Clinical Theragnostic Potential of Diverse miRNA Expressions in Prostate Cancer: A Systematic Review and Meta-Analysis

Background: Prostate cancer (PrC) is the second-most frequent cancer in men, its incidence is emerging globally and is the fifth leading cause of death worldwide. While diagnosis and prognosis of PrC have been studied well, the associated therapeutic biomarkers have not yet been investigated comprehensively. This systematic review and meta-analysis aim to evaluate the theragnostic effects of microRNA expressions on chemoresistance in prostate cancer and to analyse the utility of miRNAs as clinical theragnostic biomarkers. Methods: A systematic literature search for studies reporting miRNA expressions and their role in chemoresistance in PrC published until 2018 was collected from bibliographic databases. The evaluation of data was performed as per PRISMA guidelines for systematic review and meta-analysis. Meta-analysis was performed using a random-effects model using Comprehensive Meta-Analysis (CMA) software. Heterogeneity between studies was analysed using Cochran’s Q test, I2 and the Tau statistic. Quality assessment of the studies was performed using the Newcastle–Ottawa Scale (NOS) for the methodological assessment of cohort studies. Publication bias was assessed using Egger’s bias indicator test, Orwin and classic fail-safe N test, Begg and Mazumdar rank collection test, and Duval and Tweedie’s trim and fill methods. Findings: Out of 2909 studies retrieved, 79 studies were shortlisted and reviewed. A total of 17 studies met our eligibility criteria, from which 779 PrC patients and 17 chemotherapy drugs were examined, including docetaxel and paclitaxel. The majority of the drug regulatory genes reported were involved in cell survival, angiogenesis and cell proliferation pathways. We studied 42 miRNAs across all studies, out of which two miRNAs were found to be influencing chemosensitivity, while 21 were involved in chemoresistance. However, the remaining 19 miRNAs did not appear to have any theragnostic effects. Besides, the prognostic impact of the miRNAs was evaluated and had a pooled HR value of 1.960 with 95% CI (1.377–2.791). Interpretation: The observation of the current study depicts the significance of miRNA expression as a theragnostic biomarker in medical oncology. This review suggests the involvement of specific miRNAs as predictors of chemoresistance and sensitivity in PrC. Hence, the current systematic review and meta-analysis provide insight on the use of miRNA as PrC biomarkers, which can be harnessed as molecular candidates for therapeutic targeting.

chemosensitivity of PrC through regulation of Bmi-1 [31]. As more studies emerge involving miRNAs, it is has been recently observed that the basal cell layer from which the PrC emulates possesses progenitors or adult stem cells [32]. Though the presence of the stem cells themselves is debated in the prostate, the origin of miRNA and the presence of stem cells opens the opportunity for other research approaches in biomarker studies yet to be fully elucidated.
The challenge in choosing a therapy for prostate cancer is the inconsistent levels of PSA in the serum, due to which alternative markers are essential for firm establishment of the correct treatment [33]. miRNAs are reliable predictors, and numerous studies have developed and validated the use of miRNAs as a potential indicator of prostate cancer [34].
Thus, there is emerging evidence establishing the importance of miRNA expression in the development of drug resistance in PrC [24]. However, the effects of miRNA expression on chemoresistance and sensitivity have not been comprehensively studied in the form of a systematic review and meta-analysis. Therefore, this systematic review and meta-analysis aimed to investigate the regulatory role of miRNA in prostate cellular processes and their involvements in the development of chemoresistance in PrC.

Search Strategy and Study Selection
PubMed and Science Direct were used to identify relevant literature published until December 2018. The search strategy aimed at collecting articles relating to miRNA, chemoresistance and prostate cancer. The search terms used were "miRNA or microRNA" AND "chemoresistance" AND "prostate cancer". Only studies that were published in English or had official English translations were included in this study. Manual screening of the reference lists of eligible studies was performed to identify additional relevant articles. Full-text articles were scrutinised after initial screening by titles and abstracts. The final selection was based on predefined inclusion and exclusion criteria. The corresponding authors were contacted for collecting pertinent data if it was found missing in the full texts of the articles. Duplicates were removed, and the study was excluded if it fell within the exclusion criteria (Table S1).

