This study was carried out on the basis of Preferred Reporting Items for Systematic Reviews and Meta-analysis (PRISMA) (18).

Studies published in English language up to 20 October 2018 reporting on the association between the miRNAs polymorphism and cancer prognosis were identified by entering the following search terms into PubMed and Web of Science: “miRNA/miRNAs”; “polymorphisms/variants/variation/single nucleotide polymorphism/SNP”; and “cancer/carcinoma/tumor/neoplasm” and “prognosis/prognostic/outcome/survival.” Two independent investigators (Hanxi Ding and Qian Xu) performed the literature search. Eligible studies met the following criteria: (1) Concerning the association between miRNAs SNPs and cancer prognosis; (2) Involving prognostic indicators such as OS, DFS or RFS; (3) Including available HR and 95% CI. Articles were excluded based on the followings: (1) Duplicated studies or data; (2) Not relevant with miRNAs SNPs and cancer prognosis; (3) Lacking of available data or figures.

Two investigators (Hanxi Ding and Qian Xu) extracted the data independently and reached consensus regarding all the items. Study descriptions were obtained from each full text including author's name, year of publication, country of the origin, type of cancer, total number of the study population, the polymorphism site, the genotype, hazard ratio (HR), and corresponding 95% confidence interval (CI). In the absence of adequate information for estimation of HR or 95% CI, we have made all efforts to contact the authors to obtain sufficient information or extracted data from the Kaplan-Meier survival curves using a method suggested by Tierney et al. (19).

Two reviewers (Hanxi Ding and Qian Xu) independently evaluated the quality of selected studies according to Newcastle-Ottawa Scale (20). Eight items categorized of three dimensions were assessed, including selection, comparability and exposure were assessed. The quality scores ranged between 0 and 9 stars (Supplementary Table 2).

The association of miRNAs polymorphisms with cancer OS, DFS and RFS was estimated through forest plots. Pooled HR and 95%CI were calculated by fixed-effects model or random-effects model. Pooled HR >1 suggested poor prognosis and was considered statistically significant if the 95%CI did not contain 1(21). In the absence of inter-study heterogeneity for Q-statistic with p > 0.05 and I² < 50%, fixed-effect model was chosen to conserve statistical power, otherwise random-effect model was used (22, 23). Publication bias was assessed by Begger' test and if p > 0.05 was considered to be lack of publication bias (24). Sensitivity analysis was conducted by removing studies one by one. All analyses were performed using Stata software 11.0. All tests were two-sided and the results were considered to be statistically significant when the p < 0.05.

As was shown in the flow diagram, a total of 645 articles were enrolled in this systematic review. After multiple selections, 52 researches for 17 miRNAs SNPs including 24721 patients were involved in our meta-analysis (Figure 1) (4, 9–12, 25–68). Among the enrolled studies, 17 miRNAs were OS-related, 6 were RFS-related and 4 were DFS-related. The cancer type covered hepatocellular cancer (HCC), gastric cancer (GC), colorectal cancer (CRC), breast cancer, bladder cancer, non-small cell lung cancer (NSCLC), head and neck cancer, squamous cell carcinoma of the non-oropharynx (SCCOP). The original population came from Chinese, Korean, American, Canadian, Indian, and Polish. The characteristics of all the 52 studies were summarized in the Table 1 and the original data were shown in the Supplementary Tables 1, 4.

Other than the above-mentioned miRNAs SNPs, meta-analysis was also performed for the association of let-7a-2 (rs629367), miR-218 (rs11134527), miR-26a-1 (rs7372290), miR-27a (rs895819), miR-34b/c (rs4938723), miR-423 (rs6505162), miR-30c (rs928508), and let-7a-1 (rs10739971) with cancer OS and RFS. However, no statistical significance was demonstrated in them (Tables 2, 3).

