Clinical Theragnostic Relationship between Drug-Resistance Specific miRNA Expressions, Chemotherapeutic Resistance, and Sensitivity in Breast Cancer: A Systematic Review and Meta-Analysis

Awareness of breast cancer has been increasing due to early detection, but the advanced disease has limited treatment options. There has been growing evidence on the role of miRNAs involved in regulating the resistance in several cancers. We performed a comprehensive systematic review and meta-analysis on the role of miRNAs in influencing the chemoresistance and sensitivity of breast cancer. A bibliographic search was performed in PubMed and Science Direct based on the search strategy, and studies published until December 2018 were retrieved. The eligible studies were included based on the selection criteria, and a detailed systematic review and meta-analysis were performed based on PRISMA guidelines. A random-effects model was utilised to evaluate the combined effect size of the obtained hazard ratio and 95% confidence intervals from the eligible studies. Publication bias was assessed with Cochran’s Q test, I2 statistic, Orwin and Classic fail-safe N test, Begg and Mazumdar rank correlation test, Duval and Tweedie trim and fill calculation and the Egger’s bias indicator. A total of 4584 potential studies were screened. Of these, 85 articles were eligible for our systematic review and meta-analysis. In the 85 studies, 188 different miRNAs were studied, of which 96 were upregulated, 87 were downregulated and 5 were not involved in regulation. Overall, 24 drugs were used for treatment, with doxorubicin being prominently reported in 15 studies followed by Paclitaxel in 11 studies, and 5 drugs were used in combinations. We found only two significant HR values from the studies (miR-125b and miR-4443) and our meta-analysis results yielded a combined HR value of 0.748 with a 95% confidence interval of 0.508–1.100; p-value of 0.140. In conclusion, our results suggest there are different miRNAs involved in the regulation of chemoresistance through diverse drug genetic targets. These biomarkers play a crucial role in guiding the effective diagnostic and prognostic efficiency of breast cancer. The screening of miRNAs as a theragnostic biomarker must be brought into regular practice for all diseases. We anticipate that our study serves as a reference in framing future studies and clinical trials for utilising miRNAs and their respective drug targets.

The recent surge in the number of cancer cases along with the development of drug resistance in a large number of tumours has pushed the direction of cancer research towards new arenas that provide the grounds for the development of more effective personalised medicine treatment. MicroRNAs (miRNAs) pave the way for this by being potential biomarkers for early cancer detection, and could also help in designing a more specific treatment plan by helping in the analysis of drug resistance and sensitivity [28]. Various studies have been conducted highlighting the effect of miRNAs in chemotherapeutic resistance in cancers such as gastric cancer [29], breast cancer [30], cervical cancer [31], colorectal cancer [32], lung cancer [33], oral cancer [34], ovarian cancer [35], pancreatic cancer [36], prostate cancer [37] and skin cancer [38].
When an MCF7 (Michigan Cancer Foundation-7 cell line treated with VP-16 (etoposide) was compared with the untreated parent MCF7 cell line, it was observed that 17 miRNAs had abnormal levels of expression; the majority of them were up-regulated, whereas miR-326, miR-429, miR-187, miR-7, and miR-92-2 showed decreased expression [41]. The results were verified by RT-PCR, and it was concluded that these miRNAs could be specific regulators of MRP1 (multidrug-resistance-associated protein) and play a critical role in MDR (multiple drug resistance) [41].
A clinical study comparing the effects of the drug tamoxifen versus tamoxifen plus breast radiotherapy, carried out on 71 lymph-node-negative (LNN) breast cancer patients, revealed that the up-regulation of miRNA-301 in co-operation with SKA2 (spindle kinetochore-associated complex subunit 2) increased proliferation, migration, invasion and tumour formation through the regulation of key signalling pathways including PTEN, FOXF2 and Col2A1 [42]. According to another study, high levels of miRNA-210 expression in plasma was observed to be associated with trastuzumab resistance in HER-2 (human epidermal growth factor receptor 2)-positive breast cancer patients [43]. Xiang Ao and his colleagues examined 55 pairs of breast cancer tissues and adjacent normal tissues in total, and found that resistance to taxol in breast cancer patients increased with the loss of miRNA-17 and miRNA-20b, by the up-regulation of nuclear receptor co-activator 3 (NCOA3) levels [44].
Over the years, several studies have focused on the role of various miRNAs in the chemotherapeutic resistance or sensitivity in breast cancer. However, none of these studies have been able to conclusively define the exact mechanism by which these miRNAs are involved in chemo-sensitivity/resistance. Through this study, we aim to provide insight into the association of the expression of specific miRNAs with breast-cancer-related chemotherapeutic drug resistance and sensitivity, thereby making it relevant in a clinical setting. Further, this study paves the way to devise new treatment strategies targeting these miRNAs, and developing alternate ways to counter the occurrence of chemo-resistance in breast cancer. This study was carried out with the aid of tools including meta-analysis and systematic review.