Selection Criteria
All retrieved articles were evaluated by the reviewers (R.J., S.B., M.R.R.) for selection based on the following criteria:

Inclusion Criteria
Our primary inclusion criteria were the studies analysing the theragnostic effect of miRNA expressions in both prostate cancer patients and cell lines.
Inclusion criteria were: • Studies reporting chemoresistance in PrC.

•
Studies reporting a miRNA profiling platform. • Studies on drug regulatory genes or pathways involved in chemoresistance or sensitivity. • Studies reporting miRNA expressions analysis using in-vitro assays on chemoresistance.

Exclusion Criteria
• Studies published in languages other than English. • Letters to the editor, case studies, review articles and studies performed only in PrC patients or in vitro.

•
Studies using PrC patients' information from GenBank datasets.
Disagreements between reviewers were resolved by discussion and consultation with the corresponding author and fifth reviewer.

Data Extraction and Analysis
Studies that complied with the selection criteria were used for extracting study data. Data from the included studies were obtained by S.B., M.R.R., and P.S. and cross-checked by R.J. Corresponding authors of selected articles were contacted for further clarifications and supplementary information if required. The following list of data items was extracted from the full-text articles and supplementary information and recorded in an MS Excel (master sheet) data extraction form, designed based on PRISMA guidelines [35]; i.
First author and year of publication ii. Country iii. Patients' origin iv. Ethnicity v.
Number of samples vi. Cell lines vii. Resistant cell lines to chemotherapy viii. miRNA(s) involved ix. miRNA profiling platform x.
Drug information xi. Molecular pathways or gene associated

Quality Assessment
The Dutch Cochrane Centre's Meta-Analysis of Observational Studies in Epidemiology (MOOSE) guidelines [36] were used to assess the quality of the included studies as adopted in previous studies [37,38]. All the criteria must have been mentioned in the main text, supplementary information or later provided by the corresponding authors to qualify for systematic review. In addition to that, the quality assessment of the included studies was performed using the Newcastle-Ottawa Scale (NOS) for the methodological assessment of cohort studies.

Publication Bias
Publication bias is an integral part of a systematic review and meta-analysis [39,40]. The inverted funnel plot depicts the publication bias. Publication bias was quantified using Egger's bias indicator test [41], Orwin [42] and classic fail-safe N test, Begg and Mazumdar rank collection test, and Duval and Tweedie's trim and fill calculation [43].

Meta-Analysis
Comprehensive Meta-Analysis (CMA) software 3.0 was used to analyse the pooled hazard ratio (HR) with 95% confidence interval (CI). In case of a lack of between-study heterogeneity, fixed model effects were used, and if not, random model effects were employed [44][45][46][47]. Possible influences such as number of patients, year of publication, study period, study locations, type of studies and diagnostic procedures were investigated for heterogeneity using Cochran's Q test and Higgins I-squared statistic [48]. The statistic tau squared was used to study the variance between the studies and where it incorporates a threshold effect [39]. A Q test was used to differentiate between the observed effect and fixed effect model by summing up and squaring their differences [49].

Search Strategy and Study Selection
Initial search identified 2909 relevant studies from PubMed (n = 179) and ScienceDirect (n = 2730). After a thorough screening, 2830 articles were removed for being duplicates, irrelevant, reviews, case studies and letters to the editor. Screening based on inclusion criteria was used to narrow down to 79 potentially eligible studies, and further screening based on exclusion criteria resulted in 17 articles. Figure 1 depicts a flowchart describing our selection process.