Meta-analysis of all the included stratified data was conducted and the results indicated that CT+TT vs. CC model of miR-196a (rs11614913) was related to better OS as well as DFS in the never smoking subgroup (HR = 0.48, 95%CI 0.29–0.90, p = 0.005 for OS; HR = 0.54, 95%CI 0.35–0.84, p = 0.007 for DFS; Table 5).

In the overall comparisons, no inter-study heterogeneity was associated with cancer prognosis (OS, RFS, and DFS, Tables 2–4). With respect to the heterogeneity in other overall comparisons, sensitivity analysis was performed subsequently.

Begg's test was used to evaluate the potential publication bias of included studies. No statistically significant publication bias was indicated in any of the models for all involved miRNAs SNPs (Supplementary Table 3).

Let-7i rs10877887 is located in the promoter region of miRNA let-7, which was a well-known tumor suppressor of multiple cancers and incorporates a CpG island, TF biding sites, and DNase peak (72–74). Du et al. suggested miRNA let-7 acted as a tumor suppressor in RCC via down-regulating C-myc and C-myc's target gene (73). It has also been reported that low let-7i expression was independently associated with CRC distant metastasis and significantly linked to poor survival in CRC patients (75). Similar result was found in another research on lung cancer by Huang et al. (76).

On account of the location and incorporation of rs10877887, the T > C variant may influence the binding of transcription factors and let-7 expression (33, 77). Liu's research suggested that patients harboring rs10877887 CC genotype had a lower let-7i expression in CSCC tissues (78). In addition, the T > C variation of rs10877887 may have strong affinity with Myeloid zinc finger 1 (MAF1), a transcription factor which can promote the activity of Axl promoter, resulting in tumor cell migration, invasion and metastasis (33).

Our results suggested that the CT/CC genotype of rs10877887 plays a worse role in cancer prognosis when compared with wild homozygote. Probably it was because that deregulated let-7i expression associated with T > C mutation in cancer tissues. We could speculate that T > C variants may serve as biomarker of poor cancer prognosis as well as assistant clinical-decisions. Much more researches were needed to confirm this result as well as detect the potential mechanisms and functions in the future.

MiRNA 608 harbors one SNP, rs4919510 C > G, located in +22bp of its mature 25 bp sequence. It is at the joint of the stem with the canonical hairpin loop, which has been reported to play a key role in cancer progression and be associated with prognosis of several cancer recently (41, 46, 51, 63–65, 79–81). Zheng et al. have found that rs4919510 SNP might influence the expression of miR-608 target genes including cell growth-related genes, tumor invasion and metastasis-related genes and cancer death-related genes (82).

Among the studies included in our meta-analysis, CG vs. GG, CC vs. CG+GG, and CG+CC vs. GG models of miR-608 were associated with overall cancer prognosis, all having protective effects. The possible mechanism might be that the CG, CC, and CG+CC genotypes of rs4919510 could influence the expression level of miR-608 target genes by regulating the miR-608 expression and exert positive roles in cancer prognosis. The reason why no statistical significant results were found to be related to cancer RFS may result from the relatively small number of articles researching about the miR-608 rs4919510 polymorphisms and RFS. In conclusion, miR-608 rs4919510 is associated with cancer OS and could be used for specific prediction of cancer prognosis and may direct clinical-decisions in the future with abundant mechanisms as well as functions evidences.

As we know, mounting studies have shown that miR-492 plays an important role in cell tumorigenicity of multiple cancers (83–85). von Frowein et al found that up-regulation of miR-492 enhanced proliferation, anchorage-independent growth, migration, and invasion of hepatoblastoma by regulating CD44, which is a receptor for hyaluronan, the major component of the ECM and as well as a co-receptor for multiple cytokine signals and growth factors (83). Shen et al have revealed that ectopic expression of miR-492 contributed to deregulation of SOX7, leading to up-regulation of cyclin D1, c-Myc, and Rb phosphorylation, which could promote cell proliferation and cell cycle of breast cancer (84).

