The miR-96/RARγ signaling axis governs androgen signaling and prostate cancer progression

RAR (RARG) expression is commonly reduced in prostate cancer (PCa). Modulating RARγ levels, not retinoid ligand, had the biggest impact on prostate cell proliferation and gene expression. Genomic binding of the non-liganded, apo, RARγ was significantly enriched at active enhancers, associated with AR, and RARγ knockdown governed the AR capacity to regulate cell differentiation and gene-regulation. Altered RARγ target genes expression in TCGA significantly associated with more aggressive PCa. RARγ downregulation was explained by a stark and common increase in miR-96 in PCa cell and animal models, and human PCa. Biochemical approaches confirmed that miR-96 directly regulates RARG expression and function. Capture of the miR-96 targetome by biotin-miRNA pulldown identified a RARγ-centric network significantly associated with more aggressive PCa and worse disease free survival (hazard ratio 2.23, 95% CI 1.58 to 2.88, p=0.015). In summary, miR-96 targets an RARγ network to govern AR signaling and its disruption is a cancer-driver. STATEMENT OF SIGNIFICANCE We identified that miR-96 targets a RARγ-network, which in turn regulates AR, PCa progression and disease outcome. These findings occur independently of retinoid ligand and reveal how miRNA govern nuclear receptor functions, and can be exploited to identify aggressive prostate cancer at an early stage.


INTRODUCTION
Members of the nuclear hormone receptor (NR) superfamily are ubiquitously expressed across tissues and govern cell fate decisions. One NR, the androgen receptor (AR/NR3C4), is a key regulator of growth and differentiation in the prostate gland (1). Genomic approaches in prostate cancer (PCa) have identified that the capacity of the AR becomes skewed with disease progression (2). A normal component of AR signaling is to drive terminal differentiation of luminal epithelial cells and the phenotypic consequences of altering AR signaling capacity is, in part, to disrupt this function (3). The disruption, or re-wiring, of AR action changes both the receptor sensitivity, defined as the magnitude of the transcriptional response, and capacity, defined as the selection of gene networks governed. Although the AR is a pharmacological target in later PCa stages, its disruption at early stage disease is more nuanced (4).
In fact, multiple NRs are expressed in the normal prostate and are disrupted in PCa. NR actions are integrated by shared genomic binding regions, shared co-factors and the co-regulation of ligand availability (5,6). Non-coding RNAs, including miRNA and lncRNA target NRs, their cofactors and target genes to also exert control of signaling (7,8). Combined, these different aspects of NR regulation generate sensitivity.
As a route to identify how NR networks are disrupted in cancer we undertook a pan-cancer analysis of the NR superfamily in The Cancer Genome Atlas (TCGA). Significantly distinct NR profiles were revealed within each tumor type (9,10). Specifically, in the Taylor et al MSKCC (11) and TCGA-PRAD PCa cohorts(12) the retinoic acid receptor gamma (NR1B3/RARG, encodes RARg) and glucocorticoid receptor (NR3C1/GR) were significantly and uniquely down-regulated.
By contrast, the AR was not significantly altered in either cohort. There was only one RARG mutation and relatively few CNVs detected at the RARG locus across these approximately 600 PCa samples.
To better understand the consequences and causes of reduced RARg in prostate cells we inspected the impact of reduced RARg expression on cell phenotypes and gene expression, in the apo (non-ligand stimulated) and holo (ligand-stimulated) states, and found a substantial regulatory function specific to the apo-RARg condition. Combined with analyses of the RARg cistrome, these findings revealed that apo-RARg significantly associates with active gene enhancers, and impacts other transcription factor functions, including those of the AR. Testing how the RARg governed the capacity and sensitivity of AR was undertaken by androgendependent transcriptomic analyses in cells with altered RARg expression.
Furthermore we revealed that miRNA-96 regulates expression of RARg and several known RARg co-factors, and that miR-96 is commonly elevated in PCa associated with disease progression. Finally, across multiple PCa cohorts the miR-96 targetome and RARg-dependent transcriptome associated with aggressive PCa and disease progression.
Together these findings support several innovative concepts. Firstly, that miR-96 potently regulates RARg because it co-targets a number of RARg-interacting co-factors. Secondly, the major gene regulatory (and disease associated) functions of RARg are largely independent of retinoid activation. Thirdly, regulation of this RARg network is a potent disease driver because it is a major regulator of AR sensitivity and capacity.

