Association of mir-196a-2 rs11614913 and mir-149 rs2292832 Polymorphisms With Risk of Cancer: An Updated Meta-Analysis

Background: Accumulating evidence suggests that functional dysregulations of miRNAs, especially miR-196a-2 and miR-149, in cancers could be attributed to polymorphisms in miRNA sequences. This study was aimed at clarifying the association of mir-196a-2 rs11614913 and mir-149 rs2292832 with cancer risk by performing an updated meta-analysis of genetic association studies. Methods: PubMed, Embase, Scopus, and ScienceDirect databases were searched until 9 April 2018 to identify eligible studies. Studies should meet the following criteria to be included in the meta-analysis: evaluation of genetic association between rs11614913 and/or rs2292832 and susceptibility to cancer; A case-control design; Written in English; Availability of sufficient data for estimating odds ratio (OR) and its 95% confidence interval (95%CI). Studies that met the following criteria were excluded: review articles, meta-analysis, abstracts or conference papers; duplicate publications; studies on animals or cell-lines; studies without a case-control design; studies that did not report genotype frequencies. Pooled ORs and 95% CIs were estimated using a total of 111 studies (41,673 cases and 49,570 controls) for mir-196a rs11614913 and 44 studies (15,954 cases and 19,594 controls) for mir-149 rs2292832. Stratified analysis according to quality scores, genotyping method, ethnicity, broad cancer category and cancer type was also performed. Results: Mir-196a-2 rs11614913 T allele was associated with decreased cancer risk in overall population. The association was only significant in Asians but not Caucasians. In subgroup analysis, significant associations were found in high quality studies, gynecological cancers, ovarian, breast, and hepatocellular cancer. Mir-149 rs2292832 was not associated with cancer risk in overall population and there were no differences between Asians and Caucasians. However, the T allele was associated with a decrease risk of gastrointestinal tract cancers under the heterozygote model and an increased risk of colorectal cancer under the recessive model. Conclusions: The present meta-analysis suggests that mir-196a-2 rs11614913 may contribute to the risk of cancer especially in Asians. Mir-149 rs2292832 may modulate the risk of gastrointestinal tract cancers especially colorectal cancer. This study had some limitations such as significant heterogeneity in most contrasts, limited number of studies enrolling Africans or Caucasians ancestry and lack of adjustment for covariates and environmental interactions.


INTRODUCTION
Despite remarkable recent progress in clinical management, diagnosis and treatment, cancer has remained one of the major causes of death worldwide. According to the recent World Health Organization (WHO) report, about one in six deaths were caused by cancer in 2015. It is predicted that cancer-related death will increase up to 13.2 million by 2030 worldwide (Ferlay et al., 2010;Bray et al., 2012). Complex genetic and environmental risk factors and also interactions between these components contribute to the etiopathology of different cancers. Until recent years, much effort has been devoted to link the alteration of protein coding genes to tumorigenesis. However, latest evidence has demonstrated the emerging role of noncoding RNAs in cancer development and, especially, introduced microRNAs (miRNAs) as new players in pathobiology of cancers (Peng and Croce, 2016). MiRNAs are short noncoding functional RNAs that are involved in the regulation of transcriptome (Ha and Kim, 2014). They modulate important cellular processes both in normal physiology and disease state and are involved in almost all cellular processes altered during tumorigenesis (Osada and Takahashi, 2007;Li et al., 2009). Human mir-196a (MIR196A2, HGNC:31568) and mir-149 (MIR149, HGNC: 31536) are wellstudied miRNAs that may function either as oncomiRs, by targeting tumor suppressor genes, or as tumor suppressors, by targeting oncogenes, in different conditions Ow et al., 2018). It has been shown that single nucleotide polymorphism (SNP) in miRNA genes, such as hsa-mir-196a-2 rs11614913 and hsa-mir-149 rs2292832, may influence their functions through altering miRNA expression, maturation and/or efficiency of targeting and, thereby, contribute to the risk of cancer (Hu et al., 2008;Hoffman et al., 2009;Tu et al., 2012;Nariman-Saleh-Fam et al., 2016. Several association studies in a range of populations evaluated the contribution of mir-196a-2 rs11614913 and mir-149 rs2292832 to cancer risk; but results are inconclusive. Therefore, this study was aimed at clarifying the association of mir-196a-2 rs11614913 and mir-149 rs2292832 with cancer risk by performing an updated meta-analysis of genetic association studies.