Search Strategy and Study Selection
Initial search identified 2909 relevant studies from PubMed (n = 179) and ScienceDirect (n = 2730). After a thorough screening, 2830 articles were removed for being duplicates, irrelevant, reviews, case studies and letters to the editor. Screening based on inclusion criteria was used to narrow down to 79 potentially eligible studies, and further screening based on exclusion criteria resulted in 17 articles. Figure 1 depicts a flowchart describing our selection process.
We identified 17 studies involving 779 PrC patients eligible for the systematic review. Table 1 shows the main characteristics of the included studies for the systematic review and meta-analysis. The included studies were conducted between 2005 and 2015. The majority of studies were performed in China (n = 11) followed by USA (n = 4) and one each from Australia and Austria. The two most preferred chemotherapy agents were docetaxel and paclitaxel.  We identified 17 studies involving 779 PrC patients eligible for the systematic review. Table 1 shows the main characteristics of the included studies for the systematic review and meta-analysis. The included studies were conducted between 2005 and 2015. The majority of studies were performed in China (n = 11) followed by USA (n = 4) and one each from Australia and Austria. The two most preferred chemotherapy agents were docetaxel and paclitaxel.

In-Vitro Assays
The common in-vitro assay types and related number of studies are represented in Figure 2A, and the cell line types in Figure 2B. A total of 14 different cell lines were used in the 17 included studies, of which PC3 and DU145 were the most commonly used (in 12 and 8 studies, respectively). The highest number of cell lines used in a single study was 5 [12]. The in-vitro and in-vivo assay information collected from the included studies showed the use of -RT-PCR, luciferase assay, western blotting, chemotherapy sensitivity assay, apoptotic assay, cell viability assay, cell migration, cell proliferation, immunohistochemistry, chromatin immuno-precipitation (ChIP) assay, clonogenic assay, spheroid assay and caspase assay to determine miRNA expression and activity.

In-Vitro Assays
The common in-vitro assay types and related number of studies are represented in Figure 2A, and the cell line types in Figure 2B. A total of 14 different cell lines were used in the 17 included studies, of which PC3 and DU145 were the most commonly used (in 12 and 8 studies, respectively). The highest number of cell lines used in a single study was 5 [12]. The in-vitro and in-vivo assay information collected from the included studies showed the use of -RT-PCR, luciferase assay, western blotting, chemotherapy sensitivity assay, apoptotic assay, cell viability assay, cell migration, cell proliferation, immunohistochemistry, chromatin immuno-precipitation (ChIP) assay, clonogenic assay, spheroid assay and caspase assay to determine miRNA expression and activity.

Drug Regulatory Pathways for miRNA-Mediated Chemosensitivity and Chemoresistance
The miRNA-mediated chemoresistance pathways are represented in Figure 3. Of the 17 articles, 14 different pathways and their associated genes were elaborated upon in the individual studies.

Drug Regulatory Pathways for miRNA-Mediated Chemosensitivity and Chemoresistance
The miRNA-mediated chemoresistance pathways are represented in Figure 3. Of the 17 articles, 14 different pathways and their associated genes were elaborated upon in the individual studies. Seven pathways were described as leading to cell survival and six pathways were assessed to be involved in cell differentiation and proliferation, while one was linked to angiogenesis.
Cancers 2020, 12, x 9 of 20 Seven pathways were described as leading to cell survival and six pathways were assessed to be involved in cell differentiation and proliferation, while one was linked to angiogenesis.

Association Between miRNAs and Drug Regulatory Pathways of Chemoresistance
miRNA-34a, 200b, 200c and 205 were studied in two studies each. It was observed that miRNA-34a was downregulated on treatment in two studies (azacytidine, topotecan and doxorubicin; paclitaxel and cyclopamine) and was found to influence the AMPK/mTOR and Hedgehog signalling pathways thereby inducing chemosensitivity and resistance, respectively. PrC treatment with docetaxel was associated with downregulation of miRNA-200b through activation of the BMI-1 gene, contributing to chemoresistance. In another report, miRNA-200c was found to be downregulated during the treatment of docetaxel, paclitaxel and cyclopamine, which in turn triggered the E-cadherin and Hedgehog pathway leading to chemoresistance. Finally, miRNA-205 downregulation was observed on the treatment of docetaxel and DZNep resulting in chemoresistance mediated through EZH2 and E-cadherin gene ( Table 2 and Table 3).

