Epigenetic Silencing of miRNA-338-5p and miRNA-421 Drives SPINK1-Positive Prostate Cancer

This article is featured in Highlights of This Issue, p. 2677

Prostate cancer is characterized by extensive molecular heterogeneity and varied clinical outcomes (1). Multiple molecular subtypes involving recurrent genetic rearrangements, DNA copy number alterations, and somatic mutations have been associated with this disease (1–3). Majority of these patients harbor gene rearrangements between members of the E26 transformation–specific (ETS) transcription factor family and the androgen-regulated transmembrane protease serine 2 (TMPRSS2), most recurrent (∼50%) being TMPRSS2-ERG, involving the v-ets erythroblastosis virus E26-oncogene homolog (ERG; refs. 3, 4). The TMPRSS2-ERG–encoded ERG transcription factor is known to drive cell invasion and metastases, DNA damage in vitro, and focal precancerous prostatic intraepithelial neoplasia (PIN) lesions in transgenic mice (5, 6).

While TMPRSS2-ERG fusion forms the most frequent molecular subtype, a significant subset of ETS-negative (–) prostate cancer show overexpression of serine peptidase inhibitor, Kazal type-1 (SPINK1) in approximately 10%–15% of the total patients with prostate cancer, a distinct subtype defined by overall higher Gleason score, shorter progression-free survival, and biochemical recurrence (7–9). SPINK1 promotes cell proliferation and invasion through autocrine/paracrine signaling and mediates its oncogenic effects, in part, through EGFR interaction by activating downstream signaling. Monoclonal EGFR antibody administered in SPINK1-positive xenografted mice showed only a marginal decrease in tumor burden, suggesting involvement of EGFR-independent oncogenic pathways (10).

Although, genomic events such as gene rearrangements and somatic mutations constitute most recurrent oncogenic aberrations, many could also be attributed to epigenetic alterations. Earlier studies have shown that aberrant expression of enhancer of zeste homolog 2 (EZH2) owing to genomic loss of miRNA-101 (11) or hypermethylation of miR-26a (12) constitutes a common regulatory mechanism across several solid cancers. EZH2, being the key component of the polycomb-repressive complex 2 (PRC2) mediates trimethylation on the histone 3 lysine 27 (H3K27me3), leading to gene silencing (13). However, phosphorylated form of EZH2 is known to switch its function from polycomb repressor to transcriptional coactivator of androgen receptor in castration-resistant prostate cancers (CRPC; ref. 14). Moreover, recent studies have shown PRC2 epigenetically suppresses the expression of several tumor-suppressive miRNAs such as, miR-181a/b, miR-200b/c, and miR-203, while these miRNAs in turn directly target PRC1 members, namely BMI1 and RING2, thereby reinforcing the repressive molecular circuitry (15).

Although overexpression of SPINK1 forms the second most prevalent and aggressive subtype of prostate cancer (4, 7), the underlying mechanism involved in its upregulation is poorly understood and remains a matter of conjecture. Furthermore, SPINK1 overexpression is not ascribed to chromosomal rearrangement, deletion, or amplification (7), and thus alludes to a possible transcriptional or posttranscriptional regulation. A recent study showed the transcriptional activation of SPINK1 along with gastrointestinal lineage signature genes in patients with CRPC (16). Our study focuses on the posttranscriptional regulation of SPINK1 expression by miR-338-5p and miR-421, which are epigenetically silenced in SPINK1-positive prostate cancer. We also provide evidence that EZH2 acts as an epigenetic switch, thereby promoting transcriptional silencing of miR-338-5p/miR-421 by establishing H3K27me3-repressive marks, thus leading to SPINK1 overexpression. Collectively, our findings suggest potential benefits with epigenetic inhibitors or synthetic miR-338-5p/-421 mimics as adjuvant therapy for the treatment of aggressive SPINK1⁺ malignancies.

Fresh-frozen prostate cancer specimens used in this study were procured from King George's Medical University (KGMU, Lucknow, India). Clinical specimens were collected after obtaining written informed consent and institutional review board approvals from KGMU and Indian Institute of Technology Kanpur (Kanpur, India). A total of 20 prostate cancer specimens were selected for this study based on SPINK1 and TMPRSS2-ERG status, confirmed by qPCR, IHC, and FISH (4). Tissue microarrays (TMA) comprising prostate cancer specimens (n = 238) were obtained from Department of Pathology, Henry Ford Health System (Detroit, MI), after getting written informed consent and institutional review board approval. All patients' specimens used in this study were collected in accordance with the Declaration of Helsinki. TMAs were stained for SPINK1 and EZH2 by performing IHC and RNA in situ hybridization (RNA-ISH), respectively.

