seen in Table 1. As a result, selectively basic experiments can be performed by them (Table 4); (3) inconsistent outcomes of prognosis about miRNAs may be solved according to the basement of the current meta-analysis.

Search strategy, inclusion criteria and exclusion criteria
The details were presented in Table 5. Two authors (Yue Zhang and Dong-Hui Guan) independently performed this comprehensive online search.

Quality assessment
Yue Zhang and Dong-Hui Guan confirmed all eligible investigations that analyzed the prognostic value of miRNAs in GC, and Yue-Hua Jiang reassessed uncertain data.

Statistical analysis
All analyses were conducted using Stata version 13.0 (StataCorp, College Station, Texas, USA). The relative effect sizes for HR were characterized as moderate (protective [0.51-0.75] or contributory [1.35-1.99]) and large (≤0.50 or≥2). The HR was considered significant at the P<0.05 level if the 95%CI did not include the value 1. If the P values from OS and other survival results about corresponding miRNAs were inconsistent, the HR from OS was considered to the main reference standard. Because different types of samples (tissue, plasma and serum) from GC patients at different disease stages, cut-off values and miRNA methods were used in individual studies, random-effects models (DerSimonian-Laird method) were more appropriate than fixed-models (Mantel-Haenszel method) for most of the analyses. Consequently, the random-effects models were used in the current meta-analysis. Publication bias was estimated using the Begg's funnel plot. A two-tailed P value <0.05 was considered significant. Sensitivity analysis (influence analysis) was carried out to test how powerful the combined effect size was to removal of individual investigations. If the point assessment was out the 95%CI of the pooled effect size after it was removed from the analysis, an individual study was doubted to have excessive influence.