Reduced RARg expression promotes cell proliferation and widely alters gene expression in non-malignant and malignant prostate cell models.
To test if reduced RARg levels are pro-tumorigenic, and to what extent retinoid responses are RARg dependent, we knocked-down RARg levels in non-malignant prostate epithelial cells  2 fold change, Benjamini-Hochberg adjusted p-value < 0.05) were associated with apo RARg regulation, whereas only 237 genes were responsive to CD437 (10 nM, 24 hr) in a RARg dependent manner. In both cases the proportion of up-regulated genes was significantly higher. For example, proportionately more genes were up-(N=338 gene) than down-regulated (N=265) for the apo-transcriptome (p<0.001).
Amongst these, 196 were associated uniquely with apo function and 201 with holo, while 38 were common between categories. Frequency mining of common terms combined with hypergeometric testing established largely distinct enrichment of meta-groups (e.g. terms associated with NF-kB) between the apo and holo RARg states in both RWPE1 and LNCaP cells ( Figure 1D). In RWPE1 cells apo transcriptome was significantly enriched for terms associated with NF-kB and histone deacetylase (HDAC) function ( Figure 1E) whereas the holo transcriptome was enriched for terms related to Tretinoin (the commercial name for ATRA) ( Figure 1F Reflecting the enrichment of AR terms in the apo transcriptome in LNCaP, we used a previously compiled AR target gene panel (12) to demonstrate that expression of these genes alone significantly distinguished LNCaP-shRARG cells from controls, in a CD437-independent manner ( Figure 1H). Furthermore, knockdown of RARg in LNCaP also altered the DHT-induced expression changes in several key androgen target genes (i.e. KLK3, TMPRSS2) Figure 3H).

(Supplementary
In total, these findings indicate that apo RARg function is substantial and intertwined with other transcription factors, notably the AR and NF-kB, and functionally-independent of the holo RARg.
The apo RARg cistrome significantly overlaps with active enhancers, AR cistromes and associates with aggressive PCa.
A clear and significant overlap between the apo RARg cistrome with H3K27ac, H3K4me1 and was identified associated with enhancer status (Figure 2A, upper) and TSS of expressed target genes (Figure 2A, lower); specifically, 829 RARg binding sites overlapped with combined H3K27ac, H3K4me1 and DNAse sensitivity profiles suggesting that the apo RARg is commonly found in open chromatin active enhancer regions ( Figure 2B). Reflecting the transcriptome data of RARg function overlapping with AR and NF-kB functions (Figure 1D), significant overlaps were also identified between RARg binding and both the apo and holo AR (16), and the DHTdependent and TNF-stimulated p65 cistrome (2) (Figure 2C). This was also supported by the ChromHMM track showing binding of RARg at sites of poised and flanking transcription (e.g. Figure 2D).
Next, we sought to dissect transcriptome-cistrome relationships. Candidate level-relationships were identified. Apo RARg binds at the KRT15 locus in active enhancer sites, and its transcription is significantly altered by the knockdown of RARg (Figure 2D, E). To test the genome-wide level of significance transcriptome-cistrome relationships both apo and holo RARg cistromes were annotated to within 7.5 kb of known gene TSS. The mean expression of these cistrome proximal genes were then compared with that of genes found to be RARg dependent from expression studies using a random sampling with replacement, or bootstrapping, approach (Supplementary Table 1). Apo and holo RARg cistromes were significantly associated with genes whose expression changed upon loss of RARg, supporting a functional relationship between RARg binding and expression control ( Figure 2F).
In the TCGA-PRAD and MSKCC cohorts the RARg-annotated genes significantly positively correlated with RARg (Supplementary Figure 5A,B) filtering these RARg cistrome genes (altered by greater than 2 Z-scores in more 35% tumors relative to normal tissue) identified 58 genes. Expression of these genes stratified patients ( Figure 2G) and distinguished a cluster of high Gleason grade tumors (adjusting for age) (p-value = 0.038). We also reasoned that higher Gleason grade tumors were associated with worse outcome and therefore we tested relationships between the expression of individual genes and disease free survival. After FDR correction, four of these genes were individually significantly associated with disease free survival; the steroidogenic enzyme CYP11A ( Figure 2H); LRG6, a WNT regulator and implicated in breast cancer; PRR7, a central regulator of CLOCK;.