For mouse xenograft studies, 5- to 6-week-old NOD.CB17-Prkdc^scid/J (NOD/SCID) male mice (Jackson Laboratory) were randomized into three groups (n = 8 for each experimental condition). All procedures involving mice were approved by the Committee for the Purpose of Control and Supervision of Experiments on Animals (CPCSEA) and conform to all regulatory standards of the Institutional Animal Ethics Committee of the Indian Institute of Technology Kanpur. Detailed methodology of the xenograft studies is provided in the Supplementary Methods and Supplementary Table S1.

Full-length SPINK1 3′ untranslated region (3′UTR) wild-type and mutant with altered residues in the binding sites of miR-338-5p and miR-421 was cloned in Firefly/Renilla Dual-Luciferase reporter vector pEZX-MT01 (GeneCopoeia). Cells were seeded in a 24-well plate at 30%–40% confluency and cotransfected with 30 pmol of miRNA mimics along with 25 ng of pEZX-MT01 constructs using Lipofectamine RNAiMax (Invitrogen). Luciferase assay was performed using Dual-Glo Luciferase Assay (Promega) 24 hours after the second transfection. Firefly luciferase activity was obtained by normalizing with Renilla for each sample.

Total RNA was isolated from stable 22RV1-miR-338-5p, 22RV1-miR-421, and 22RV1-CTL cells and subjected to Whole Human Genome Oligo Microarray profiling (dual color) using Agilent platform (8 × 60K format) according to the manufacturer's protocol. Microarray hybridization was performed using three independent stable miRNA-overexpressing clones against control cells. Microarray data was normalized by following locally weighted linear regression (Lowess normalization). Furthermore, details of differential expression analysis are provided in the Supplementary Methods and Supplementary Table S1.

Statistical significance was determined by either two-tailed Student t test for independent samples or one-way ANOVA, otherwise specified. The differences between the experimental groups were considered significant if the P value was of less than 0.05. Error bars represent mean ± SEM. All experiments were repeated three times in triplicates.

Additional methodologic details are available in the Supplementary Methods and Supplementary Table S1.

The gene expression microarray data from this study has been submitted to the NCBI Gene Expression Omnibus (GEO, http://www.ncbi.nlm.nih.gov/geo/) under the accession number GSE108558.

We employed four miRNA prediction algorithms, namely PITA (omicstools.com), miRmap (mirmap.ezlab.org), miRanda (microRNA.org), and RNAHybrid (BiBiserv2-RNAhybrid) to examine putative binding of miRNAs to the 3′UTR of SPINK1 transcript. Notably, three miRNAs (miR-338-5p, -421, and -876-5p) were predicted as strong candidates by all four algorithms (Fig. 1A; Supplementary Table S2), and were taken forward for further investigation. To examine the differential expression of these miRNAs between SPINK1⁺ and ERG⁺ patients' specimens, publicly available RNA-sequencing (RNA-seq) data from The Cancer Genome Atlas Prostate Adenocarcinoma (TCGA-PRAD) dataset were analyzed. Interestingly, hierarchical clustering of TCGA-PRAD RNA-seq data exhibit reduced expression of miR-338-5p and miR-421 (miR-338-5p/-421) in SPINK1⁺/ERG⁻ patients (Fig. 1B). To validate further, we examined the expression of miR-338-5p/-421 and miR-876-5p in our prostate cancer patients' specimens. A significant lower expression of miR-338-5p and miR-421 was observed specifically in SPINK1⁺ as compared with ERG⁺ specimens, while no difference in miR-876-5p expression was noticed (Fig. 1C). To understand the clinical significance of miR-338-5p/-421, we stratified TCGA-PRAD data into high and low miRNA-expressing groups. Intriguingly, patients with low miR-338-5p expression showed significant (P = 0.0024) association with decreased survival probability compared with high miRNAs group, while no such association was found in case of miR-421 (Fig. 1D). We speculate that this could be attributed due to narrow range of miRNA-421 expression in TCGA-PRAD cohort. Moreover, lower expression of miR-338-5p also associates with higher Gleason score (Fig. 1E). Similar association for both miRNAs was further confirmed in another independent cohort (GSE45604) (Fig. 1F). In summary, SPINK1⁺ subtype shows lower expression of miR-338-5p/-421, which strongly associates with overall poor survival and aggressiveness of the disease.

