IFNγ-Induced IFIT5 Promotes Epithelial-to-Mesenchymal Transition in Prostate Cancer via miRNA Processing

IFNγ is first characterized as a cytokine associated with antivirus and antitumor activities during cell-mediated innate immune response (1, 2). Mechanistically, IFNs can activate JAK–STAT signaling pathway leading to the transcriptional activation of a variety of IFN-stimulated genes (ISG), resulting in diverse biologic responses (3). Among ISGs, IFN-induced tetratricopeptide repeat (IFIT) family members are highly inducible. They are viral RNA-binding proteins (4) and a part of antiviral defense mechanisms. Among IFIT orthologs, human IFIT1, IFIT2, and IFIT3 form a complex through the tetratricopeptide repeats (TPR) to degrade viral RNA. However, the functional role of IFIT5 is not fully understood because it acts solely as a monomer that can not only bind directly to viral RNA molecules via its convoluted RNA-binding cleft, but also endogenous cellular RNAs with a 5′-end phosphate cap, including transfer RNAs (tRNA; refs. 5, 6). In this study, we demonstrate a new function of IFIT5 in regulating miRNAs turnover.

miRNAs regulate approximately 60% of protein-coding genes via posttranscriptional suppression, mRNA degradation, or translation inhibition (7, 8). Many miRNAs associated with different stages of tumor development are regulated at transcriptional or posttranscriptional level (9). Unlike most eukaryotic protein genes, several miRNAs such as miR-106a-363 (10) and miR-17-92 are clustered together to generate a polycistronic primary transcript (11–13), which implies a complicated regulation of miRNA biogenesis. For example, miR-363 belongs to the polycistronic miR-106a-363 cluster containing six miRNAs. Unlike the other five miRNAs with similar seed sequences and functions as the oncogenic miR-17-92 cluster (14), the seed sequence of miR-363 is distinct from the rest of miRNAs, suggesting different function. Indeed, based on the specific interaction with IFIT5, miR-363 biogenesis is mediated by miRNA turnover, which appears to be a new function of IFIT5.

Based on the mechanism of IFIT5-elicied miR-363 degradation, additional miRNAs such as miR-101 and miR-128 are subjected to IFIT5 complex and target several epithelial-to-mesenchymal transition (EMT) transcriptional factors. Clinically, loss of these miRNAs is associated with tumor grade of prostate cancer, which is inversely correlated with elevated IFIT5 mRNA level. However, IFIT5 mRNA expression is correlated with ZEB1 and Slug mRNA expression in prostate cancer specimens. Taken together, IFIT5 regulated by IFNγ is involved in unique miRNA degradation that can promote EMT, leading to prostate cancer metastasis.

Cells were obtained from ATCC: LAPC4 in Iscove DMEM containing 10% FBS; RWPE-1 in keratinocyte medium containing 10% FBS; LNCaP and PC3 in RPMI1640 medium containing 10% FBS. All the DAB2IP-KD and control (Con) prostate cell lines (such as RWPE1, PC3, and LAPC4) were described previously (15). Stable IFIT5-knockdown (KD; shIFIT5) and control (shCon) prostatic cell lines were generated from PC3, LAPC4-KD, RWPE1-KD, and C4-2Neo cell lines using pLKO-shIFIT5. Stable IFIT5-overexpressing (IFIT5) and control (Vec) cell lines were generated from LAPC4-Con, RWPE1-Con, and C4-2D2 cell lines using pcDNA3.1-3XFlag-IFIT5 plasmid from Dr. Collins (5). All these cell lines were used within 20 passages and authenticated with the short tandem repeat (STR) profiling by Genomic Core in UT Southwestern (UTSW) periodically and Mycoplasma testing was performed by MycoAlert Kit (Lonza Walkersville Inc.) every quarterly to ensure Mycoplasma-free.