Methods
To obtain studies to perform the meta-analysis, two databases were extensively used: PubMed and Science Direct. This systematic review required articles related to the chemotherapeutic resistance specific to miRNA in breast cancer. To obtain relevant papers, the selection was performed using of the following MeSH (Medical Subject Heading) terms: "miRNA" or "microRNA", "drug resistance" and "breast cancer". To further refine the process of selection, only papers published within 2012-2018 were selected. This systematic review and meta-analysis study adheres to the Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) guidelines [45].
The study search ended on (31 December 2018). After the initial screening process, additional studies were obtained via the reference section of relevant articles. The relevance of articles was determined by reading the title and abstract followed by the analysis of the complete text. The search was conducted in an orderly and elaborate manner, and was designed to meet the requirements of the study.

Selection Criteria
Studies that were to be used in the systematic review and meta-analysis had to adhere to certain selection criteria. These criteria were of two types: inclusion criteria and exclusion criteria. The inclusion criteria set the guidelines for the studies that could be included in the analysis process and included the following factors: 1.
An analysis of the association between miRNA and breast cancer; 2.
Studies with both breast cancer patients as well as in vitro studies with cell lines; 3.
Studies that focused on cancer tissues that had resistance to some form of therapy; 4.
Information about the genes or pathways involved in chemotherapeutic resistance or sensitivity; 6.
Inclusion of some in vitro assays to analyse the expression of miRNA or gene-related studies.
Some studies were not considered because of certain exclusion criteria. These included studies that were not in the English language, did not involve drug resistance in breast cancer, studies involving microbes and those focusing only on long non-coding (lnc) RNA. Additionally, review articles, editorials and studies with only in vitro or only breast cancer patient samples were excluded.

Data Analysis
The studies were evaluated separately by both authors (RJ and MRM), and further elaboration was performed with the help of corresponding authors. All articles were subject to the exclusion and inclusion criteria. An MS Excel worksheet (Master) was used to structurally store all the information obtained from the studies that qualified for final inclusion. After a complete survey of full-text and supplementary material, the data from all the studies were broken down under the following important headings: First author, Year of publication, Patients information, Location of the study, Ethnicity, Gender, Drug used, Clinical stages, Number of samples, Lymph node metastasis, Cell lines used, miRNA(s) involved, miRNA profiling platform and Drug pathways or gene associated. A number of biochemical and molecular assays were used to qualitatively and quantitatively analyse the miRNA expression in various studies. The frequency of their usage in all studies were compared and duly represented in a graphical form.
For further qualification of the studies, they had to pass a set of criteria that ensured a degree of quality control [46][47][48]. Two of the authors (RJ and MRM) critically assessed the quality of eligible articles for epidemiological studies based on some checklists derived from Dutch Cochrane Centre represented by Meta-analysis Of Observational Studies in Epidemiology (MOOSE) [49]. The studies that were finally selected had to meet all the criteria as determined by the authors. This process of sorting all the information obtained was a step that was crucial to ensure efficient examination of the studies.