Having established an association between miR-338-5p/-421 and SPINK1 expression in prostate cancer specimens, we next examined the ability of these miRNAs to bind to SPINK1-3′-UTR. The wild-type (3′-UTR-WT) and mutant (3′-UTR-mut) SPINK1-3′-UTR cloned in Firefly/Renilla dual-luciferase reporter vectors were cotransfected with synthetic mimics for miR-338-5p or miR-421 in HEK293T cells, a significant reduction in the luciferase activity was noted with 3′-UTR-WT, while no suppressive effect was observed in 3′-UTR-mut constructs (Fig. 1G). We next evaluated the expression of these miRNAs in various prostate cancer cell lines including 22RV1 (SPINK1⁺), ETS-fusion positive VCaP (TMPRSS2-ERG⁺), and LNCaP (ETV1⁺) cells. Supporting our observation in clinical specimens, the cell line data also showed lower expression of miR-338-5p/-421 in 22RV1 cells relative to ETS-fusion–positive cell lines (Supplementary Fig. S1A). To further ascertain that miR-338-5p/-421 specifically regulates SPINK1, we used antagomiRs (anti-miR) to abrogate miR-338-5p and miR-421 expression (anti-338-5p and anti-421, respectively) in VCaP cells (Supplementary Fig. S1B). As expected, anti-338-5p or anti-421 significantly induced SPINK1 expression with concomitant increase in cell invasion and migration (Fig. 1H and I; Supplementary Fig. S1C and S1D), while there was no change in the endogenous ERG expression (Fig. 1H; Supplementary Fig. S1C). We next established stable miR-338-5p or miR-421–overexpressing 22RV1 cells (22RV1-miR-338-5p and 22RV1-miR-421, respectively) and examined SPINK1 expression, a significant reduction in SPINK1 both at transcript (∼80%–90%) and protein (Fig. 1J) levels was observed. Because, SPINK1 overexpression has also been implicated in colorectal, lung, pancreatic, breast, and ovarian cancers (17, 18), we sought to examine whether SPINK1 is regulated by a similar mechanism in cancers of different cellular/tissue origins. Thus, we determined the status of SPINK1 expression in multiple cancer cell lines (Supplementary Fig. S2A and S2B). Furthermore, SPINK1⁺ cancer cell lines, namely, colorectal (WiDr), melanoma (SK-MEL-173), pancreatic (CAPAN-1), and prostate (22RV1) transfected with mimics for miR-338-5p or miR-421 showed a significant decrease in SPINK1 expression (Supplementary Fig. S2C and S2D). This provides irrevocable evidence that these two miRNAs modulate the expression of SPINK1 transcript irrespective of the tissue background. To ascertain whether decrease in oncogenic properties is indeed due to miR-338/-421–mediated reduction in SPINK1 expression, a rescue cell migration assay using human recombinant SPINK1 (rSPINK1) was performed. As expected, 22RV1-miR-338/-421 cells show decrease in migration, while adding rSPINK1 to these miRNA-overexpressing cells rescued the invasive phenotype, indicating that miR-338/-421–mediated effects are indeed due to reduced SPINK1 expression (Supplementary Fig. S2E).

SPINK1 overexpression is known to contribute to cell proliferation, invasion, motility, and distant metastases (4, 7, 19). Hence, to understand the functional relevance of miR-338-5p/-421, we examined 22RV1-miR-338-5p and 22RV1-miR-421 cells for any change in their oncogenic properties. Both 22RV1-miR-338-5p (C1 and C2) and 22RV1-miR-421 (pooled and C1) cells showed a significant decrease in cell proliferation compared with 22RV1-CTL cells (Fig. 2A). Similarly, a significant reduction in invasive properties of 22RV1-miR-338-5p and 22RV1-miR-421 cells was noted (∼40% and 60%, respectively; Fig. 2B), whereas only a modest decrease in cell proliferation and invasion was observed in pooled 22RV1-miR-338-5p cells (Supplementary Fig. S3A–S3C). To assess any change in neoplastic transformation, soft-agar colony formation assay was performed, where both 22RV1-miR-338-5p and 22RV1-miR-421 cells exhibited marked reduction (∼60% and ∼80%, respectively) in number and size of the colonies (Fig. 2C). Likewise, 22RV1-miR-338-5p and 22RV1-miR-421 cells demonstrate significantly lower numbers (∼70% and ∼60%, respectively) of dense foci (Fig. 2D). However, overexpressing miR-338-5p/-421 in immortalized benign prostate epithelial RWPE-1 cells show no significant change in cell proliferation or migration with miR-338-5p mimics' transfection, but a marginal decrease in proliferation and migration was noted with miR-421 (Supplementary Fig. S3D). Furthermore, to demonstrate that miR-338-5p/-421 modulate SPINK1 expression and attenuate SPINK1-mediated oncogenicity irrespective of the tissue background, we performed functional assays using colorectal carcinoma WiDr cells (SPINK1⁺) stably overexpressing these miRNAs. As anticipated, a significant decrease in the oncogenic potential of the miR-338-5p/-421–overexpressing WiDr cells was observed (Supplementary Fig. S3E and S3F).

