Meta-analysis of the association between three microRNA polymorphisms and breast cancer susceptibility

Single nucleotide polymorphisms (SNPs) in three microRNAs (miRNAs), rs2910164 in miR-146a, rs11614913 in miR-196a2, and rs3746444 in miR-499, have been associated with breast cancer (BC) susceptibility, but the evidence is conflicting. To obtain a more robust assessment of the association between these miRNA variants and BC risk, we carried out a meta-analysis through systematic literature retrieval from the PubMed and Embase databases. A total of 9 case-control studies on rs2910164, 12 on rs11614913, and 7 on rs3746444 were included. Pooled odds ratios and 95% confidence intervals were used to evaluate associations with BC risk. Overall analysis showed that rs2910164 was not associated with BC susceptibility in any genetic model, whereas rs11614913 was associated with a decreased risk in both the allelic contrast and recessive models, and rs3746444 imparted an increased risk in all genetic models. Stratified analyses showed that rs11614913 may decrease the risk of BC in the heterozygote model in Asians, and in all genetic models, except the heterozygote model, when the sample size is ≥ 500. Subgroup analysis indicated that rs3746444 was associated with increased risk of BC in Asians, but not Caucasians, at all sample sizes. This meta-analysis suggests that rs11614913 in miR-196a2 may decrease the risk of BC, while rs3746444 in miR-499 may increase it, especially in Asians when the sample size is large. We propose that rs11614913(C > T) and rs3746444 (A > G) may be useful biomarkers predictive of BC risk.


INTRODUCTION
One of the most surprising advances in understanding the mechanisms of gene regulation in health and disease has been the discovery of microRNA (miRNA) [1].

Meta-Analysis
Oncotarget 68810 www.impactjournals.com/oncotarget Breast cancer (BC) is the most common type of cancer affecting women worldwide. Although its etiology is multifactorial, its development and outcome are especially influenced by genetic factors. In this regard, several studies pointed out that alterations in miRNAs may contribute to the pathogenesis of BC [6,7]. Since approximately 50% of miRNA genes are located in cancer-related chromosomal regions [8], in recent years their usefulness as biomarkers to evaluate cancer risk has been the subject of intense research.
Single nucleotide polymorphisms (SNPs) represent the most common form of genetic variation. When present in miRNA genes, SNPs may influence miRNAs' properties by altering their expression, maturation, and/or function [9], and may thus increase the risk of cancer, or influence its progression [10,11].
Among several common miRNA SNPs purportedly related with BC risk, the association of rs2910164 in miR-146a, rs11614913 in miR-196a2, and rs3746444 in miR-499 and BC risk remains inconclusive. For instance, Bansal et al. found that the heterozygous variant of rs2910164 in miR-146a is associated with reduced risk of BC [12], while a separate report indicated associations for rs11614913 in miR-196a2 and rs3746444 in miR-499 [13]. Nevertheless, some studies reported that these polymorphisms were not related to BC risk [14,15]. Therefore, in order to evaluate the association of these three miRNA SNPs and BC susceptibility, we performed this meta-analysis by systematically summarizing published data.

Association between miRNA-146a rs2910164 polymorphism and BC susceptibility
We firstly assessed the association between miRNA-146a rs2910164 polymorphism and BC susceptibility. Significant heterogeneity was identified by Q-test and I 2 statistic under all genetic models except the heterozygote. Therefore, except for the latter, the random-effects model was used for all models. No significant associations were identified for any genetic model ( Figure 3A, Table 2).
Next, subgroup analysis was carried out according to race. No significant association was found in any genetic model for Asians and Caucasians. Subgroup analysis based on the source of controls revealed no significant association between any genetic model and either population-based or hospital-based groups; also, no associations were detected for sample sizes < 500 and ≥ 500 (Table 2).