Inclusion and Exclusion Criteria
Studies should meet the following criteria to be included in the meta-analysis: (1) evaluation of genetic association between rs11614913 and/or rs2292832 and susceptibility to cancer; (2) a case-control design; (3) Written in English; (4) Availability of sufficient data for estimating odds ratio (OR) and its 95% confidence interval (95%CI). Studies that met the following criteria were excluded: (1) review articles, meta-analysis, abstracts or conference papers; (2) duplicate publications; (3) studies on animals or cell-lines; (4) studies without a case-control design (5) studies that did not report genotype frequencies.

Data Extraction
Data was extracted from each eligible study and manually checked. Then, items were recorded for each eligible study: the first author, publication year, category of cancer, type of cancer, country, ethnicity, source of controls, genotyping method, number of subjects in the case and the control groups, genotype counts for each SNP in the case and the control groups. A broad cancer category was assigned for each study according to the following scheme: gastrointestinal tract cancers (GI, including gastric, esophageal, colorectal, bladder, pancreatic, or hepatocellular cancers), head and neck squamous cell carcinoma (HNC, including oral, non-oral, and nasopharyngeal cancers), gynecologic cancer (GyC, including endometrial, ovarian, and cervical cancers), hematological malignancies (HM, including leukemia and lymphomas), urogenital cancers (UG, including prostate, renal cell, and bladder cancers), or other cancers.

Quality Assessment
The quality of each study was assessed using quality assessment criteria for genetic association studies used elsewhere (Thakkinstian et al., 2011;Xue et al., 2015). This score is based on seven items including representativeness of cases, representativeness of controls, ascertainment of cancer case, control selection, genotyping examination, Hardy-Weinberg Equilibrium (HWE) status in controls, and total sample size. Quality scores ranged from 0 (lowest) to 15 (highest). Studies which were scored equal to or less than eight were regarded as low quality, while those with scores of greater than eight were regarded as high quality.

Statistical Analysis
The Meta package for R was used to perform meta-analysis (Schwarzer, 2007). Association of rs11614913 and rs2292832 with cancer was estimated by calculating pooled ORs and their 95% CIs assuming homozygote, heterozygote, dominant, recessive, and allelic models. Heterogeneity was assessed using the Chi-squared based Q test (Lau et al., 1997). In the presence of a significant heterogeneity (i.e., P-value of Qtest < 0.05 or I 2 > 50%), the random effect (RE) model (DerSimonian and Laird, 1986) was used to calculate pooled ORs and 95% CIs. Otherwise, the fixed effect (FE) model was used (Mantel and Haenszel, 1959). Significance of the pooled OR was determined by the Z-test (P < 0.05 was considered significant). In cases of remarkable heterogeneity (i.e., I 2 > 50%), the potential sources of heterogeneity across studies was explored using univariate meta-regression and stratified analysis. Moreover, subgroup analyses based on genotyping method, study quality, ethnicities, broad cancer categories, and cancer types were carried out. To assess consistency of results and influence of each study on the pooled OR, sensitivity analysis was done by omitting one study at a time and recalculating summary OR and 95% CI. Publication bias was evaluated by the Begg's rank correlation test of funnel plot asymmetry (Begg and Mazumdar, 1994) the "Trim and Fill" approach was used to correct for asymmetry in cases of significant rank correlation test (Duval and Tweedie, 2000a,b). All Pvalues were two-sided and P-value < 0.05 was considered statistically significant. All statistical analyses were performed in R (version 3.3.1).