To examine the tumorigenic potential of 22RV1-miR-338-5p and 22RV1-miR-421 cells in vivo, chick chorioallantoic membrane (CAM) assay was performed, and relative number of intravasated cancer cells was analyzed. Consistent with in vitro results, 22RV1-miR-338-5p and 22RV1-miR-421 cells showed significant reduction in the number of intravasated cells and tumor weight, compared with control (Supplementary Fig. S4A–S4C). To evaluate distant metastases, lungs and liver excised from the chick embryos were characterized for metastasized cancer cells. The groups implanted with miRNA-overexpressing cells revealed approximately 80% reduction in cancer cell metastases to lungs (Supplementary Fig. S4D), while no sign of liver metastases was observed in either group. Furthermore, tumor xenograft experiment was recapitulated in immunodeficient NOD/SCID mice (n = 8 per group) by subcutaneously implanting 22RV1-miR-338-5p, 22RV1-miR-421, and 22RV1-CTL cells into flank region, and trend of tumor growth was recorded. A significant reduction in the tumor burden was observed in the mice bearing miR-338-5p and miR-421–overexpressing xenografts as compared with control (∼70% and 85% reduction, respectively; Fig. 2E and F). To examine spontaneous metastases, lung, liver, and bone marrow were excised from the xenografted mice, and genomic DNA was quantified for the presence of human-specific Alu sequences. A significant decrease (∼85% for miR-338-5p and ∼90% for miR-421) in cancer cell metastases was observed in the groups implanted with miRNA-338-5p/-421–overexpressing cells (Fig. 2G). Similar to CAM assay, cancer cells failed to metastasize to murine liver (data not shown). Furthermore, a significant drop (∼50%) in Ki-67–positive cells in the miRNA-overexpressing xenografts confirms that tumor regression was indeed due to decline in cell proliferation (Supplementary Fig. S4E). Taken together, our findings indicate that miR-338-5p/-421 downregulate the expression of SPINK1 and abrogate SPINK1-mediated oncogenic properties and tumorigenesis.

To explore critical biological pathways involved in the tumor-suppressive properties rendered by miR-338-5p/-421 in SPINK1⁺ cancers, we determined global gene expression profiles of miRNA-overexpressing 22RV1 cells. Our analysis revealed 2,801 and 2,979 genes significantly dysregulated in 22RV1-miR-338-5p and 22RV1-miR-421 cells, respectively, relative to control (log₂ fold change of 0.6, FDR < 0.05 and P < 0.05). Approximately 22% (704 genes) of the downregulated and approximately 15% (506 genes) of the upregulated transcripts show an overlap in miRNA-overexpressing cells (90% confidence interval; Fig. 3A), indicating that these miRNAs regulate a significant number of common gene sets and cellular processes. To examine biological processes commonly regulated by miR-338-5p/-421, we employed the Database for Annotation, Visualization and Integrated Discovery (DAVID) and gene set enrichment analysis (GSEA). Most of the downregulated genes were associated with DNA double-strand break repair by homologous recombination, cell-cycle regulation including G₂–M-phase transition, stem cell maintenance, histone methylation, and negative regulation of cell–cell adhesion, whereas genes involved in negative regulation of gene expression or epigenetics, intrinsic apoptotic signaling pathways, and negative regulation of metabolic process and cell cycle were significantly upregulated (Fig. 3B; Supplementary Table S3). Moreover, GSEA also revealed a significant decrease in enrichment of genes associated with sustaining proliferative signaling (EGFR and MEK/ERK) and cell-cycle regulators (E2F targets and G₂–M transition) in miRNA-overexpressing cells, while tumor-suppressive p53 signaling was found to be positively enriched (Fig. 3C), indicating their role in reduced oncogenicity. In addition, an overlapping network of pathways using Enrichment map revealed regulation of cell-cycle phase transition and DNA repair pathways (overlap coefficient = 0.8, P < 0.001, FDR = 0.01), as one of the significantly enriched pathways for both miRNAs (Supplementary Fig. S5).

Because MAPK signaling pathways involving a series of protein kinase cascades play a critical role in the regulation of cell proliferation, we examined the phosphorylation status of MEK (pMEK) and ERK (pERK), as a read-out of this pathway. In agreement with our in silico analysis, a significant decrease in pMEK and pERK was observed in 22RV1-miR-338-5p and 22RV1-miR-421 cells (Fig. 3D). E2F transcription factors are known to interact with phosphorylated retinoblastoma and positively regulate genes involved in S-phase entry and DNA synthesis (20); thus, we next examined the E2F1 level in miRNA-overexpressing cells, surprisingly a notable decrease in E2F1 was observed (Fig. 3D). Furthermore, a significant decrease in the expression of genes involved in G₁–S transition such as cyclin E2 (CCNE2), cyclin A2 (CCNA2), and cyclin-dependent kinase (CDK1 and CDK6), including mini-chromosome maintenance (MCM3 and MCM10), required for the initiation of eukaryotic replication machinery was recorded (Fig. 3E). Thus, these findings corroborate with previous literature that during DNA damage, CDKs being cell-cycle regulators crosstalk with the checkpoint activation network to temporarily halt the cell-cycle progression and promote DNA repair (21). Intriguingly, presence of putative miR-338-5p/miR-421–binding sites on the 3′UTRs of these cell-cycle regulators (Supplementary Table S4) further support that these targets could be directly controlled by these miRNAs. Next, to validate that miR-338-5p/-421 overexpression leads to cell-cycle arrest, cell-cycle analysis using miRNA mimic–transfected 22RV1 cells revealed a significant increase in cells arrested in S-phase (Supplementary Fig. S6A). To delineate that this increase in the S-phase cells is indeed due to cell-cycle arrest and not because of DNA replication, BrdU-7AAD–based cellcycle analysis was performed, which revealed a significant decrease in the percentage of BrdU-incorporated S-phase cells (Fig. 3F). Taken together, our findings strongly indicate that miR-338-5p/-421 overexpression led to S-phase arrest, thus elucidating the mechanism for reduced proliferation and dramatic regression in tumor growth.