Association between miRNA-196a2 rs11614913 polymorphism and BC susceptibility
The association between miR-196a2 rs11614913 polymorphism and the risk of BC was tested using the random-effects model, due to the presence of significant heterogeneity, for the allelic contrast model and the homozygote, dominant, and recessive models, while the fixed-effects model was used for the heterozygote model. A significantly decreased risk of BC was observed under the allelic contrast model and the recessive model (T vs. C, OR = 0.90, 95% CI: 0.81-1.00; TT vs. CC + TC, OR = 0.86, 95% CI: 0.74-1.00; Table 3; Figure 3B).
In subgroup analysis by race, a significantly decreased risk of BC was observed for Asians under the heterozygote model (TC vs. CC: OR = 0.85, 95% CI: 0.74-0.99). In Caucasians, in contrast, no association was www.impactjournals.com/oncotarget detected between miR-196a2 rs11614913 and BC risk for any genotype model. Similarly, no relationship was found for any genotype model in the mixed-race subgroup. Results of subgroup analysis based on the source of controls showed no significant association between any genetic model and either population-based or hospitalbased controls. We also found no significant association for sample size < 500 under any genetic model, although a sample size ≥ 500 was associated with decreased BC risk in all models except the heterozygote (T vs.  Table 3).

Association between miRNA-499 rs3746444 polymorphism and BC susceptibility
The association between miR-499 rs3746444 polymorphism and the risk of BC was examined by applying the fixed-effects model to assess all genetic models. A significantly increased risk of BC was observed for all genetic models (G vs Figure 3C).
Subgroup analysis was performed for the Asian population, where a positive association was  Table 4).

Publication bias
We utilized Begg's funnel plot and Egger's test to evaluate publication bias. No evidence of publication bias was found for the association between miR-146a rs2910164, miR-196a2 rs11614913, and miR-499 rs3746444 polymorphisms and BC susceptibility using

Sensitivity analysis
To examine the influence exerted by individual studies on the pooled ORs, sensitivity analysis in the allelic contrast model was performed by successively deleting each participant study. We confirmed that the omission of any single study did not significantly affect the overall results ( Figure 5).