Dealing With HWD (Departure From Hardy-Weinberg Equilibrium)
Departure from Hardy-Weinberg equilibrium (HWE) may be caused by a range of factors, among which genotyping error is more importantly relevant to the association study context. Currently there is no consensus on the way of handling association studies with the controls not in HWE, but it has been recommended that such studies should not be excluded from meta-analysis (Minelli et al., 2008). However, sensitivity analysis should be performed to evaluate the possible effects of such studies on the pooled estimates (Attia et al., 2003;Thakkinstian et al., 2005;Zintzaras and Lau, 2008;Wang X. B. et al., 2014). In the present metaanalysis, the following approach with regards to HWE-deviated studies was followed. Departure of genotype distributions from HWE (i.e., HWD) in the control group of each study was evaluated using the Chi-squared or the exact goodness of fit test. Meta-analyses, including the overall and subgroup analyses, were performed considering all eligible studies including HWD studies. However, to evaluate possible impacts of HWE-deviated studies, HWD sensitivity analysis was performed by evaluating the influence of excluding these studies on point estimates and identifying the influenced genotype contrasts. In cases that excluding HWD studies altered the result of meta-analysis, ORs of such studies were adjusted for HWE deviation by means of incorporating the HWEexpected genotype counts in the control group as recommended (Trikalinos et al., 2006;Zintzaras et al., 2006;Zintzaras, 2008;Zintzaras and Lau, 2008;Srivastava and Srivastava, 2012) and the HWD-adjusted pooled ORs were calculated in genotype contrasts.
Moreover, 43 articles comprising 44 studies (15,954 cases and 19,594 controls) evaluated the association of mir-149 rs2292832 and cancer risk (     Figure S4 for forest plots and Supplementary Figure S11 for funnel plots of these models). Significant associations are shown in boldface.
CI]: 0.89 [0.83-0.95], P: 0.001) models (Table 3). Mir-196a-2 rs11614913 was not associated with cancer risk in the heterozygote and dominant models and there was a nonsignificant borderline association in the allele contrast (Table 3). Supplementary Figure S1 shows the forest plots of association of mir-196a-2 rs11614913 and cancer risk in five models. The results of subgroup analysis for mir-196a-2 rs11614913 are shown in    (Figure 3). No significant findings were observed for gastrointestinal, head and neck, hematological or urogenital cancers (Table 4). Supplementary Figure S2 presents the forest plots for subgroups according to the broad cancer categories. Further subgrouping by cancer type revealed significant association of mir-196a-2 rs11614913 with hepatocellular carcinoma (Figure 4) Figure S3 presents the forest plots for subgroups according to cancer type. HWD sensitivity analysis (i.e., excluding studies with controls deviated from HWE) revealed stable results in the overall analysis under the homozygote, heterozygote, dominant, and recessive models ( Table 4). However, excluding HWD studies made the borderline allele contrast statistically significant (Table 4). Moreover, excluding HWE violating studies had no dramatic effects on subgroup meta-analyses using quality of studies, genotyping methods, the ethnicity and the broad cancer category (Supplementary Table S1). In meta-analysis subgrouped by cancer type, the results were also stable for gastric, bladder, oral, ovarian, prostate, and esophageal cancer subgroups after excluding HWD studies (Supplementary Table S1). However, excluding such studies altered the results for the breast, hepatocellular, colorectal, and lung cancer subgroups. Therefore, for these subgroups, pooled ORs were estimated to account for departures from HWE (denoted as HWD-adjusted ORs) (Supplementary Table S2). When corrected for HWD, mir-196a-2 rs11614913 was found to be significantly associated with breast cancer under the homozygote (TT vs. CC, HWDadjusted OR RE Table S2).