Association between epithelial-to-mesenchymal transition (EMT) and cancer stem cells (CSC) has been well-established, indicating that a subpopulation of neoplastic cells, which harbor self-renewal capacity and pluripotency, are associated with highly metastatic and drug-resistant cancers (22). To delineate the pathways involved in miR-338-5p/-421–mediated regression in tumor burden and metastases (Fig. 2E–G), we analyzed microarray data and noted a marked decrease in the expression of genes involved in EMT and stemness (Fig. 4A), including the EMT-inducing transcription factors namely, SNAI1 (SNAIL), SNAI2 (SLUG), and TWIST1 (Fig. 4B and C). Because SNAIL and SLUG are known to negatively regulate CDH1 (E-Cadherin; ref. 23), an epithelial marker involved in cell–cell adhesion, we next examined miR-338-5p/-421–overexpressing cells for any change in E-Cadherin expression: a prominent increase in the membrane localization of E-Cadherin, while a significant decrease in the expression of vimentin, a mesenchymal marker, was observed (Fig. 4D).

A subpopulation (CD117⁺/ABCG2⁺) of 22RV1 cells, known as prostate carcinoma–initiating stem-like cells, has been shown to exhibit stemness and multi-drug resistance (24). Therefore, we examined the expression of genes associated with CSC-like properties in 22RV1-miR-338-5p/-421 cells. Strikingly, the expression of well-known pluripotency markers, such as AURKA, SOX9, and OCT-4, and stem cell surface markers EPCAM, CD117 (c-Kit), and ABCG2, an ATP-binding cassette transporter, were markedly downregulated in miRNA-overexpressing cells (Fig. 4E and F). Having confirmed that miR-338-5p/-421 downregulate the expression of ABCG2 and c-Kit, we next evaluated the efflux of Hoechst dye via ABC transporters in the absence or presence of verapamil, a competitive inhibitor for ABC transporters (25). As expected, 22RV1-miR-338-5p/-421 cells show a significant reduction (∼91% and 89%, respectively) in the side population (SP) cells involved in Hoechst dye efflux (Fig. 4G). As expected, efflux assay performed in the presence of verapamil shows substantial reduction in the SP cells due to inhibition of ABC transporters (Fig. 4G). Furthermore, to confirm whether miRNA overexpression leads to decrease in CSC-like properties, prostatosphere assay, a surrogate model for testing enhanced stem cell–like properties, was performed: a significant decrease in the size and prostatosphere-forming efficiency was observed in 22RV1-miR-338-5p/-421 cells (Fig. 4H and I). Furthermore, prostatospheres formed by miRNA-overexpressing cells exhibit a significant reduction in the expression of genes implicated in self-renewal and stemness (Supplementary Fig. S6B). Intriguingly, miR-338-5p/-421 putative binding sites on the 3′UTR of EPCAM, c-Kit, SOX9, SOX2, and ABCG2 were also noticed (Supplementary Table S4), suggesting a possible mechanism involved in the downregulation of these genes.

Epigenetic regulators, such as ten-eleven-translocation (TET) family member, TET1, convert 5′-methylcytosine (5mC) to 5′-hydroxymethylcytosine (5hmC) and are well-known to induce pluripotency and maintain self-renewal capacity (26). Thus, we analyzed the expression of TET family members. A striking decrease in TET1 expression was observed in miRNA-overexpressing cells (Fig. 4J). Because major players involved in drug resistance, such as ABCG2 and c-Kit were downregulated in miRNAs-overexpressing cells, we examined the sensitivity of these cells to chemotherapeutic drugs. Interestingly, 22RV1-miR-338-5p/-421 cells show enhanced sensitivity to doxorubicin as compared with 22RV1-CTL (Supplementary Fig. S6C). Collectively, miR-338-5p/-421 downregulates the expression of genes implicated in multiple oncogenic pathways namely EMT, stemness, and drug resistance, signifying that these two tumor suppressor miRNAs could represent alternative novel approaches for integrative cancer therapy (Fig. 4K).