Reduced RARg expression alters AR signaling capacity and sensitivity.
The cellular phenotype, transcriptome and cistrome data support the concept that apo RARg actions are a substantial portion of all RARg genomic functions and intertwined with other TFs including the AR. Therefore, we sought to test the genome-wide impact of RARg expression on AR function directly in non-malignant prostate epithelial HPr1AR cells. These cells constitutively express AR, and undergo androgen-induced differentiation from a basal-like towards a more luminal-like phenotype (17). RARg knockdown reduced RARg levels 60-80%, which was not altered by DHT (Supplementary Figure 6A-D).
The anti-proliferative response induced by DHT was significantly dampened in RARg knockdown HPr1AR cells ( Figure 3A). We therefore investigated whether the apo RARg  Figure 7D).
Together, this suggest that RARg expression governs AR function in normal prostate cells independently of exogenous retinoid ligand, and that RARg loss dampens both phenotypic and transcriptomic androgen responses, which manifest as reduced luminal differentiation and reduced antiproliferative signaling.

Elevated miR-96 drives reduced RARg expression and associates with aggressive prostate cancer.
The frequent downregulation of RARG reflected neither mutation nor copy number variation (10) and therefore we considered epigenetic mechanisms, but found no evidence for altered DNA methylation in the TCGA-PRAD cohort (Supplementary Figure 8). Therefore, we considered miRNA, and used in silico prediction tools (18) to define a cohort of miRNAs that target the most common previously identified downregulated NRs (i.e. RARG, GR) in the TCGA-PRAD and MSKCC cohorts (10). Together, these findings support the concept that elevated miR-96 targets and suppresses RARG expression and function, an interaction which occurs early in PCa progression and which has pro-tumorigenic properties.

MiR-96 targets a network of RARg interacting co-factors.
While these data strongly support miR-96 targeting of RARg, other studies have identified additional targets, including the pioneer factor FOXO1 (20), which have biological consequence in prostate cells. Therefore, to reveal all miR-96 targets (the miR-96 targetome), we undertook a biotin-miRNA (bi-miR) pulldown approach coupled with microarray analyses (21). Biotin labelling of miR-96 mimics did not interfere with either transfection or knockdown of target gene expression (Supplementary Figure 15A-B), and was able to capture known (e.g. RARG and FOXO1) and predicted (TBL1X) miR-96 targets (Supplementary Figure 15C). PCA revealed strong separation of experimental groups (Supplementary Figure 15D).
The bi-miR-96 pulldown revealed 111 and 389 miR-96 targets in RWPE1 and LNCaP cells, respectively, which were largely shared, but also had unique genes ( Figure 5A Interestingly, in LNCaP cells FOXO1 was neither significantly enriched in the bi-miR-96 fraction upon microarray analysis and was only validated by RT-qPCR in RWPE1 cells. Also surprisingly, RARG transcript was not significantly detectable via microarray in either LNCaP or RWPE1 cells, but did validate by RT-qPCR in both cell types. It was revealed that the Illumina HT12v4 Bead Chip array contains only a single exon probe targeting a single RARG isoform. RARG expression in LNCaP cells has been reported (22) suggesting that the microarray finding is a false negative one.
The capacity of miR-96 to suppress target gene expression was also assessed by independent transfections of biotinylated and non-biotinylated miR-96 mimics. In LNCaP cells, 6 out of 7 bi-miR-96 identified targets were significantly downregulated by both miR-96 and bi-miR-96 relative to respective non-targeting controls, while a negative control target CDH1 was not affected ( Figure 5D). Intriguingly, the single transcript not downregulated in this experiment, p27 (encoded by CDKN1B), showed significantly reduced protein level upon miR-96 overexpression, suggesting that in some cases the result of miR-96 binding is translational inhibition not degradation of transcript (Supplementary Figure 15G, H). To assess whether experimentally determined targets had relationships in clinical samples, TCGA-PRAD data were again examined. Genes within the miR-96 targetome showed a significantly more negative correlation with miR-96 expression than the background transcriptome, supporting a functional relationship (p = 5.17e -7 ) ( Figure 5E). Stringent filtering of expression of the miR-96 targetome genes (+/-2 Z scores in 35% of tumors relative to normal prostate tissue) identified 22 altered genes segregated invasive tumors (Pathological N1), after adjusting for age at diagnosis (p-value = 0.041) (Figure 5F). Five of these 22 genes were known to regulate RAR signaling (i.e. TACC1 (23), ZIC2 (24), PRKAR1A (25), EIF4G2 (26), and others were such are ONECUT2 are putative co-regulator (27). Interestingly, the motif for ONCUT2 was enriched in the experimentally derived holo RARg cistrome.