Association of mir-149 rs2292832 and Cancer Risk
The overall analysis showed no significant association with cancer risk under any genetic model ( Table 3). Supplementary Figure S4 shows the forest plots for the association of mir-149 rs2292832 and cancer risk under different genetic models. However, in subgroup analyses (  (Table 5). Supplementary Figures S5, S6 show the forest plots for subgroup analysis according to the broad cancer category and cancer type, respectively. Sensitivity analysis revealed that HWD studies had no significant effect on point estimates in the overall meta-analysis of mir-149 rs2292832 and cancer risk, and still no significant association was observed in overall analysis (Table 5). Moreover, most subgroup analyses were also stable after removing HWD studies (Supplementary Table S3). However, removing studies with HWD controls influenced comparisons in three subgroups: (i) non-PCR-RFLP subgroup (heterozygote model); (ii) the breast cancer subgroup (recessive model); (iii) the colorectal cancer subgroup (recessive model). Therefore, for these subgroups, HWE-expected genotype distributions in controls were used for pooling ORs (denoted as HWD-adjusted OR) (Supplementary Table S4). Adjusting for HWD in these subgroups confirmed the results of original analyses and showed that rs2292832 is associated with cancer risk in non-RFLP subgroup under the heterozygote model (Figure 8,

Heterogeneity, Meta-regression, and Sensitivity Analysis
Heterogeneity was evaluated for both polymorphisms in all genetic models (Tables 3, 4, 5). Significant between study heterogeneity was observed in the overall estimation under all genetic models for mir-196a-2 rs11614913 and consequently random effect model was used. Univariate meta-regression using cancer type, country, ethnicity, the quality of study (either high or low), genotyping method, source of controls (PB or HB) or HWE was performed to identify potential sources of heterogeneity. For mir-196a-2, meta-regression showed that at least a part of the observed between study heterogeneity in the heterozygote (R 2 : 24.23%, P: 0.007) and dominant (R 2 : 19.86%, P: 0.028) models could be attributed to the country moderator. However, there was still significant unaccounted heterogeneity even after correcting for the effect of country moderator (Heterozygote I 2 : 63.78 and dominant I 2 : 72.05, P for test of residual heterogeneity < 0.0001). Moreover, Galbraith plot analysis demonstrated three studies Wang et al., 2013;Dikeakos et al., 2014) as the most extreme outliers in all genetic models that account for a considerable portion of the observed heterogeneities (Supplementary Figure S7). Excluding these studies led to a 11.6% reduction of I 2 in the homozygote model (from 77.1 to 65.5%), a 12.8% reduction in the heterozygote model (from 70.6 to 57.8%), a 12.3%   reduction in the dominant model (from 76.9 to 64.6%), an 9.1% reduction in the recessive model (from 68.8 to 59.7%) and a 11.3% reduction in the allelic model (from 79 to 67.7%). However, excluding these studies did not alter any genotypic contrast and results were comparable to the overall analyses (data not shown). Sensitivity analysis by omitting one study at a time revealed that no individual study significantly influenced the genotype contrasts (Supplementary Figure S8).
In the allele contrast, omitting no single study dramatically influenced pooled OR or its 95%CI. However, given that the original 95%CI was borderline (0.90-1.00), omitting some studies lead the upper limit of 95%CI to fall slightly below one (Supplementary Figure S8-e).
Statistically significant heterogeneity was also observed in the overall analysis of miR-149 rs2292832 and cancer risk and, consequently, RE model was used to estimate pooled OR (Table 3). Subgrouping by study level moderators led to a reduction in heterogeneity in some subgroups (Table 5). However, univariate metaregression showed no statistically significant source of heterogeneity (All P > 0.05). Sensitivity analysis by omitting one study at a time revealed no single influential study (Supplementary Figure S9).

Publication Bias
Rank correlation test of the mir-196a-2 rs11614913 Begger's funnel plot asymmetry revealed no statistically significant evidence of publication bias in any contrast (Table 3 and Supplementary Figure S10). However, rank correlation test for asymmetry of mir-149 rs2292832 funnel plots showed statistically significant results in the homozygote, recessive and the allelic contrasts (Table 3 and Supplementary Figure S11). Consequently, the "trim and fill" approach (Duval and Tweedie, 2000a,b) employed to correct for funnel plot asymmetry arising from publication bias in these models. The results of overall analysis using original studies or trim and fill method under the three models were comparable (see Supplementary Figure S4 to compare forest plots of the original studies vs. trim-and-fill method and Supplementary Figure S11 for funnel plots). After excluding studies with controls deviating from HWE in sensitivity analysis, rank correlation test was still significant in the mentioned three contrasts.