Publication Bias
On the basis of a few distinct methods, two of the authors (RJ and MRM) individually assessed the risk of bias [50][51][52][53][54]. This included the number of patients, year of publication, study period, study location and diagnostic procedure. With the information obtained from the eligible studies, the reviewers arrived at a decision [55][56][57][58][59]. Egger's and Begg's bias indicator tests were employed to infer the publication bias along with the inverted funnel plot [60][61][62][63]. The effect size of statistically non-significant, unpublished and small studies was addressed using classic [64] and Orwin [65] fail-safe N tests. Duval and Tweedie's trim and fill calculation was also performed to compute the new size effect, after the removal of an extremely positive and small study, until a symmetric funnel plot was obtained [66]. A third reviewer was consulted to resolve any disagreement regarding the decision of the team.

Statistical Analysis
We used the Comprehensive Meta-Analysis (CMA) 3.0 software for the meta-analysis and calculated the hazard ratios (HRs) with 95% confidence intervals (CIs). Cochran's Q test and Higgins' I 2 statistic [67] were used to obtain the heterogeneity, and statistical significance was defined as a p-value less than 0.01. A fixed-effect model [67] or random-effects model [68] was used to calculate 95% CI in cases where significant heterogeneity was not observed. The overall standard deviation (SD) of each sample from the main sample was calculated using the statistical Z-test.

Results
The eligible studies for our systematic review and meta-analysis through search results identified are shown in the form of the flow chart in Figure 1. Of the 4584 potential studies, 600 were screened for further proceeding and 92 articles were analysed in depth. Finally, 85 studies were found to be confined to the inclusion and exclusion criteria and the eligible studies involved 5159 tissues. The main characteristics of the patients are represented in Table 1. The systematically reviewed articles met all the criteria, and of the 85 articles included only 6 had hazard ratios and 95% confidence intervals and among these 3 articles denoted them directly in the article and 3 were extracted from Kaplan-Meier curves through online software. Between the 85 articles published, 57 were from China, 9 were from the USA, 5 were from Japan, 3 were from India, 2 each were from France, Italy and Taiwan, and there was 1 from each of Argentina, Canada, Finland, South Korea and Spain. Thirty studies used frozen tissues samples, 15 studies used formalin fixed paraffin embedded (FFPE) samples, 6 studies used core needle biopsy and 1 used blood sample. Meanwhile, 33 studies did not mention the type of material used.      [135].

miRNA Pathway Relation
The miRNA and pathways involved in chemoresistance and chemosensitivity are represented in Tables 2 and 3, respectively.      1.191). An extensive examination found that 89 out of 95 articles did not mention the HR and 95% confidence interval values and of the six remaining articles, only three mentioned them in their manuscript and three HR values were obtained from Kaplan-Meier curve through online software. So, 89 studies were excluded from our meta-analysis due to insufficient data. Cumulatively, a meta-analysis was done for six studies encompassing 852 samples (Figure 2).
An unbiased correlation was observed from Begg and Mazumdar rank collection test results. Regarding Duval and Tweedie's trim and fill calculation for the fixed-effect model, the point estimate and 95% confidence interval for the combined studies was 0.83921 (0.69115-1.01899). Under the random-effects model, the point estimate and 95% confidence interval for the combined studies was 0.79909 (0.50575-1.26256). Using trim and fill, these values were unchanged. Egger's regression intercepted at −0.132 with 95% CI from −5.141 to 4.877; t = 0.07, p = 0.945. The 1-tailed p-value was 0.47237, and the 2-tailed p-value was 0.94473. The funnel plot is represented in Figure 3.
An unbiased correlation was observed from Begg and Mazumdar rank collection test results. Regarding Duval and Tweedie's trim and fill calculation for the fixed-effect model, the point estimate and 95% confidence interval for the combined studies was 0.83921 (0.69115-1.01899). Under the random-effects model, the point estimate and 95% confidence interval for the combined studies was 0.79909 (0.50575-1.26256). Using trim and fill, these values were unchanged. Egger's regression intercepted at −0.132 with 95% CI from −5.141 to 4.877; t = 0.07, p = 0.945. The 1-tailed p-value was 0.47237, and the 2-tailed p-value was 0.94473. The funnel plot is represented in Figure 3.