Aberrant transcriptional regulation, genomic loss, or epigenetic silencing are well-known mechanisms involved in miRNA deregulation (27, 28). Because SPINK1⁺ patients exhibit reduced expression of miR-338-5p/-421, we sought to decipher the role of epigenetic silencing involved in reduced expression of these miRNAs. Moreover, one of the well-studied key players of epigenetic silencing is EZH2, implicated in many malignancies; thus, we interrogated for any plausible association between SPINK1 and EZH2 using Memorial Sloan Kettering Cancer Center (MSKCC) patients' cohort on cBioPortal (http://cbioportal.org). Interestingly, most of the SPINK1⁺ specimens show concordance with EZH2 expression (Fig. 5A). Furthermore, TCGA-PRAD patients harboring higher expression of EZH2 show increased levels of SPINK1 and decreased expression of miR-338-5p/-421 as compared with EZH2-low patients (Fig. 5B). To further confirm the association between these two oncogenes, we performed IHC and RNA-ISH for SPINK1 and EZH2 expression, respectively, using TMAs comprising a total of 238 prostate cancer specimens. In accordance with the previous reports (3, 4), we found that 21% of patients with prostate cancer (50 of 238 cases) were positive for SPINK1 expression. Interestingly, 88% (44 cases) of these SPINK1⁺ specimens show positive staining for EZH2, of which approximately 14% of the SPINK1⁺/EZH2⁺ patients fall into the high EZH2 range (score 3 and 4), approximately 36% in medium EZH2 (score 2) and approximately 50% into low EZH2 expression group (score 1) indicating a significant association between SPINK1 and EZH2 expression (Fig. 5C; χ² = 13.66; P = 0.008). Although 75% (141 cases) of the SPINK1⁻ patients were also found positive for EZH2, while majority of these cases (∼71%) exhibit low EZH2 levels (score 1). Thus, in corroboration with the previous reports (1, 12), our data indicate a more pronounced role of epigenetic alterations involved in ETS fusion–negative prostate cancer. Although, six SPINK1⁺ cases failed to show any expression of EZH2, indicating that an alternative mechanism may be involved in SPINK1 regulation or possibly miRNA-338/-421 genomic deletion could be a cause.

To investigate whether epigenetic silencing of these miRNAs is mediated by EZH2, we screened the promoters of miR-338, miR-421, and FTX (miR-421 host gene) for putative transcription factor–binding sites and identified MYC and MAX (Myc-Associated Factor X) elements within approximately 2-kb upstream of Transcription Start Site (TSS). MYC is known to form a repressive complex with EZH2 and histone deacetylases (HDAC), and downregulates multiple tumor-suppressive miRNAs, which in turn target PRC2-interacting partners (29). In addition, EZH2-silenced DU145 cells' miRNA expression data (GSE26996) indicates an increase in the expression of numerous EZH2-regulated miRNAs including miR-338-5p and miR-421 (Supplementary Fig. S7A). We therefore examined the promoters of miR-338 and FTX for the recruitment of EZH2 in stable 22RV1-miR-338-5p, 22RV1-miR-421, and 22RV1-CTL cells. Interestingly, a significant enrichment of EZH2 over input was observed on the promoters of miR-338 and FTX in control cells, while a substantial decrease in miR-338-5p/-421–overexpressing cells was noticed (Fig. 5D), suggesting the presence of negative feedback–regulatory network between miR-338-5p/-421 and EZH2. Subsequently, we checked for EZH2 recruitment on miR-421 promoter and no enrichment was observed (Supplementary Fig. S7B), indicating that FTX promoter regulates the expression of this intronic miRNA. Furthermore, to confirm EZH2-mediated methyltransferase activity, we sought to identify H3K27me3 marks on these promoters. A remarkable enrichment of H3K27me3 marks on the miR-338, miR-421, and FTX promoters were noted relative to IgG control (Fig. 5D; Supplementary Fig. S7B), confirming the role of EZH2-mediated epigenetic silencing of miRNA-338-5p/-421.

Comprehensive GSEA revealed that miRNA-338/-421–overexpressing cells show an enrichment for EZH2-interacting partners, including PRC2 members (30) and EZH2-regulated genes (refs. 31, 32; Fig. 5E; Supplementary Fig. S7C), indicating that these two miRNAs in turn regulate EZH2 partners and their target genes. Thus, we next examined the putative binding of miRNA-338-5p/-421 on the 3′UTR of the PRC2 members, interestingly both miRNAs show negative mirSVR binding score, indicating miRNAs binding strength (Supplementary Table S4). Moreover, a significant decrease in the transcript levels of EZH2, and its interacting partners SUZ12, RBBP4, RBBP7, and MTF2, were observed in miRNA-338-5p/-421–overexpressing cells (Fig. 5F; Supplementary Fig. S7D). Collectively, our data indicate that overexpression of miR-338-5p/-421 downregulates EZH2 expression and its interacting members, leading to impaired histone methyltransferase activity of PRC2, thereby establishing a double-negative feedback loop.