Association between miRNAs and Drug Regulatory Pathways of Chemoresistance
miRNA-34a, 200b, 200c and 205 were studied in two studies each. It was observed that miRNA-34a was downregulated on treatment in two studies (azacytidine, topotecan and doxorubicin; paclitaxel and cyclopamine) and was found to influence the AMPK/mTOR and Hedgehog signalling pathways thereby inducing chemosensitivity and resistance, respectively. PrC treatment with docetaxel was associated with downregulation of miRNA-200b through activation of the BMI-1 gene, contributing to chemoresistance. In another report, miRNA-200c was found to be downregulated during the treatment of docetaxel, paclitaxel and cyclopamine, which in turn triggered the E-cadherin and Hedgehog pathway leading to chemoresistance. Finally, miRNA-205 downregulation was observed on the treatment of docetaxel and DZNep resulting in chemoresistance mediated through EZH2 and E-cadherin gene (Tables 2 and 3).  The association between miRNA expressions and patient survival was analysed using HR and 95% CI values through meta-analysis. High expression of miR-132 (HR = 1.9; 95% CI = 1.1-3.2) and low expression of miR-20a (HR = 1.8; 95% CI = 1-3.  The association between miRNA expressions and patient survival was analysed using HR and 95% CI values through meta-analysis. High expression of miR-132 (HR = 1.9; 95% CI = 1.1-3.2) and low expression of miR-20a (HR = 1.8; 95% CI = 1-3.

Publication Bias
The funnel plot represented in Figure 5 is slightly asymmetrical across the studies, which indicates the presence of publication bias. The funnel plots from Figures 5 and 6 have been constructed using the software 'Comprehensive Meta-Analysis Software' (version 3.3.070, USA). Each funnel plot was constructed alongside the forest plot using HR values and 95% CI for each cohort. Each point in the funnel plot represents an individual cohort. If these points were symmetrically present across the regression line, it would indicate the lack of any publication bias in

Publication Bias
The funnel plot represented in Figure 5 is slightly asymmetrical across the studies, which indicates the presence of publication bias. The funnel plots from Figures 5 and 6 have been constructed using the software 'Comprehensive Meta-Analysis Software' (version 3.3.070, USA). Each funnel plot was constructed alongside the forest plot using HR values and 95% CI for each cohort. Each point in the funnel plot represents an individual cohort. If these points were symmetrically present across the regression line, it would indicate the lack of any publication bias in the conducted meta-analysis. Whereas, in the current study, the points in the funnel plot are not symmetric, indicating the existence of bias.
Cancers 2020, 12, x 12 of 20 the conducted meta-analysis. Whereas, in the current study, the points in the funnel plot are not symmetric, indicating the existence of bias.

Classic Fail-Safe N
This meta-analysis incorporates data from nine studies, which yield a z-value of 6.82789 and a corresponding two-tailed p-value of 0.00001. The fail-safe N is 101. This means that we would need to locate and include 101 'null' studies for the combined two-tailed p-value to exceed 0.050. This could alternatively be expressed that there would be 11.2 missing studies for every observed study for the effect to be nullified.

Orwin Fail-Safe N
Here, the hazard ratio in observed studies is 1.96, which did not fall between the mean hazard ratio in the missing studies, so we could not calculate the Orwin fail-safe N. the conducted meta-analysis. Whereas, in the current study, the points in the funnel plot are not symmetric, indicating the existence of bias.