The PCR-amplified DNA fragment of T7- precursor-miRNAs (pre-miRNA; Supplementary Table S1) was separated by 2% agarose gel electrophoresis and purified using Mermaid SPIN Kit (MP Biomedicals) then subjected to in vitro transcription assay using T7 High Yield RNA Synthesis Kit (New England Biolabs). The pre-miRNA molecules were treated with DNase I for 15 minutes at 25°C and purified by acid phenol–chloroform extraction and ethanol precipitation at −20°C for 1 hour. The molecular size and sequence of each purified precursor miRNA was confirmed by gel electrophoresis using 15% TBE-Urea gel and qRT-PCR, respectively.

The in vitro transcribed pre-miRNA was subjected to RNA pull-down assay using Pierce Magnetic RNA-Protein Pull-Down Kit (ThermoScientific). An approximate 100 pmol of pre-miRNA were incubated with 10X RNA Ligase reaction buffer, RNase inhibitor, Biotinylated Cytidine Bisphosphate, and T4 RNA ligase at 16°C for 16 hours. The biotinylated pre-miRNA was then purified and incubated with streptavidin magnetic beads for 30 minutes at room temperature. Whole cell lysates (200 μg) were incubated with the biotinylated pre-miRNA beads at 4°C for 1 hour. After elution, proteins–pre-miRNA complex was separated by SDS-PAGE using Bolt 4% to 12% Bis-Tris Plus gel and stained with Coomassie blue then subjected to LC/MS-MS analysis. For identifying the interaction between mutant pre-miR-363s (SS6Mut, DSMut) and IFIT5 or miRNA processing machinery, the proteins–pre-miRNA complex was subjected to Western blot analysis and blotted by using primary antibodies against IFIT5, Dicer, or Drosha proteins.

The in vitro transcribed pre-miRNAs were incubated with recombinant IFIT5 protein and/or XRN1 (New England Biolab) at 37°C, then the RNA-containing buffer were collected at indicated time points and subjected to 15% TBE-Urea gel electrophoresis. To quantify the degradation of precursor miRNA, the 15% TBE-Urea gel was then stained with GelRed Nucleic Acid Gel Stains (VWR) and visualized under UV light in the AlphaImager device (Protein Simple). The RNA bands were quantified by Multiplex band analysis (AlphaView Software) and the rate of degradation was calculated from each time point normalized to time zero.

Cells (4–10 × 10⁴) in the serum-free medium were plated on the upper chamber (8-μm pore size) of Transwell (Corning) with or without Matrigel for invasion or migration assay, respectively, whereas bottom chamber contained medium supplemented with 10% FBS. After 5 days, cells in the bottom chamber were fixed by 4% paraformaldehyde, stained, and visualized under microscope. Quantification of migratory cells was carried out with Crystal Violet staining and measurement at OD_555nm. Each experiment was performed in triplicates.

All animal work was approved by the Institutional Animal Care and Use Committee from UTSW. Stable clones of PC3-shCon or -shIFIT5 were infected with luciferase lentivirus. One million PC3 cells pretreated with vehicle (PBS) or IFNγ (20 ng/mL, 48 hours) were resuspended in 50 μL PBS and then injected into the tail vein of male SCID mice, followed by intravenous injection of IFNγ (5 ng/mL,) weekly for 4 weeks. At eighth-week post-injection, lung metastasis of PC3 tumor was observed by bioluminescent imaging (BLI) using IVIS system then lungs were excised, fixed in 10% formalin, paraffin-embedded, and stained with hematoxylin and eosin for pathologic identification of tumor nodules presented in the lung parenchyma.

A total of 41 prostate cancer specimens obtained from UT Southwestern Tissue Bank were collected from 6-mm core punch from radical prostectomy and examined by pathologist to determine tumor grade then subjected to RNA extraction and 12 fresh prostate cancer tissues were obtained from men undergoing radical prostatectomy at UT Southwestern University Hospital.