Discussion
This systematic meta-analysis of "the miRNAs that influence the chemoresistance or chemosensitivity to drugs in breast cancer" carefully reviewed over 400 research articles through a systematic PubMed search query from which 80 research articles were scrutinized based on the inclusion criteria.
From the meta-analysis, the results indicate that many miRNAs could intricately orchestrate cellular functions including chemosensitivity/resistance through post-transcriptional control on target gene expression, either canonically or non-canonically. Of the studies included in this metaanalysis, anthracyclines like doxorubicin and epirubicin were predominantly tested in patients/cell lines to study the differential expression of miRNAs followed by tamoxifen in the case of Estrogen Receptor (ER) positive subjects and trastuzumab in the case of Human Epidermal growth factor Receptor (HER) positive subjects. A major limitation in our research is that less than 10% of the 80 papers (6 papers) had direct hazard values that could be utilized for the meta-analysis, reducing the accuracy of the results obtained since only a small fraction of papers were used to give results of the whole, leading to the biasing of the results. There is a possibility of our interpretation being wrong in the context of heterogeneous disease.

Role of miRNAs in Guiding Diagnosis and Prognosis
We extracted the prognosis results of six miRNAs from six different studies. Among the selected miRNAs, two miRNAs (miR200c and miR489) were downregulated and the remaining four miRNAs (miR484, miR4443, miR520h and miR125b) were upregulated. Both downregulated miRNAs were associated with better prognosis; similarly, both miRNAs (miR484 and miR4443) from the overexpressed miRNAs were expressed as better prognosis whereas miR520h and miR125b were associated with poor prognosis.
The overall hazard ratio (95% CI) of the prognostic significance was 0.78 (0.508-1.100) at a pvalue of 0.140 which was analysed by random-effect model. This overall combined sized effect estimate indicates that the miRNAs decreased the likelihood of death of breast cancer patients by 22%. This means an HR value >1 indicates an increased risk of breast cancer survival whereas an HR <1 indicates a decreased risk of breast cancer patient survival. The Z-value of the overall effect size was −1.476. The individual overall hazard ratios (95% CIs) of upregulated and downregulated

Discussion
This systematic meta-analysis of "the miRNAs that influence the chemoresistance or chemosensitivity to drugs in breast cancer" carefully reviewed over 400 research articles through a systematic PubMed search query from which 80 research articles were scrutinized based on the inclusion criteria.
From the meta-analysis, the results indicate that many miRNAs could intricately orchestrate cellular functions including chemosensitivity/resistance through post-transcriptional control on target gene expression, either canonically or non-canonically. Of the studies included in this meta-analysis, anthracyclines like doxorubicin and epirubicin were predominantly tested in patients/cell lines to study the differential expression of miRNAs followed by tamoxifen in the case of Estrogen Receptor (ER) positive subjects and trastuzumab in the case of Human Epidermal growth factor Receptor (HER) positive subjects. A major limitation in our research is that less than 10% of the 80 papers (6 papers) had direct hazard values that could be utilized for the meta-analysis, reducing the accuracy of the results obtained since only a small fraction of papers were used to give results of the whole, leading to the biasing of the results. There is a possibility of our interpretation being wrong in the context of heterogeneous disease.

Role of miRNAs in Guiding Diagnosis and Prognosis
We extracted the prognosis results of six miRNAs from six different studies. Among the selected miRNAs, two miRNAs (miR200c and miR489) were downregulated and the remaining four miRNAs (miR484, miR4443, miR520h and miR125b) were upregulated. Both downregulated miRNAs were associated with better prognosis; similarly, both miRNAs (miR484 and miR4443) from the overexpressed miRNAs were expressed as better prognosis whereas miR520h and miR125b were associated with poor prognosis.
The overall hazard ratio (95% CI) of the prognostic significance was 0.78 (0.508-1.100) at a p-value of 0.140 which was analysed by random-effect model. This overall combined sized effect estimate indicates that the miRNAs decreased the likelihood of death of breast cancer patients by 22%. This means an HR value >1 indicates an increased risk of breast cancer survival whereas an HR <1 indicates a decreased risk of breast cancer patient survival. The Z-value of the overall effect size was −1.476. The individual overall hazard ratios (95% CIs) of upregulated and downregulated miRNAs were estimated 0.662 (0.403-1.087) and 0.904 (0.487-1.678), respectively. On observing the overall effect size of the individual subgroups, the significant prognosis was associated with a good prognosis, and hence the miRNAs could be considered as better prognostic biomarkers for breast cancer patients.
The Z-value of upregulated and downregulated miRNAs for the null hypothesis test (the mean risk ratio of which is 1.0) were −1.636 and −0.319, respectively. Both the differently expressed miRNA subgroups were associated with lower risk of death in breast cancer patients and hence we cannot accept the null hypothesis that the risk is lower in both differently expressed miRNAs. Similar to our study, two other studies have studied the subgroup analysis of higher and lower expressed miRNAs in meta-analysis studies of the prognosis of melanoma and nasopharyngeal carcinoma patient survival. Those studies demonstrated different risk levels among the subgroups, whereas in our study both subgroups exhibited better prognosis for cancer patients. More studies are required to obtain better prognostic significance of miRNAs in breast cancer patients [147].