DISCUSSION
MiRNAs mediate degradation or translational repression by binding to the 3′UTR and 5′UTR of the target mRNA [2]. SNPs, the most common source of genetic sequence variation, can affect the function of miRNAs by altering primary transcript formation, pre-miRNA maturation, or miRNA-mRNA interactions [11,37]. Even minor variations in miRNAs could have an enormous effect on the expression of different target genes and thus lead to susceptibility to several diseases including BC [38]. Murria Estal et al. identified miRNA profiles related to breast cancer features like node involvement, histological grade, ER, PR, and HER2 expression [39]. SNPs in miR-146a (rs2910164), miR-196a2 (rs11614913), and miR-499 (rs3746444) have been suggested to be predictive biomarkers for patients with BC [12, 14, 15, 26-28, 30, 31, 35, 40], although the studies in question provided inconsistent results. This lack of consensus prompted us to perform a comprehensive meta-analysis on the association of these three miRNAs polymorphisms and BC risk.
The miR-146a human gene is located on chromosome 5 at locus 5q34 and has been linked with BRCA1/BRCA2 activity. The SNP rs2910164 is located in the middle of the miRNA stem hairpin and leads to a change from a      Oncotarget 68820 www.impactjournals.com/oncotarget G:U pair to a C:U mismatch in the stem structure of the precursor molecule, altering the expression of mature miR-146a [41]. This SNP has been associated with the risk of various cancers, among them hepatocellular and bladder carcinomas [41,42], evidencing also cancer-specific and ethnicity-dependent effects [43,44]. Our analysis of miR-146a rs2910164, which included 4,441 cases and 3,899 controls from nine studies, revealed however no association with BC in both overall comparison and subgroup analysis by race, source of control, and sample size. In accordance with this conclusion, studies by Catucci et al. [14] and Alshatwi et al. [24] also failed to demonstrate a link between rs2910164 and BC risk.
The rs11614913 polymorphism in miR-196a2, located in the mature sequence of miR-196a-3P, may affect pre-miRNA maturation and confer diminished capacity to regulate target genes [11,27]. Epidemiology studies have also revealed an association between rs11614913 and risk for multiple cancers; however, results were conflicting [42,45]. Similarly, Linhares et al. [32] found that individuals carrying the CC genotype of rs11614913 in miR-196a2 had decreased BC risk, whereas Gao et al. [46] found instead a positive association between this genotype and BC risk. On the other hand, Dai et al. concluded that rs11614913 may reduce the risk of BC under the recessive model [47]. However, our meta-analysis involving twelve studies with 5,792 cases and 7,159 controls demonstrated an association of rs11614913 with decreased risk of BC both in the allelic contrast model and the recessive model. In subgroup analysis, a significant association was observed between rs11614913 and reduced risk of BC for the heterozygote model in Asians, and for all, except the heterozygote, genetic models when sample size ≥ 500. The discrepancies may derive from different sample sizes, races, and genetic backgrounds of the studies' groups.
Rs3746444, located at the 3p region of mature miR-499, involves a A:U to G:U mismatch in the stem structure of the precursor molecule, leading to altered processing and expression of the mature transcript [48]. The presence of this mismatch would affect Sox6 and Rod1 genes, which are important in the etiology of several cancers [49]. A number of studies investigating the association between rs3746444 and cancer risk have found that this SNP has distinct effects on different populations and cancer types. Dai et al. found that rs3746444 may be related to increased risk of BC under the allelic contrast, homozygote, and recessive models [47]. Our meta-analysis, assessing seven studies with 4,019 cases and 4,683 controls, showed that carriers of the rs3746444 GG genotype and GG + GA genotypes are at a significantly increased risk of developing BC when compared with those carrying the AA genotype. Also, Asians and hospital-based control subgroups demonstrated a significant association with increased risk of BC under all genetic models, but no significant association was found for Caucasians and for population-based source of control under any model. Thus, our results suggest oncogenic mechanisms are distinctly influenced by specific genetic backgrounds across populations.
Although the studies included in our meta-analysis differed from one another in numerous aspects, sensitivity analysis of miR-146a rs2710164, miR-196a2 rs11614913, and miR-499 rs3746444 indicated that the associations detected were not driven by any single one. Moreover, no publication bias was identified with either Begg's funnel plot or Egger's regression test. Finally, no limitations were imposed on our literature search, thus selection bias was well controlled.
Nevertheless, some limitations in this meta-analysis are noteworthy. Firstly, the number of included studies for the miR-499 rs3746444 polymorphism was limited. Secondly, there exists a certain degree of heterogeneity in some genetic models of rs2710164, rs11614913 and rs3746444. After subgroup analysis stratified by race, it could be established that heterogeneity was significantly reduced for Asians in some genetic models of rs11614913 and in all the genetic models of rs3746444. Thus, it could be assumed that the observed heterogeneity resulted, at least in part, from racial differences, which may have impacted the results of our study.
In conclusion, our results indicated that the rs2910164 (G > C) polymorphism in miR-146a may not be associated with susceptibility to BC; the rs11614913 (C > T) polymorphism in miR-196a2 is significantly associated with decreased BC risk; and the rs3746444 (A > G) polymorphism in miR-499 is associated with increased BC risk, especially in Asians. Thus, rs11614913(C > T) and rs3746444 (A > G) appear to be both promising biomarkers to forecast BC risk and potential therapeutic targets. However, owing to the limitations mentioned above, these results should be treated with caution. To further verify and confirm these findings, well-designed, large scale case-control studies will be required.

Search strategy and selection criteria
This meta-analysis was performed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement [50]. To identify all published studies addressing the relationship between miRNA polymorphisms and BC risk, PubMed and Embase databases (last updated on July 20, 2016) were searched without publication type or date restrictions using the following keywords: breast cancer/carcinoma, miR-146a/ rs2910164, miR-196a2/rs11614913, miR-499/rs3746444, and polymorphism/SNP/variation. The literature search was limited to English articles. We selected all potentially eligible studies for review. www.impactjournals.com/oncotarget