The ex vivo culture was performed as previously described (16). Briefly, fresh prostate cancer tissue was dissected into 1 mm³ cube and placed on a Gelatin sponge (Novartis) bathed in RPMI1640 media supplemented with 10% heat-inactivated FBS, 100 units/mL penicillin–streptomycin, 0.01 mg/mL hydrocortisone, and 0.01 mg/mL insulin (Sigma). In addition, to the media, was added either vehicle, IFNγ (25 ng/mL) or IFNγ (100 ng/mL). Tissues were cultured at 37 °C for 48 hours then snap-freeze in liquid nitrogen for RNA purification. The Institutional Review Board of UTSW approved the tissue procurement protocol for this study and written informed consent was obtained from all patients.

Statistics analyses were performed by using GraphPad Prism software. Statistical significance was evaluated using Student t test. P < 0.05 or P < 0.0001 was considered a significant difference between compared groups and marked with an asterisk. The statistical association between miR-363, miR-101, miR-128, and IFIT5 expression among different grades of human prostate cancer was evaluated with regression correlation analysis.

IFNγ is known to modulate cancer immunity and increase cytotoxicity. In addition, we observed that IFNγ was able to induce the expression of Slug and ZEB1, both are potent EMT transcription factors, in prostate cancer cell lines C4-2 (Fig. 1A) and PC3 cells (Supplementary Fig. S1A) in a dose-dependent manner. In contrast, Disabled homolog 2–interacting protein (DAB2IP) protein expression was reduced in treated cells (Fig. 1A; Supplementary Fig. S1A). We have previously identified DAB2IP as an EMT inhibitor in prostate cancer (15, 17). Thus, we believe that the mechanism of IFNγ-induced EMT is mediated through DAB2IP-regulated pathway. Because emerging evidence demonstrates a critical role of miRNA in EMT process, we decided to profile miRNA between DAB2IP-positive and -KD cells. From miRNA microarray screening (Supplementary Fig. S1B), a significant reduction of miR-363 (1176 folds decrease) was detected in DAB2IP-KD cells. The downregulation of miR-363 in DAB2IP-KD cells was further validated in immortalized normal prostatic epithelial cell (RWPE1 and PNT1A) and prostate cancer lines (LAPC-4 and PC3; Fig. 1B). Ectopic expression of DAB2IP in C4-2Neo or LAPC4-KD cells (Fig. 1C) was able to induce mature miR-363 levels, indicating that DAB2IP could modulate miR-363 expression.

miR-363 is located in the polycistronic miR-106a-363 cluster that is first transcribed into a single primary miRNA containing the entire sequence of the cluster genes. We therefore examined the effect of DAB2IP on the expression levels of primary transcript of miR-106a-363. In contrast to the significant downregulation of mature miR-363 in DAB2IP-KD cells, the expression levels of either primary miR-106a-363 (Fig. 1D) or pre-miR-363 (Fig. 1E) were similar between DAB2IP-positive and -KD cells. Noticeably, among the miRNA members in the miR-106a-363 cluster, only mature miR-363 levels dramatically decreased in DAB2IP-KD cells (i.e., LAPC4-KD and RWPE-1-KD; Fig. 1E; Supplementary Fig. S1C and S1D). However, only mature miR-363 levels increased significantly in C4-2D (Fig. 1E; Supplementary Fig. S1E) with ectopic expression of DAB2IP. These findings indicate that DAB2IP specifically regulates miR-363 maturation from the miR-106a-363 cluster.