DISCUSSION
The possible contribution of miRNAs, especially mir-196a-2 and mir-149, to the risk of cancer has stimulated great attention in recent years. Many studies evaluated the functional alterations of these micro-regulators in a wide range of cancers. Accumulating evidence suggests that, at least a part of functional dysregulations of miRNAs in cancers could be attributed to polymorphisms in miRNA sequences (Hu et al., 2008;Hoffman et al., 2009;Tu et al., 2012;Ghaedi et al., 2015;Nariman-Saleh-Fam et al., 2016. Two mature miRNAs, miR-196a-5p and miR-196a-3p, are generated from the stem-loop structure of hsa-mir-196a-2 (Kozomara and Griffiths-Jones, 2014) with the studied polymorphism, rs11614913, residing in the 3 ′ arm (Figure 9A). This polymorphism, therefore, may potentially alter miRNA processing and also binding to related target mRNAs (Hoffman et al., 2009) (Figure 9B). Previous studies have shown that the expression level of mature miR-196a-3p was higher in CC carriers with lung cancer compared to CT and TT individuals (Hu et al., 2008). More evidences have been provided by Hoffman et al. (2009) who observed elevated expression of mature mir-196a-2 forms in MCF-7 cells transfected with pre-mir-196a-C vector compared with cells transfected with pre-mir-196a-T vector. The potential of rs11614913 in influencing targeting function of mir-196a-2 has also been documented by whole-genome expression microarrays which found different numbers of FIGURE 6 | Forest plot of the original and HWD-adjusted meta-analysis of breast cancer risk and mir-196a-2 rs11614913 under the recessive (TT vs. CT+CC) and homozygote (TT vs. CC) models. Red boxes represent HWD-adjusted log ORs and 95%CIs of studies with HWE-deviated controls (i.e., Linhares et al., 2012;Ma et al., 2013;Afsharzadeh et al., 2017;Doulah et al., 2018). The blue and red diamonds represent the estimated pooled effects of the original and HWD-adjusted meta-analysis, respectively. dysregulated mRNAs after transfecting cells with pre-mir-196a-C or pre-mir-196a-T vector (Hoffman et al., 2009). Hsa-mir-149 also generates two mature miRNAs (miR-149-5p and miR-149-3p) and the studied polymorphism, rs2292832, does not reside in the mature sequence of neither miR-149-5p or miR-149-3p ( Figure 10A). Therefore, it has been hypothesized that rs2292832 is not a structure-shifting polymorphism for pri-mir-149 or pre-mir-149 (Wei et al., 2014). However, Tu et al. reported that the T allele may disrupt the maturation process compared with the C allele and, consequently, decrease miR-149 expression (Tu et al., 2012) (Figure 10B) in head and neck squamous cell carcinoma patients.  The overall, HWD-sensitivity and subgroup analyses. a n represents number of cases and controls in each group. Increasing number of association studies evaluating miRNA polymorphisms and cancer risk with contradictory results merits the need for comprehensive systematic reviews and meta-analyses. Several meta-analyses have evaluated the risk of cancer associated with mir-196a-2 rs11614913 or mir-149 rs2292832 (Chu et al., 2011;Feng et al., 2016;Yan et al., 2017;Liu et al., 2018). However, the conclusion of these studies with regards to the subgroup analysis and the significant genetic model varies due, at least in part, to differences in the number of studies included or in the methodology. Moreover, several recent genetic association studies have not been included in previously published metaanalyses. Therefore, it was necessary to perform an updated meta-analysis with larger number of studies to clarify the association of mir-196a-2 rs11614913 or mir-149 rs2292832 with cancer risk. Therefore, compared to the previous meta-analysis, we included more studies in the analyses. The present metaanalysis also evaluated and corrected for the possible influence of departure of the control group of association studies from HWE. Although checking for departure from HWE has been recommended, currently there is no consensus about how HWD studies should be handled in meta-analysis (Minelli et al., 2008). Result of simulations suggests no advantage for excluding these studies (Minelli et al., 2008). However, sensitivity analysis to detect any possible bias imposed by such studies and/or using HWE-expected counts instead of the observed genotype frequencies have been recommended and implemented in several studies (Attia et al., 2003;Thakkinstian et al., 2005;Trikalinos et al., 2006;Zintzaras et al., 2006;Zintzaras, 2008;Zintzaras and Lau, 2008;Srivastava and Srivastava, 2012;Wang X. B. et al., 2014). The current study noticed that most analyses, especially those with sufficiently large number of studies, were not influenced by excluding HWD studies. However, as it is rationally expected, excluding HWD studies from some subgroup with relatively small number of studies may influence the analysis, and therefore in such situations adjusted analyses were preferred.