Study selection and data extraction
All the included studies were selected following the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [51]. Eligible studies met the following criteria: (1) assessed the relationship between miR-146a/-196a2/-499 polymorphisms and BC risk; (2) had a case-control design; (3) addressed histologically confirmed BC; (4) had sufficient genotype data for further calculating odds ratios (ORs) and their 95% confidence intervals (95% CIs); (5) met Hardy-Weinberg equilibrium (HWE) in the control group (P > 0.05). Exclusion criteria included: (1) duplications of previous publications; (2) comments, meeting reports, reviews or editorials; (3) family-based studies of pedigrees; (4) studies with no detailed genotype data. When there were multiple publications from the same population, only the largest study was included. Study selection was done by two investigators independently, by screening the title, abstract and full-text. Any dispute was settled by discussion.
Data from eligible studies were extracted in duplicate by two investigators independently (Mu and Guo). Extracted data included author, year, country, race, genotyping method, source of control, genetic models of cases and controls, and P value for HWE (Table 1). These two authors checked the extracted data and approved it by consensus. If dissent existed, an additional investigator (Liu) would intervene to settle the disagreement. The quality of selected studies was assessed by two or more investigators independently, according to the Newcastle-Ottawa Scale (NOS) criteria [52]. As per the latter, studies must ascertain or include: cases with independent validation (NOS01); representativeness of the cases (NOS02); selection of controls from community controls (NOS03); controls with no history of disease (endpoint) (NOS04); appropriate study controls for the most important study factor (NOS05); study controls for any additional factor (NOS06); secure record (NOS07); structured interview where interviewer is blind to case/ control status (NOS08); same method of ascertaining exposure for cases and controls (NOS09); same nonresponse rate for both groups (NOS10). The maximum NOS score is 10 points, and studies scoring 6 or higher were included in the meta-analysis.

Statistical analysis
We calculated the P value of HWE in the control group using an online tool (http://ihg.gsf.de/ cgi-bin/hw/ hwa1.pl). The departure from HWE of SNP frequencies in the control group was assessed by X 2 test, and a P value < 0.05 was regarded as significant. Odds ratios (ORs) and 95% confidence intervals (CIs) were obtained to evaluate the strength of the association between miR-146a/-196a2/-499 SNPs and susceptibility to BC. Pooled ORs were determined for the allelic contrast model (miR-146a: C vs G, miR-196a2: T vs C, miR-499: G vs A), homozygote model (miR-146a: CC vs GG, miR-196a2: TT vs CC, miR-499: GG vs AA), heterozygote model (miR-146a: GC vs GG, miR-196a2: TC vs CC, miR-499: AG versus AA), dominant model (miR-146a: CC + GC vs GG, miR-196a2: TT + TC vs CC, GG + AG vs AA), and recessive model (miR-146a: CC vs CG + GG, miR-196a2: TT vs TC + CC, miR-499:GG vs AG + AA). The statistical significance of the pooled OR was evaluated by Z test and a P value of < 0.05 was regarded as significant. Inter-study heterogeneity was tested using a X 2 -based Q-test (with significance level P < 0.1) and I 2 statistic (with values greater than 50% indicating significant heterogeneity) [53]. According to the result of the heterogeneity test, the random model was chosen to assess OR and 95% CI when P < 0.05; conversely, the fixed model was selected when P > 0.05. Subgroup analysis was performed by race, source of control, and sample size. Sensitivity analysis was performed to evaluate the effect of each study on the combined ORs by omitting individual studies one at a time. Publication bias was checked by Begg's funnel plots [54] and Egger's regression test [55]. An asymmetric plot and a P < 0.05 for the Egger's test denoted a noteworthy publication bias. The trim-and-fill computation was used to estimate the effect of publication bias on the interpretation of the results [56]. Statistical analysis was conducted utilizing Stata12.0 Software.

Authors ̕ contributions
K.M., W.G. and L.J.W. designed the study. J.P.Y., H.Z. and J.T.L. collated the data, designed and developed the database, carried out data analyses and produced the initial draft of the manuscript. Z.Z.W., N.W. and J.Z. contributed to drafting the manuscript. All authors have read and approved the final submitted manuscript.