To elucidate the machinery responsible for miR-363 maturation process, the protein candidates were pulled down by synthetic pre-miR-363 RNA and subjected to LC/MS-MS analysis. IFIT5 (Supplementary Table S2) is consistently showing higher affinity with pre-miR-363 among all three DAB2IP-negative cell lines (LAPC-KD, RWPE1-KD, and C4-2Neo) when compared with DAB2IP-positive cell lines, LAPC4-Con, RWPE1-Con, and C4-2D2, respectively. Moreover, IFIT5 appeared to be a viral-RNA-binding protein, which match with our criteria while searching for a RNA-interacting protein as a potential candidate from the pre-miR-363 RNA pull-down results. The steady-state levels of IFIT5 mRNA and protein were inversely correlated with DAB2IP (Supplementary Fig. S2A). IFIT5 has not been shown to bind to miRNAs; therefore, the interaction between pre-miR-363 and IFIT5 was confirmed in DAB2IP-negative and -positive prostate cancer cell lines (Fig. 2A). This inhibitory effect of DAB2IP on IFIT5 expression was also demonstrated by the ectopic expression of DAB2IP in LAPC4-KD (Fig. 2B), C4-2 (Fig. 2C), and LNCaP cells (Supplementary Fig. S2B).

To determine the role of IFIT5 in miR-363 maturation, we used siRNA to KD IFIT5 in DAB2IP-negative cells. In LAPC4-KD cells treated with several IFIT5 small-interfering RNA (siRNA), the elevated levels of mature miR-363 were detected, which was inversely correlated with the reduction of the endogenous IFIT5 mRNA levels (Supplementary Fig. S2C). Because IFIT5-C siRNA exhibited high efficiency, this siRNA was further tested with RWPE1-KD cells and data showed a significant elevation of mature miR-363 (Supplementary Fig. S2D). Despite the significantly elevated mature miR-363 in IFIT5-siRNA KD cells (Fig. 2D; Supplementary Fig. S2E), the levels of mature miR-106a, miR-18b, miR-20b, miR-19b-2, miR-92a-2 remained relatively unchanged (Supplementary Fig. S2F). However, we ectopically expressed IFIT5 in DAB2IP-positive cells and observed a significant reduction of mature miR-363 levels but not other mature miRNAs from the same cluster (Fig. 2E; Supplementary Fig. S2G). Overall, our finding indicates that IFIT5 can specifically inhibit miR-363 maturation from the miR-106a-363 cluster.

Knowing IFIT5 as a typical ISG, we further confirmed that all IFNs can induce IFIT5 in PC3 cells (Fig. 2F). Meanwhile, a significant reduction of miR-363 was observed under the same condition (Fig. 2F). Among three IFNs, the type II IFN, IFNγ, has the most potent effect on inducing IFIT5 mRNA and suppressing miR-363. We therefore, used IFNγ to examine its impact on IFIT5 downstream target miRNAs. We first identified that the induction of IFIT5 mRNA by IFNγ was the result of transcriptional activation mediated by STAT1 signaling using IFIT5 gene promoter construct (Fig. 2G). Moreover, a dose-dependent induction of IFIT5 protein and mRNA by IFNγ was detected in LAPC4-KD and PC3 cells (Fig. 2H and I) and significantly reduce mature miR-363 levels compared with other miRNAs in miR-106a-363 cluster (Fig. 2H and I). These data support a key mediator role of IFIT5 in IFNγ-mediated miR-363 suppression. Noticeably, the other member of IFIT family, such as IFIT1, was absent in PC3 cell after IFNγ treatment in contrast to RWPE1 cell (Supplementary Fig. S2H), implying IFIT5 plays a unique role in prostate cancer progression.

Based on the predicted sequences and gene profiling, Slug/SNAI2 mRNA appears to be a potential target gene of miR-363. By transfecting miR-363 vector into DAB2IP-KD cells, the suppression of Slug/SNAI2 mRNA levels was detected in miR-363 expressing cells compared with controls (Fig. 3A and B; Supplementary Fig. S3A). Using both wild-type Slug/SNAI2 3′UTR (Slug-WT3′UTR) and mutant Slug/SNAI2 3′UTR (Slug-Mut³⁶³3′UTR) luciferase reporter genes, a significant reduction of the Slug-WT3′UTR but not the Slug-Mut³⁶³3′UTR activity was detected in RWPE-1-KD (Fig. 3C) and LAPC4-KD cells (Supplementary Fig. S3B).