For mir-196a-2 rs11614913, the previous largest metaanalysis, conducted by Liu et al. (2018), included 84 studies compared to 111 studies in the present meta-analysis. By including 41,673 patients and 49,570 control subjects, the present meta-analysis showed a decreased risk of cancer in the homozygote and the recessive models (Table 3). Although, the association was not significant in allele contrast, the OR and 95%CI of the allele contrast were borderline and influenced by excluding some individual studies. Excluding HWE-deviated or low quality studies yielded significant associations under allelic model. As there is, currently, no way to adjust allele frequencies for departure from HWE, the possibility that HWD studies may bias the allele contrast cannot be rolled out and a definite conclusion cannot be drawn under allelic model. Apart from the allele contrast, the results of other genetic models were statistically stable and not influenced by removing any single study, HWE deviated or low quality studies. The results also suggest that mir-196a-2 rs11614913 may pose an ethnic dependent effect on cancer risk as associations with cancer were only observed in Asians. However, it should be noted that most  Lv et al. (2013). The blue and red diamonds represent the estimated pooled effect of the original and HWD-adjusted meta-analysis. studies enrolled Asian patients, mainly Chinese patients, and the number of studies involving other ethnicities were relatively small. Moreover, different minor allele frequencies (MAF) may partly contribute to the observed differences among ethnicities (Average MAF in Asians T: 0.501 ± 0.127, Caucasians T: 0.410 ± 0.1, Others T: 0.338 ± 0.066). In-line with previous studies, the current meta-analysis also confirmed that rs11614913 is associated with decreased risks of hepatocellular cancer under three genetic models  and that it may not modulate risk of urogenital cancers . The results of meta-analysis of all studies, subgroup analysis by ethnicity and hepatocellular cancer are in agreement with findings of the previous largest meta-analysis . However, the increase in the number of analyzed studies led to discrepancies with regards to conclusions in some subgroup analysis. (i) In contrast to the studies by Liu and Pan (Pan et al., 2017;Liu et al., 2018), the present meta-analysis did not find a significant association with head and neck carcinoma in any genetic model. This discrepancy may be attributed to the number of studies included in meta-analyses [nine studies in the present meta-analysis compared to four studies in the meta-analysis by Liu et al. (2018)]. Furthermore, differences in defining head and neck cancer may also explain different conclusions drawn from the present study and the study by Pan and colleagues (Pan et al., 2017). They included esophageal cancer as a type of head and neck cancer, whereas we considered it as a type of gastrointestinal tract cancers (according to the ICD-10-CM C15-C26). (ii) Additionally, the present meta-analysis found significant associations between mir-196a-2 rs11614913 and decreased risks of gynecologic cancers (especially ovarian cancer), which have not been reported in any previous metaanalysis. Interestingly, low heterogeneity was observed in the gynecological cancers subgroup assuming the two significant contrasts (i.e., recessive and allelic). (iii) Although previous metaanalyses Zhang H. et al., 2017;Liu et al., 2018) failed to find a significant association between mir-196a-2 rs11614913 and breast cancer, the current study showed, by incorporating more association studies and performing FIGURE 8 | Forest plot of the original (Left) and HWD-adjusted (Right) meta-analysis of cancer risk associated with mir-149 rs2292832 in subgroup of studies which used a non-RFLP genotyping method assuming the heterozygote model. In the right plot, HWD-adjusted log ORs and 95%CI were used for the underlined studies (i.e., Kontorovich et al., 2010;Pratedrat et al., 2015;Jiang et al., 2016).