Because inhibitors for chromatin modifiers are known to erase epigenetic marks, we tested 3-Deazaneplanocin A (DZNep), a histone methyltransferase inhibitor; 2′-deoxy-5-azacytidine (5-Aza), a DNA methyltransferase (DNMT) inhibitor, and Trichostatin A (TSA), a HDAC inhibitor, in 22RV1 cells and examined miR-338-5p/-421 expression. Treatment with TSA, DZNep, 5-Aza alone or a combination of DZNep and TSA in 22RV1 cells showed a modest increase in miR-338-5p/-421 expression, while 5-Aza and TSA together resulted in approximately 9-fold increase (Fig. 6A). Intriguingly, 5-Aza and TSA in combination results in significant increase in miRNAs expression accompanied with a notable decrease (∼60%–80%) in SPINK1 levels (Fig. 6B). Because, 3′-arm of miR-338 (miR-338-3p) is known to negatively regulate Apoptosis Associated Tyrosine Kinase (AATK) expression (33), likewise a significant reduction in the AATK expression was noticed (Fig. 6B). Previously, a deletion construct of FTX showed decreased expression of miR-374/-421 cluster (34). Corroborating with this, we also observed a significant increase in the FTX and miR-421 expression upon 5-Aza and TSA combinatorial treatment, signifying the importance of FTX in the regulation of miR-421 (Fig. 6B).

In addition, EZH2 is also known to interact with DNMTs, thus enabling chromatin remodeling and DNA methylation (35). We next examined the presence of methylated CpG marks on miR-338-5p and FTX promoters. Interestingly, methylated DNA immunoprecipitation (MeDIP) revealed locus-specific enrichment in 5mC levels over 5hmC on these regulatory regions (Fig. 6C). To ascertain the presence of DNA methylation marks, we performed bisulfite sequencing using prostate cancer cell lines, a relative increase in the methylated CpG sites on miR-338 and FTX promoters was observed in 22RV1 (SPINK1⁺) as compared with VCaP (ERG⁺) cells (Fig. 6D). No significant difference in the methylated CpG sites on the AATK and miR-421 promoters was observed (Supplementary Fig. S7E and S7F). To recapitulate this finding in patients with prostate cancer, bisulfite sequencing was carried out on SPINK1⁺ (n = 5) and ERG fusion–positive (n = 5) patients' specimens. Interestingly, all SPINK1⁺ specimens exhibit increased methylation marks on the promoters of miR-338 and FTX as compared with ERG⁺ (Fig. 6D). Taken together, our results strongly indicate that epigenetic machinery comprising of EZH2 and its interacting partners play a critical role in the epigenetic silencing of miRNA-338-5p and miR-421 in SPINK1⁺ subtype, which, in turn, reaffirms its silencing by a positive feedback loop.

In this study, we unraveled the underlying molecular mechanism involved in the overexpression of SPINK1 exclusively in ETS fusion–negative prostate cancer. Our study provides a molecular basis for SPINK1 overexpression, brought about by epigenetic repression of the key posttranscriptional negative regulators of SPINK1 namely, miR-338-5p and miR-421. We demonstrate miR-338-5p/-421 exhibits functional anticancer pleiotropy in SPINK1⁺ subtype, by attenuating oncogenic properties, tumor growth, and metastases in murine model. Conversely, miR-421 has been reported to be a potential oncogenic miRNA in multiple cancers (36, 37). However, in corroboration with our findings, a recent report suggested tumor-suppressive role of miR-421 in prostate cancer (38). We also established that miR-338-5p/-421–overexpressing cells display perturbed cell-cycle machinery triggered by dysregulated cyclins and CDKs, subsequently leading to S-phase arrest. Recently miRNAs targeting multiple cyclins/CDKs are shown to be more effective than the FDA-approved CDK4/6 inhibitor in triple-negative breast cancer (39), thus supporting our findings that replenishing miRNA-338-5p/-421 may prove advantageous in SPINK1⁺ cancers.

Emerging evidences suggest a complex interaction between EMT and CSCs during cancer progression, and in developing resistance toward anticancer drugs. Previous studies have implicated role of several miRNAs, such as miR-200 family, and miR-34a (40, 41) in regulating the expression of genes involved in metastases, stemness, and drug resistance. Furthermore, miR-338 exhibits tumor-suppressive role, and inhibits EMT by targeting ZEB2 (42) and PREX2a (43) in gastric cancer. Here, we identified miR-338-5p/-421 as critical regulators of EMT-inducing transcription factors and -associated markers, which in turn led to decreased stem cell–like features. Moreover, CSCs are known to express ABC transporters, which efflux the chemotherapeutic drugs during resistance (25). Remarkably, miR-338-5p/-421 overexpression shows decreased expression of ABCG2 and c-KIT, and consequently a significant drop in the drug-resistant side population, indicating that miR-338-5p/-421 are highly effective in conferring drug sensitivity and reducing the therapy-resistant CSCs. Collectively, our findings provide a solid foundation for qualifying these miRNAs as an adjuvant therapy for the SPINK1⁺ and other drug-resistant malignancies.