Slug/SNAI2 is known to promote EMT by suppressing E-cadherin. As expected, an elevation of E-cadherin mRNA and protein was observed in miR-363 overexpressing RWPE1-KD (Fig. 3D and E) and LAPC4-KD cells (Supplementary Fig. S3C). In contrast, vimentin, a mesenchymal marker, was suppressed (Fig. 3D; Supplementary Fig. S3C). Functionally, ectopic expression of miR-363 was able to reduce cell migration in RWPE1-KD (Fig. 3F) and LAPC4-KD cells (Supplementary Fig. S3D). Noticeably, cells collected from the bottom chamber of a Transwell exhibited lower miR-363 levels than those from the upper chamber (Supplementary Fig. S3D). In contrast, inhibition of miR-363 in RWPE1-Con (Fig. 3G), Du145, and C4-2 cells (Supplementary Fig. S3E) increased cell invasion and migration, respectively. Moreover, restored Slug/SNAI2 levels in miR-363–expressing cells were resulted in a dose-dependent reduction of E-cadherin and elevation in vimentin in RWPE1-KD (Fig. 3H) and LAPC4-KD cells (Supplementary Fig. S3F). These data indicate that miR-363 can suppress EMT by targeting Slug in prostate cancer.

IFIT5 has been suggested to suppress virus replication by targeting the 5′-phosphate end of single-stranded viral RNAs for rapid turnover (6). Thus, we examined whether IFIT5 has a direct impact on the stability of pre-miR-363. In fact, pre-miR-363 RNA prepared from in vitro transcription was relatively stable at 37°C (Supplementary Fig. S4A) but quickly degraded in the presence of IFIT5 protein complex (Supplementary Fig. S4A), indicating that the degradation of pre-miR-363 is accelerated by the IFIT5 protein complex. To examine the specificity of IFIT5 in the acceleration of pre-miR-363 degradation, we found no significant change for in vitro degradation rate of pre-miR-92a-2 (immediate adjacent to miR-363) under the same condition (Fig. 4A). Previous studies (4, 5) indicate that IFIT5 protein binds to viral RNA molecules at either 5′-phosphate cap or 5′-tri-phosphate group. By comparing the 5′-end structure between pre-miR-92a-2 and pre-miR-363, we hypothesized that a single nucleotide (uracil) overhang in pre-miR-363, in contrast to the double-stranded blunt end in pre-miR-92a-2, is critical for IFIT5 recognition. Therefore, we generated two mutant pre-miR-363 constructs: one with 5′-end six nucleotides single-stranded overhang (SS⁶Mut) and the other with double-stranded blunt end (DSMut; Fig. 4B) to test their stabilities. The in vivo result (Fig. 4C; Supplementary Fig. S4B) indicated that the expression levels of pre-miR363 or mature miR-363 derived from SS⁶Mut were significantly lower than those from native or DSMut form (Fig. 4C; Supplementary Fig. S4B), indicating that the 5′-end structure of pre-miR-363 dictates the stability of miR-363 maturation. By determining the in vitro degradation rates of native, pre-SS⁶Mut- and pre-DSMut -miR-363 RNA molecules, as we expected, pre-SS⁶Mut-miR-363 was very sensitive to IFIT5 whereas pre-DSMut-miR-363 was the most resistant one (Fig. 4D). Furthermore, we observed a steady elevation of SS⁶Mut -derived mature miR-363 level in a dose-dependent manner in the presence of an incremental IFIT5 siRNA, whereas the expression of mature DSMut-miR-363 was not impacted by IFIT5 siRNA (Fig. 4E). Meanwhile, using RNA pull-down assay, pre-SS⁶Mut-miR-363 exhibited higher affinity to IFIT5 protein than pre-DSMut -miR-363 (Fig. 4F). Noticeably, the biogenesis of these artificial constructs is similar to native one (Fig. 4C; Supplementary Fig. S4B). As expected, all these precursor constructs exhibited low binding affinity to Drosha (Supplementary Fig. S4C), compared with the primary transcript containing both miR-92a-2 and miR-363 (Pri-92a-2+363). However, pre-DSMut-miR-363 exhibited the highest binding affinity to DICER among native and pre-SS⁶Mut-miR-363 (Supplementary Fig. S4C), suggesting IFIT5 could prevent Dicer from binding to pre-miR-363. Knowing the high stability of pre-DSMut-miR-363 in vivo, it exhibited more potent effect on inhibiting EMT (Supplementary Fig. S4D) evidenced by elevated E-cadherin and reduced Slug protein expression in RWPE1 (Fig. 4G) and PC3 cells (Supplementary Fig. S4E). Also, DSMut exhibited a greater impact on diminishing PC3 and LAPC4-KD cell invasion (Fig. 4H; Supplementary Fig. S4F) and migration (Supplementary Fig. S4F). These data conclude that IFIT5 recognizes the unique 5′-end overhanging structure of pre-miR-363 for its degradation.