The miR-96 governed RARg network drives aggressive prostate cancer.
Several lines of evidence support the concept that miR-96/RARg is a significant signaling axis in the prostate; the levels alone of apo RARg control gene expression and govern AR function; miR-96 governs the levels of apo RARg and a number of interacting co-factors; the expression of miR-96 and RARg are reciprocal in murine and human tumors; RARg targets and miR-96 targets predict aggressive PCa tumors.
Of the RARg-network genes, TACC1 had the strongest positive correlation with RARg, and only the cross-correlations with either RARg or TACC1 to RARg target genes were significantly different between tumors expressing high or low levels of miR-96 (Supplementary Figure 17).
Therefore we sought to establish the combined significance of miR-96/RARg axis in PCa by segregating tumors in the TCGA-PRAD cohort into low expression of RARg and TACC1, and high miR-96 expression. These were compared to the reciprocal (high RARg and TACC1, low miR-96) ( Figure 6A). Differential expression identified ~1700 genes, which were in turn overlapped with the gene sets developed through the study. Several of the overlaps support the concept of a miR-96/RARg axis. Nearly 200 genes were shared between the apo RARgdependent gene expression and the DHT-regulated and shRNA RARg-dependent genes, and 31 of these significantly overlapped (p=7.1e-6) with the genes identified in TCGA-PRAD ( Figure   6A).
We reasoned that if genes central to the miR-96/RARg axis were critical in PCa, they would be commonly distorted and associate with aggressive disease. Filtering the ~1700 differentially expressed genes in RARg low/miR-96 high tumors (altered by +/-2 Z scores in more than 45% of tumors relative to normal prostate tissue) revealed 47 genes. These genes clustered tumors separated that significantly associated with worse disease free survival (X_REFS_IND) in the TCGA-PRAD cohort (hazard ratio 2.23, 95% CI 1.58 to 2.88, p=0.015), and also clustered high Gleason score tumors (p-val = 0.012) (Figure 6B). Genes were also annotated for whether they were an apo RARg cistrome gene; DHT-regulated RARg-dependent genes; and member of the highest rank GSEA term (LIU_PROSTATE_CANCER_DN; a cohort of genes significantly downregulated in PCa(28)). Five RARg cistrome genes were all down-regulated, including the tumor suppressor MCC (29) and other regulators of cell-fate such as GPX3 (30) and SOX15 (31) and furthermore MCC and SOX15 were also DHT-regulated RARg-dependent genes ( Figure 6C).
These findings support the concept that the miR-96/RARg axis can be detected in PCa cohorts and underscores both its cross talk with AR signaling and its clinical relevance.

DISCUSSION
Developmental roles for RARg have been identified in skin, skeletal and reproductive systems, including the prostate (32). Prostate epithelial squamous metaplasia arises in Rarg-/-mice, due to improper glandular function (32). More broadly, RARg has emerged as an important transcription factor. For example, the Roadmap Epigenome consortium identified RARg as a member of the most highly enriched in transcription factor-enhancer interactions across the human epigenome (33) and others have considered a role for the receptor to govern pluripotency (34). RARG has also been suggested to be a tumor suppressor in keratinocytes (35). Previously, we identified a common RARg down-regulation in the TCGA-PRAD and MSKCC cohorts.
We have now identified that the apo actions of RARg are substantial, and more significant than the impact of retinoid exposure, as seen with our microarray and ChIP studies, but also our analyses of microarray studies in RARg -/murine ES cells and also F9 cells, (36) where similar patterns were observed (Supplementary Figure 18). These findings are also compatible with other receptors such as RARβ being more prominent in the sensing of exogenous ligand in PCa cells (37).
The current study also revealed RARg regulation of AR signaling in PCa. The apo RARg transcriptomic studies revealed enrichment of AR signaling processes, notably in LNCaP, whereas in RWPE1 cells the transcriptome was enriched for ESR function, further supporting RARg functional crosstalk with steroid receptor signaling. Links between RARs and ER signaling which are well established in breast cancer (13).
Genomic apo RARg binding very significantly overlapped with active enhancers (RWPE-1) and AR binding (LNCaP), and also significantly regulated the DHT-AR transcriptome in HPr-1AR cells. RARg loss profoundly altered AR regulated events including DHT-mediated antiproliferative effects (17) and reduced the DHT-dependent transcriptome, notably by restricting the capacity of AR to repress MYC signaling. Recently MYC has been shown to antagonize AR function in prostate cells (38), and the current study suggest that RARg regulates these events, in part through direct binding at the MYC locus. Together these findings supported the concepts that the apo RARg is significantly associated with enhancers, and governing AR functions.
Although miR-34c targets RARG in embryonic stem cells(39), we found no significant relationship between miR-34c and RARg in the current study. Rather we identified miR-96, which has been investigated in various cancer settings including in PCa, but not associated with RARg (20,40). Furthermore, miR-96 is one of the 60 miRNAs whose expression most differentiated PCa clinical groups in TCGA-PRAD cohort. The current study extends these reports of oncogenic actions of miR-96 to identify the RARg network as a major biological and clinically-relevant target.
The inverse relationships between miR-96 and RARg was robust and significant across cell models, murine PCa models and clinical cohorts, and equal or greater than with previously identified targets including FOXO1. Biochemical tests established that miR-96 binds directly to multiple target sites contained within the 3'UTR of RARG and using the bi-miR approach revealed ~400 direct targets enriched for predicted miR-96 target sequences, and identified targets were directly bound and downregulated upon miR-96 overexpression. Together these date strongly support the concept that the miR-96/RARg is a significant and yet under-explored regulator of the AR and drives PCa progression.