Our findings were further strengthened by TCGA study (1), wherein a subset of prostate cancer patients harboring SPOP-mutation/CHD1-deletion exhibits elevated DNA methylation levels accompanied with frequent events of SPINK1 overexpression. Recently, a new subtype of ETS fusion–negative tumors has been defined by frequent mutations in the epigenetic regulators and chromatin remodelers (44). Yet another study, using genome-wide methylated DNA immunoprecipitation sequencing, revealed higher number of methylation events in TMPRSS2-ERG fusion–negative as compared with normal and TMPRSS2–ERG fusion–positive prostate cancer specimens (12). Collectively, these independent findings reaffirm the critical role of epigenetic pathways engaged in the pathogenesis of SPINK1⁺ subtype.

Interestingly, increased methylated regions in the ETS fusion–negative patients have been attributed to hypermethylation of miR-26a, a posttranscriptional regulator of EZH2 (12). Thus, given the central role played by EZH2 and the epigenetic mechanism involved in ETS fusion–negative cases, our findings rationalize the role of EZH2-mediated epigenetic regulation of miR-338-5p/-421 in SPINK1⁺/ETS⁻ subtype. Hence, we propose a molecular model involving SPINK1, EZH2, and miR-338-5p/-421, wherein EZH2 acts as an epigenetic switch and by its histone methyltransferase activity establishes H3K27me3-repressive marks on the regulatory regions of miR-338 and FTX, a miR-421 host gene (Fig. 6E).

In consonance with this, miR-338-5p/-421 overexpression also results in decreased Tet1 expression. Converging lines of evidences suggest dual role of Tet1 in promoting transcription of pluripotency factors and recruitment of PRC2 on CpG-rich promoters (45). Collectively, miR-338-5p/-421–mediated decrease in Tet1 expression might possibly contribute to reduced stemness and drug resistance. We also conjecture that decrease in Tet1 expression may result in reduced PRC2 occupancy on the miRNA promoters, diminish epigenetic silencing marks, and consequently downregulate their targets including SPINK1.

Currently, there is no effective therapeutic intervention for SPINK1⁺ malignancies including prostate, although use of monoclonal EGFR antibody has been suggested (10). Nevertheless, outcome of phase I/II clinical trials using cetuximab (46) and EGFR small-molecule inhibitors was largely unsuccessful (47, 48). For instance, in a phase Ib/IIa clinical trial using cetuximab and doxorubicin combination therapy, only a fraction of patients with CRPC (∼8%) showed >50% PSA decline (46), revealing its limited efficacy. Owing to the pleiotropic anticancer effects exhibited by miRNA-338-5p/-421, we propose miRNA replacement therapy as one of the potential therapeutic approaches for SPINK1⁺ cancers; nonetheless in vivo delivery methods and stability are some of the major challenges for successful translation into the clinic (49). While not restricted to miRNA replacement therapy, this study also suggests alternative therapeutic avenues for SPINK1⁺ malignancies, for instance, adjuvant therapy using inhibitors against DNMTs, HDACs, or EZH2, several of which are already in clinical trials (50). Conclusively, we moved the field forward by addressing an important question that how SPINK1 is aberrantly overexpressed in ETS⁻ prostate cancer, and the stratification of patients based on SPINK1-positive and miRNA-338-5p/-421-low criteria could further improve therapeutic modalities and overall management strategies.

No potential conflicts of interest were disclosed.

Conception and design: V. Bhatia, A. Yadav, B. Ateeq

Development of methodology: V. Bhatia, A. Yadav, B. Ateeq

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): V. Bhatia, A. Yadav, R. Tiwari, S. Nigam, S. Goel, S. Carskadon, N. Gupta, A. Goel, N. Palanisamy, B. Ateeq

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): V. Bhatia, A. Yadav, R. Tiwari, S. Nigam, S. Goel, N. Gupta, N. Palanisamy, B. Ateeq

Writing, review, and/or revision of the manuscript: V. Bhatia, A. Yadav, R. Tiwari, S. Nigam, S. Goel, N. Palanisamy, B. Ateeq

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): V. Bhatia, B. Ateeq

Study supervision: B. Ateeq

Other (directed the overall project): B. Ateeq

B. Ateeq is an Intermediate Fellow of the Wellcome Trust/DBT India Alliance. This work is supported by the Wellcome Trust/DBT India Alliance Fellowship [grant number IA/I(S)/12/2/500635; to B. Ateeq). Research funding from Department of Biotechnology (BT/PR8675/GET/119/1/2015; to B. Ateeq) and Science and Engineering Research Board (EMR/2016/005273; to B. Ateeq), Government of India, is also acknowledged. The authors thank Yuping Zhang, Mahendra Palecha, Ayush Praveen for their technical support, and Anjali Bajpai for critically reading the manuscript. We also thank Jonaki Sen for extending the use of fertilized eggs facility. The IIT Kanpur has filed a patent (IN 201611016564) on the therapeutic applicability of miR-338-5p and miR-421 described in this study in which B. Ateeq, V. Bhatia, and A. Yadav are named as inventors.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.