Classic Fail-Safe N
This meta-analysis incorporates data from nine studies, which yield a z-value of 6.82789 and a corresponding two-tailed p-value of 0.00001. The fail-safe N is 101. This means that we would need to locate and include 101 'null' studies for the combined two-tailed p-value to exceed 0.050. This could alternatively be expressed that there would be 11.2 missing studies for every observed study for the effect to be nullified.

Orwin Fail-Safe N
Here, the hazard ratio in observed studies is 1.96, which did not fall between the mean hazard ratio in the missing studies, so we could not calculate the Orwin fail-safe N.

Classic Fail-Safe N
This meta-analysis incorporates data from nine studies, which yield a z-value of 6.82789 and a corresponding two-tailed p-value of 0.00001. The fail-safe N is 101. This means that we would need to locate and include 101 'null' studies for the combined two-tailed p-value to exceed 0.050. This could alternatively be expressed that there would be 11.2 missing studies for every observed study for the effect to be nullified.

Orwin Fail-Safe N
Here, the hazard ratio in observed studies is 1.96, which did not fall between the mean hazard ratio in the missing studies, so we could not calculate the Orwin fail-safe N.

Begg and Mazumdar Rank Correlation Test
In this case, Kendall's tau b (corrected for ties, if any) is 0.41667, with a one-tailed p-value (recommended) of 0.05893 or a two-tailed p-value of 0.11785 (based on continuity-corrected normal approximation).

Duval and Tweedie's Trim and Fill
This method suggests that three studies are missing ( Figure 6 Publication bias analysis of the included studies was conducted to study the effect of publication bias on the results of this study. Figure 6 shows the funnel plot results with imputed studies. The symmetric nature of the plot denotes the presence of no bias. From the funnel plot, it is obvious that the smaller included studies place towards the bottom of the funnel plot, and the more extensive studies look towards the top of the graph, with clustering near the mean effect size. Large studies appear outside the funnel and tend to cluster on one side of the funnel plot. Smaller studies appear toward the top of the graph, since there is more sampling variation in effect size estimates in the smaller studies, which will be dispersed across a range of values.

Discussion
The drug resistance mechanisms in PrC are being extensively studied, as indicated by the emerging studies [20,[59][60][61][62]. The purpose of this systematic review is to investigate and evaluate the miRNA biomarkers as potential predictors in chemoresistance/sensitivity in PrC, and their association with different drug-regulatory genetic pathways.
The previous meta-analysis on prostate cancer was carried out deciphering the role of miRNAs targeting the androgen receptor [63]. miRNA-21 is downregulated in our study, similar to in another report which demonstrates that the reduction in expression could inhibit the cancer growth, while its overexpression in androgen-independent prostate cancers makes them resistant to androgen ablation [64,65].
The contradicting results of miRNA-375 expression in one of our studies show it is downregulated while another study reported upregulation of this miRNA in serum samples, indicating their role in the development of AIPCs [66]. miRNA-34a, 146a, 205 are downregulated similar to the results stated in the Li and Mahato (2014) review [67]. Synthetic miRNAs or genetic precursors can be used to restore the levels of tumour-suppressive miRNAs, which is known as miRNA replacement therapy [68].
EZH2 plays a key role in cancer progression [83], and its function as a key regulator is studied in several cancers, such as breast [84], prostate [29] and nasopharyngeal carcinomas [85]. Hedgehog functions as a mediator of tumorigenesis, and its signalling is essential in the regulation of cancer and tumorigenicity [86]. Of the numerous existing RAS proteins, KRAS has a crucial role in proto-oncogene activity, and its mutation status has been highly explored for therapeutic response in cancer research [87]. E-cadherin is an invasion-and tumour-suppressor protein [88] and plays a role in the transition of adenoma to carcinoma, and its repressed expression is a poor prognostic indicator in cancer [89,90]. ZEB1 and its two proteins are studied widely as they promote the epithelial-mesenchymal transition in cancer and are direct repressors of E-cadherin [91]. TCF7 activates transcription through the Wnt/β-catenin pathway and is observed to be involved in the growth and metastasis of cancer [92,93].
PC3 and DU145 cell lines are being used for evaluating transcriptional activity [94], in cancer studies [95][96][97] and for evaluating genetic activities [98,99], which is examined in most of the authors' included studies. The results of this meta-analysis show a significant pooled effect when correlating drug-resistance-related miRNAs to patient survival, thereby reinforcing the possibility of use of miRNA as prognostic markers. However, we also have to take into consideration that only one study reported the HR and 95% confidence interval.