HWD sensitivity analysis, that adjusting for departure from HWE may reveal significant associations under the homozygote and recessive contrasts (Figure 6 and Supplementary Table S2). (iv) Moreover, contradictory to previous meta-analyses, no association with gastric , colorectal (Xie et al., 2015;Yan et al., 2017) or lung cancer Yan et al., 2017;Liu et al., 2018), was found. Apart from larger sample sizes and correcting for HWD, sometimes this discrepancy in results may also be related to methodological differences in the design, specifically the inclusion criteria, of meta-analyses. As a case in point, studies by Hu et al. (2008) and Yoon et al. (2012) did not meet the inclusion criteria of our study, as they deal with the survival or recurrence risk of lung cancer patients with approaches that differed from routine case-control genetic association studies. However, we noticed that these studies were included in a previous meta-analysis .
The current meta-analysis also showed that mir-149 rs2292832 is not associated with risk of cancer in any genetic model and the results were statistically reliable, as summary effects were not influenced by excluding any single study, HWE-deviated or low quality studies. No differences in cancer risk was observed between ethnicities. Similar to miR-196a-2 polymorphism, the Asian subgroup comprised a large proportion of studies and relatively few studies with limited sample sizes were performed in Caucasians. Therefore, a definite conclusion cannot be drawn in Caucasians and more studies are needed further clarify the association of this SNP with cancer risk in Caucasians. The results of overall analysis were comparable to previous meta-analyses Feng et al., 2016). By pooling the results of 22 studies, this meta-analysis found a decreased risk of gastrointestinal tract (GI) cancers for individuals who carry the CT genotype compared to those with the CC genotype in a heterozygote model. Interestingly, there was no significant heterogeneity in the GI subgroup assuming the heterozygote model indicating the reliability of meta-analysis in this subgroup. Previous meta-analyses yielded different results with regard to GI cancers. A previous meta-analysis of seven studies on GI cancers suggested a marginally elevated risk under the recessive model (TT vs. CT+CC), while another pooling of 10 studies found a borderline decreased risk for the CT vs. TT contrast. Although no significant association was identified in the head and neck cancers subgroup, it should be noted that significant heterogeneity was present in all models except the recessive contrast and number of samples were relatively small. For colorectal cancer the association was in reverse direction FIGURE 9 | Schematic illustration of hsa-mir-196a-2 locus and effect of rs11614913 on its processing and targeting. (A) Top: Hsa-mir-196-a-2 generates two mature miRNAs, miR-196a-5p and mir-196a-3p, and rs11614913 lies in miR-196a-3p at GRCh37 chr12: 54,385,599. Bottom: Stem-loop structure of hsa-mir-196, including mature miR-196-5p (Red) and miR-196-3p (Blue) sequences. (B) Rs11614913 alters miRNA processing and/or binding to related target mRNAs. and an increased risk was observed in individuals who carry the TT genotype compared with subject who carry at-least one C allele (Table 5). A similar association based on three studies was previously reported (Rong et al., 2017), but not reproduced in other meta-analyses Feng et al., 2016). Taken together, the current results based on five studies suggest an increased risk for colorectal cancer that was stable after correcting for departure from HWE. Although no significant heterogeneity was detected in the colorectal cancer subgroup under any genetic model, it should be noted that the limited number of studies may influence heterogeneity evaluation and more definite conclusion may be drawn by analyzing larger sample sizes. In the case of breast cancer, a previous meta-analysis of three studies found a significant association (Feng et al., 2016). We found no significant association in the original and HWD-adjusted analysis. However, number of studies in the colorectal and breast cancer subgroups are relatively limited and results should be interpreted with caution. More studies with large sample sizes are needed for a definite conclusion.