Single nucleotide polymorphisms of miRNAs have been identified to affect miRNA processing and alter miRNA expression (50). Some researchers have shown that miR-492 rs2289030 G > C was associated with various cancers (50, 62, 67, 86). Lee et al found CRC patients carried CG and GG genotype demonstrated worse RFS when compared with CC genotype. However, all enrolled studies in our meta-analysis showed that the GC and GC/GG genotypes played a positive role in cancers OS (87). Although our findings suggested that the GC and GG variation of miR-492 may improve cancers prognosis, more rounded investigations are needed to elucidate the association between miR-492 rs2289030 and cancer prognosis as well as the special mechanisms for the limited studies and inconsistent results.

It has been reported that the A to G base change of SNP rs1076064 at 222 bp upstream from miR-378 may alter the expression of miR-378 (60). Accumulating researches have indicated that miR-378 was down-regulated in CRC, GC, and oral cancer (88–90). MiR-378 may exert tumor suppressor roles by deregulating the expression of CDK6 and VEGF in GC. Wang et al suggested that miR-378 inhibited cell proliferation by targeting CDC40 (88, 90). However, other studies have demonstrated that the up-regulation of miR-378 was related to several kinds of cancer including breast cancer, RCC and AML (91–93). Lee et al found that miR-378 could promote cell survival, tumor growth and angiogenesis by targeting at SuFu and Fus-1 (94).

An's research revealed that the variant genotype of rs1076064 acted as a transcription regulator of miR-378 and the G allele of rs1076064 may exert higher promoter activity in many cancer cell lines and was associated with a better prognosis in HCC (60). The meta-analysis results of the two involved articles also showed that AG vs. AA model was associated with better cancer OS, which was consistent with previous study. However, more studies were needed to confirm this result and translate the mechanisms in the future.

MiR-423, relevant to NSCLC, CRC, and breast cancer, is located in frequently amplified region of chromosome 17q11.2 (95–97). Zhao et al have revealed that miR-423 played a potentially oncogenic role in breast carcinogenesis by promoting cell proliferation of breast cancer cell lies (97). Similarly, Sun et al suggested the overexpression of miR-423 could decrease BRMS1 level obviously and promote cell invasion of HCC (98).

The study of Xing et al demonstrated that the variant-containing genotype AC/AA of rs6505162 in miR-423 was significantly associated with poor OS and RFS of CRC patients (65). However, another research by Lin et al suggested the A allele of miR-423 rs6505162 was associated with decreased RCC recurrence and better prognosis by weakening the capacity to target KLF2 mRNA, leading to inhibition of angiogenic pathways and cancer recurrence (99). The meta-analysis results of rs6505162 in our study showed that AC and AC/AA genotypes were associated with poor cancer RFS, which was consistent with Xing's research. However, only two related studies were enrolled in this meta-analysis, further relevant investigations are needed to obtain more reliable results.

It has been well acknowledged that miRNAs SNPs could influence cancer prognosis by affecting miRNAs maturation or ability to combine with mRNAs target gene (100, 101).

Researches showed the variant G allele of miR-146 rs2910164 G > C, located in the 3′ miRNA passenger strand, may enhance the expression of mature miR-146 and the miR-146 overexpression has been found to suppress breast cancer metastasis (102, 103). This meta-analysis results of miR-146 rs2910164 showed that only the GG vs. GC+CC model was associated with better cancer DFS, which was consistent with Xia's meta-analysis (14). It could be inferred that the variant G allele might lead to increased expression of miR-146a and thus better prognosis.