Current Challenges
Systematic reviews and meta-analytic studies face a number of challenges when investigating the theragnostic relationship between miRNA and chemotherapeutic response in breast cancer. The primary limiting factor for detailed analysis and clinically applicable insights/results is the scarcity of data. The literature in this specific niche of breast cancer treatment is sparse, with few high-quality studies being available for comparison and analysis. This challenge is exacerbated by the lack of homogeneity between similar studies. The variance in study parameters and the methodology makes assessment difficult by introducing uncertainty in the reliability of the results. Furthermore, a large number of studies have explored this topic via the use of in-vitro models, which cannot be directly applied to clinical theragnostics. The lack of well-documented, large-scale, patient-based clinical studies is a significant challenge faced by this study. Furthermore, the mechanisms of miRNA and chemotherapeutic response are not currently understood in detail, requiring further assessment in the future if meta-analytic studies are to provide conclusions viable for application in the clinical sphere.
The strengths of our paper include its large set of research papers, varied results in terms of miRNAs and pathways that show a change in function in cancerous cells. The result of this exhaustive analysis has provided us with a large number of miRNAs that can be focused on for prognostic or diagnostic purposes. Many miRNAs play a role in regulating many vital cellular pathways, and these regulations are observed to be significantly potentiated or deregulated during treatment with chemotherapeutics. A single miRNA can regulate multiple genes, and this regulation down the cascade can affect many pathways. Many reports have independently observed several genes or pathways as targets of many miRNAs. Of those, the treatment of doxorubicin has been frequently observed to affect the PTEN/Akt and MAPK signalling pathways, and increases chemoresistance (Table 3). In the case of miRNA 21 which is also an oncomiR, treatment with Fulvestrant; Selective Estrogen Receptor Degrader (SERD) or trastuzumab (HER2 antagonist) leads to downregulation, affecting the EMT. Whereas treatment with tamoxifen; Selective Estrogen Receptor Modulators (SERM) downregulates the expression of miRNA-21 via estrogen-dependent functions, leading to chemosensitivity.
In case of miRNAs 221 and 222, the treatment with fulvestrant, doxorubicin or trastuzumab also leads to the downregulation with increased expression of ABC transporters. The treatment with Paclitaxel leads to the downregulation of miRNA 320a with downregulation of TRPC5, NFATC3 and the FTS-1 genes, ultimately causing chemoresistance. miRNA 125b is upregulated when treated with tamoxifen, letrozole, anastrazole or fulvestrant due to its interaction with the Akt/mTOR pathway, leading to chemoresistance. The same pattern is observed when treatment of 5-FU, Paclitaxel and cyclophosphamide is applied, which affects the EMT pathway; or when 5-FU is used, which affects the transcription factor E2F3.
miRNAs Let-7, 181a and 145 are also majorly downregulated when treated with drugs like doxorubicin, tamoxifen, or epirubicin, with increases in chemosensitivity. Thus, myriad miRNAs take centre stage in the search for theragnostic miRNAs indicating drug resistance. However, Our study has tried its best to bridge the gaps, and serves as a benchmark for further clinical studies in personalized treatment research.