However, the present study has some limitations. First, significant heterogeneity was present in most analyses especially for mir-196a-2 polymorphism. We, therefore, used random effect model and performed meta-regression; but no significant source of heterogeneity was observed for most analyses, suggesting that other unknown study level moderators may contribute to the heterogeneity. Second: The molecular mechanisms underlying association of these miRNAs-SNP with risk of cancer are complex FIGURE 10 | Schematic illustration of hsa-mir-149 locus and effect of rs2292832 on its processing. (A) Top: Hsa-mir-149 generates two mature miRNAs (miR-149-5p and miR-149-3p), and rs2292832 resides in pre-miRNA sequence, but not in either mature miRNAs, at GRCh37 chr2: 241,395,593. Bottom: Stem-loop structure of hsa-mir-149, including mature miR-149-5p (Red), and miR-149-3p (Blue) sequences. (B) Rs12292832 may alter miRNA processing and decrease mature miR-149 (−5p and −3p forms) expression. and might be strongly affected by different genetic background as well as other masked variables. This, in turn, may limit the efficacy of the overall analysis especially in the case of miR-196a-2 rs11614913. Stratified analyses based on a specific cancer category or a cancer type may help to reduce this heterogeneity and, therefore, are considered to be more reliable. Third, this study was based on unadjusted ORs of the original studies and no adjustment for covariates like age and gender or interaction with environmental factors were done and this fact may also potentially contribute to the between study heterogeneity. Fourth, some limitations such as language restriction or lack of access to the genotype counts of mir-196a-2 rs11614913 in four studies with insufficiently reported data may bring in publication bias. The trim and fill method has been shown to reduce the bias in estimates in the presence of publication bias and heterogeneity (Peters et al., 2007). However, it has been recommended that this method should be considered as sensitivity analysis as we cannot be sure that asymmetry in funnel plot is truly caused by publication bias (Peters et al., 2007). Although rank correlation test of funnel plots of mir-149 rs2292832 was significant in three genetic models raising the possibility of publication bias, adjusting for such a bias using trim-and-fill method did not afford any change in analysis of overall studies in these models (Supplementary Figure S4). Fifth, number of studies in some subgroup analyses was limited and, consequently, results of such analysis should be interpreted with caution. Most studies were performed enrolling Asian patients and the number of studies on Caucasians or Africans was limited. Therefore, more association studies with larger sample sizes on Africans and Caucasians are needed to make precise estimations of cancer risk associated with the studied polymorphisms. Assigning ethnicity to each study population could be another limitation of meta-analysis of association studies as each ethnicity may regroup several sub-populations with somewhat different genetic background. Sixth, the control groups of association studies were not uniformly defined and non-differential misclassification bias may have occurred.
In conclusion, this meta-analysis showed that mir-196a-2 rs11614913 T allele is associated with decreased cancer risk in overall population, high quality studies and studies on Asian populations. It is also associated with a decreased risk of gynecological cancers, ovarian, breast and hepatocellular cancer. Mir-149 rs2292832 was not associated with cancer risk in overall population, high quality studies, Asians or Caucasians. However, the T allele was associated with a decrease risk of gastrointestinal tract cancers under the heterozygote model and an increased risk of colorectal cancer under the recessive model.

AUTHOR CONTRIBUTIONS
MB and AM conceived the original idea and supervised the project. JC, ZN-S-F, and ZS contributed to the literature search and data management. MB, JC, and ZN-S-F wrote the manuscript with support from all authors. MB contributed to the data analysis, interpretation of results, and data visualization with inputs from AM. EO and ZS assisted with data visualization. All authors provided critical feedback, discussed the results, and contributed to the final manuscript.