Strengths of Our Study
This systematic review and meta-analysis follow appropriate best practice research and statistical guidelines, and the following points accentuate this approach. Initially, the studies chosen were selected from a global arena, where there is a diversity in the selection of patients and their outcomes; secondly our selection criteria (which follow the PRISMA guidelines) help us to engage and have access to an extensive global literature database. An analysis of the publication bias was minimised to minimise the heterogeneity between the studies selected in our meta-analysis.
This comparison will help future researchers to evaluate and publish articles from separate patients, which will collectively add to the knowledge base around miRNA prognosis in PrC patients. From this point of view, this study brings out more value for future researchers and technicians regarding predicting miRNA as a valuable prognostic biomarker. The use of subgroup analysis also took into consideration demographic characteristics and repeated miRNAs to provide a better understanding of the survival outcomes of PrC patients. Additionally, the authors believe that this is the foremost systematic review and meta-analysis study on the prognostic utility of miRNAs in PrC patients.

Limitations of Our Study
The majority of the included studies in our meta-analysis were on patients from Asian and Caucasian backgrounds, which might limit the applicability of the results. From the prognostic point of view of these specific miRNAs, only one study reported the HR and 95% CI value, which is a major limitation, as the rest of the values were estimated from KM curves. Finally, potential biases and confounders cannot be avoided completely, since all studies included had observational designs. Our hallmarks of the PrC provide an overview of the miRNAs involved in drug resistance. The identified miRNAs can be used as diagnostic, prognostic and therapeutic biomarkers in PrC.

Future Work
It is expected that analysing the graph based on the correlation between pairs of miRNAs using cluster editing will reveal other drug regulatory pathways that are not detected yet. This is because clusters (cliques) of mutually pairwise correlated miRNAs represent a strong indication of potential chemoresistance pathways. In fact, clustering can be used as a verification phase or to possibly disclose other pathways that are not detected yet.

Conclusions
There is extensive anticipation regarding the use of miRNAs as a possible predictor of chemoresistance in prostate cancer as the available markers are unreliable. There are a significant number of published studies on the evaluation of miRNA regulation in prostate cancer. Our systematic review and meta-analyses from recent clinical evidence demonstrate that miRNAs facilitate drug resistance and sensitivity in prostate cancer patients. The hallmarks of specific miRNAs in prostate cancer as highlighted by our study might provide insights into the regulation of miRNAs and the genes associated with the process of drug resistance, thereby helping future clinicians and researchers in designing efficient clinical trials and in-vitro studies. We have provided a list of miRNAs and their respective pathway targets for routine therapeutic purposes. Further research is required to highlight which specific miRNAs may be intricately involved in chemoresistance and sensitivity.
Author Contributions: R.J. was involved in the design, data analysis, interpretation and conceptualisation, supervised the study and contributed to all sections of the manuscript. G.R., S.K., K.S.T., S.B., M.R.R., S.K.G., H.C.C., F.N.A.-K. and P.S. provided input into the study design, supervision, ensured the absence of errors and arbitrated in case of disagreement. R.J., M.R.R. and S.K. performed the literature searches, data extraction, data interpretation and contributed to the analysis. All authors have read and agreed to the published version of the manuscript.
Funding: There was no funding source for this study.