MiR-196a has been regarded as an oncogene in cancers pathogenesis such as proliferation, migration and invasion (55). MiR-196a2 rs11619413 T > C, located in the 3′ messenger strand of miR-196a2, could influence the combination with target gene and the CC genotype was related to increased expression of miR-196a2 (34, 49, 104). The Meta-analysis showed that TC+TT vs. CC model was associated with better cancer OS while CC vs. CT+TT model was associated with poor cancer DFS, suggesting that CC allele might have risk effects on cancer prognosis. These results were contrary to Wang et al but consistent with Hu's research about lung cancer (11, 34). The differences may result from the different types of cancers or the baseline characteristics and more large number samples researches are needed to clarify the relationship between miR-196a2 rs11614913 and cancer prognosis.

MiR-149 is reported to be a pro-apoptotic miRNA, which can inhibit the expression of Akt1 and E2F1 and thus induce cancer cell lines apoptosis (99, 105). A function study of Xia et al suggested that C allele could increase the expression of miR-149, therefore leading to better prognosis in NSCLC (46). In this meta-analysis, we found four models of miR-149 rs2292832 including CT vs. TT, CC vs. TT CT+CC vs. TT, and CC vs. CT+TT were associated with better cancer OS, which was consistent with previous meta-analysis (106). We could infer that variant C allele may enhance miR-149 expression which was associated with better cancer prognosis and may play role as specific biomarker of cancer prognosis.

MiR-499 rs3736444 is located in the stem region opposite to the mature miR-499 sequence genetic and the T to C variation may influence miR-499 expression (39, 107). Ma et al have reported that the C allele could suppress miR-499 expression, resulting in decreased expression of Ets1, which therefore promotes HCC development and cause poor cancer outcome (108). The present study showed that TT vs. CT+CC model was associated with better cancer OS which was consistent with the results of Qiu's study that rs3746444 C to T variant could contribute to unfavorable cancer prognosis by regulating the expression of cancer-related genes (109). Therefore, we could speculate that variants containing C allele may play the role of miR-499 expression inhibitor and results in poor cancer outcome. Although there were statistically significant relationships, it should be noticed that only two relevant studies were involved, thus more convincing results need more studies on this field.

In addition to the above-mentioned 9 miRNAs polymorphisms, other miRNAs SNPs involved were also reviewed and processed in this meta-analysis. No significant association with cancer prognosis was discovered in our meta-analysis, but some of them were reported to influence the development and invasion of cancers and the corresponding mechanisms have been elaborated. For example, the transition from A to G of rs1113427 may alter the local secondary structure of miR-218 and then influence the expression of miR-218 which could inhibit the invasion and metastasis of GC by targeting the Robol receptor (66, 110, 111). Zhang et al. found rs895819 could affect the secondary structure of pre-miR-27a that subsequently influence the processing and maturation of miR-27a (4). Additionally, only two researches were available in the meta-analysis for multiple SNPs, so more relevant investigations should be included for updating the findings on the association between these polymorphisms and cancer prognosis in the future.

Further, we have collected all the available stratified data of included researches. It is suggested only the CT+TT genotype of miR-196a rs11614913 was associated to better cancer prognosis in the never smoking subgroup. We may speculate that CT +TT genotype of miR-196a rs11614913 maybe a potential biomarker for cancer prognosis in the specific subgroup. More evidences were needed to strengthen this conclusion in the future.

Our study had some advantages. First, we collected all the published articles related to miRNAs SNPs and cancer prognosis, making the systematic review and meta-analysis comprehensive and complete. Second, this is the first meta-analysis concerning the association between miRNAs SNPs and cancer prognosis. Moreover, this study is reliable and stable due to the large number of enrolled patients (24,721) and the strict inclusion criteria.

Undoubtedly, some limitations should be acknowledged in our study. First, only English literature was searched, which may results in publication bias. Second, although the overall sample size were up to 24,721 patients enrolled, the number of some miRNAs SNP relevant studies was too small thus we could only preliminarily analyze the relationships between miRNAs polymorphisms and cancer prognosis with the currently published literatures. What's more, many of the included articles were without clear statistical power thus we could have no idea whether there were type I error and type II error in the original data. Finally, several original texts have no available data and we extracted data from figures which might lead to some bias.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.