Data analyses and integration
All analyses, were undertaken using the R platform for statistical computing (version 3.1.0) (43) (43) and a range of library packages were implemented in Bioconductor (44).

Cell culture and materials.
All cells were maintained in standard conditions in media recommended by ATCC. 2µg/mL puromycin was used for selection of pGIPz-shRARG and BAC-RARG-EGFP containing cell lines.

Cell proliferation and cell cycle analyses.
Bioluminescent detection of cellular ATP (proliferation) and cell-cycle distribution was determined utilizing FACSCalibur TM Flow Cytometer (Becton-Dickinson) and software.(8)

Quantitative real-time reverse transcription-polymerase chain reaction (RT-qPCR) undertaken via Applied Biosystems 7300 Real-Time PCR System (Applied Biosystems), for both TaqMan® and SYBR® Green (SYBR® Green PCR Master Mix (Thermo Fisher Scientific)) applications(8).
Fold changes were determined using the 2 -ΔΔCt method. Significance of experimental comparisons was performed using Student's t-test.

Western Immunoblotting
Western blot analysis of protein expression was undertaken as described previously .

Expression determination in PCa mouse models
Snap frozen tissue and/or RNA from previously harvested normal or malignant prostate tissues of Hi-MYC, PTEN-/-and TRAMP (6,8,10,15,(20)(21)(22)(23)(24)(25) week) models, as well as from agematched wild type (FVB:BL6, C57BL/6) mice, was obtained from the lab of Dr. Barbara Foster at RPCI. Relative expression of Rarg and microRNAs was determined in these tissues by RT-qPCR after normalization to Gusb or RNU6B, respectively. Experiments were performed in technical triplicates, and experiment replicated a total of 5 times.

Stable knockdown of RARg
Knockdown of RARγ in RWPE1, LNCaP and HPr1AR cells was achieved by stable selection after transduction with lentiviral shRNA constructs targeting RARG. Two targeting constructs (V2LHS_239272, V2LHS_239268) and one non-silencing control construct were selected from the V2LHS pGIPZ based lentiviral shRNA library (Thermo Fisher Scientific) for testing. Viral packaging and cellular infection was performed through the RPCI shRNA Resource. All pGIPZ containing cells were subsequently maintained in media supplemented with puromycin (2µg/mL), including during all experiments.

BAC-RARG-EGFP construct (CTD-2644H7) was a generous gift of Dr. Kevin White (University of Chicago). RWPE1 cells were transfected with BAC-RARG-EGFP construct using
Lipofectamine® 3000 and selected with G418 and consistently maintained under antibiotic selection for all subsequent passaging of cells and also during experiments.

Chromatin Immunoprecipitation
ChIP was performed in BAC-RARG-EGFP containing RWPE1 cells in the presence of CD437 (10nM, 2hr) or DMSO as previously described (8). The RARg cistrome was analyzed with
Harvested cell pellets were resuspended in cell lysis buffer (10mM KCl, 1.5mM MgCl2, 10mM Tris-Cl pH 7.5, 5mM DTT, 0.5% Sigma-IGEPAL CA-630) containing SUPERase·In (Ambion) and 1x cOmplete Mini protease inhibitor (Roche) and cleared by centrifugation. 5% of cell lysate was collected to serve as lysate RNA input. Dynabeads® MyOne TM Streptavidin C1 (Thermo Fisher Scientific) were washed 3 times with bead wash buffer (5mM Tris-Cl pH 7.5, 0.5mM EDTA, 1M NaCl), and then blocked (1µg/µL bovine serum albumin, 1µg/µL Yeast tRNA, 50U/mL RNaseOUT) for 2 hr. Resuspended beads were added 1:1 to cell lysate, and mixed for 30 minutes. Bead-lysate mixtures were collected with a magnetic rack, and bead-bi-miR complexes washed a total 3 times with wash buffer. Bead-bi-miR complexes and input control samples were resuspended in water and purified using the Qiagen RNeasy® Mini kit (Qiagen) according to manufacturer's RNA clean-up protocol. To concentrate samples for downstream analyses, eluted RNA was brought up to a total volume of 500µL in H 2 O and filtered through Amicon® Ultra-0.5mL Centrifugal Filters (EMD Millipore) according to manufacturer's instructions. Subsequent amplification and labeling of 50ng of pulldown and input RNA was performed at the Roswell Park Cancer Institute Genomics Core Facility, using the Illumina® TotalPrep RNA Amplification kit, including a 14 hour incubation for the IVT step. Hybridization of cRNA samples onto the Illumina® Human HT-12v4 bead arrays, and successive scanning and raw intensity measurement extraction were also performed at the RPCI Genomics Core Facility.

miRNA prediction determination & TCGA miRNA analysis
To reveal putative NR-targeting miRNA, miRWalk, a comprehensive database on predicted and validate miRNA targets, was employed. If at least 5 out of 9 algorithms positively predicted an interaction, it was considered in subsequent analyses. MiRNA expression was queried in PCa tissue samples and matched normal tissue from TCGA cohort data as previously described (10).
To examine if NR-targeting miRNA expression alterations significantly deviated from what would be expected by chance, bootstrapping approaches were utilized as previously described (10).

Microarray / RNA-seq analyses
Global changes in mRNA, biological triplicate samples per experimental condition were analyzed using Illumina microarray (Illumina HT12v4) or by RNA-seq (limma (46) or DESeq2 (47)). For RNA-seq data, raw sequence reads were aligned to the human genome (hg19) using tophat2, and aligned reads translated to expression counts via featurecounts, followed by a standard DESeq2 pipeline.

Functional annotation of shRARG defined gene sets
Functional annotations were performed using GSEA v3.0 and gene sets from the Molecular signatures database (MSigDB). Specifically, gene sets were compiled to assess enrichment of all BROAD Hallmark pathways, curated pathways (KEGG, BioCarta, Canonical, Reactome, Chemical/Genetic perturbations), and transcription factor motif gene sets. Additionally, several candidate gene sets were included from previous studies, including microarray analyses of HPr1AR cells treated with DHT(48), compiled gene sets previously utilized to encompass androgen response in PCa patients(12), as well as gene sets differentiating basal from luminal prostate epithelial cells (49). In total, 4204 gene sets were queried.
To identify meta groups within enriched (NES > 1.8, FDR q-val < 0.05) gene sets, keywords from gene set identifiers were compiled, and frequency tables determined for all keywords across all gene sets and within enriched gene sets. To account for background frequencies of given terms across gene sets hypergeometric testing was used to determine if the frequency of key words within enriched gene sets was greater than expected.

Validation of miR-96 target sites & Functional annotation
Unbiased assessment of miRNA seed sequence binding motifs in experimentally determined miR-96 target transcript 3'UTR regions was performed using the MSigDB microRNA targets collection.Functional annotation of GO terms enriched in miR-96 target genes was accomplished via DAVID Bioinformatics Resources 6.7 (via BiNGO) and additionally by the topGO package implemented in R. Significantly enriched terms (FDR < 0.05) by both methods were simultaneously visualized using Cytoscape 3.2.0. Survival outcomes were determined using Survival.         (based on lower/upper tertile expression) to generate RARg/TACC1 low , miR-96 high (n = 60) and RARg/TACC1 high , miR-96 low (n = 66) tumors and differential expression undertaken. (B) Filtering the differentially expressed 1728 gene sets as in Figure 2G revealed which were most altered in the TCGA-PRAD cohort. These are represented as a heatmap in which tumor clusters significantly associated, after adjusting for age with higher Gleason Grade (Gleason_score >7).
These clusters also significantly identified patients who experienced treatment failure and worse disease free survival following radical prostatectomy (X_RFS_IND; log-rank test p-val = 0.026).       Identify most significant miR-96-RARg axis genes