To further demonstrate the specificity of this unique 5′-end structure of pre-miRNA, we also generated a mutant construct of pre-miR-92a-2 with single nucleotide at 5′-overhang (SS¹Mut pre-miR-92a-2; Supplementary Fig. S4G), which is similar to the 5′-end of pre-miR-363 (Fig. 4B). Using RNA pull-down assay, we observed an increased interaction between SS¹Mut pre-miR-92a-2 and IFIT5 protein, compared with native pre-miR-92a-2 (Supplementary Fig. S4G). Moreover, the degradation rate of pre-SS¹Mut-miR-92a-2 increased in the presence of IFIT5 complex, compared with that of pre-miR-92a-2 (Supplementary Fig. S4H). Thus, the 5′-end overhanging structure of pre-miRNAs dictates IFIT5-elicited miRNA turnover.

Although IFIT5 can elicit miR-363 turnover, IFIT5 does not possess ribonuclease activity. To determine whether a ribonuclease is associated with the IFIT5–pre-miR-363 complex, we further examined LC/MS-MS results derived from pre-miR-363 pull-down protein candidates and identified an exoribonuclease candidate-XRN1. XRN1 is known to regulate mRNA stability via cleavage of de-capped 5′-monophosphorylated mRNA (18, 19) and a recent study also implied its potential role in miRNA turnover (20). Indeed, an interaction was observed between IFIT5 and XRN1 protein in LAPC4-Con cells transfected with Flag-tagged IFIT5 (Fig. 5A). Meanwhile, an interaction between endogenous IFIT5 and XRN1 protein is also observed in PC3 cells (Supplementary Fig. S5A). Also, the expression levels of miR-363 were correlated with the diminished level of XRN1 protein (Fig. 5B; Supplementary Fig. S5B). Similar to IFIT5-KD, data from XRN1-KD cells clearly demonstrated that only mature miR-363 exhibited a significant accumulation whereas the levels of other mature miRNAs (miR-106a, miR-18b, miR-20b, miR-19b-2, and miR-92a-2) remained relatively unchanged (Fig. 5C). By incubating XRN1 immunocomplex (Supplementary Fig. S5C) with native, SS⁶Mut or DSMut pre-miR-363 RNA in vitro, a significantly increased degradation of both native and pre-SS⁶Mut-miR-363 was detected in a time-dependent manner, whereas pre-DSMut-miR-363 levels remained relatively unchanged (Fig. 5D; Supplementary Fig. S5C), implying that the IFIT5 binding structure in the 5′-end of pre-miR-363 is critical for recruiting XRN1. In addition, by increasing IFIT5 expression in XRN1-positive LAPC4-Con cells, XRN1–IFIT5 immunocomplex apparently increased the in vitro degradation of SS⁶Mut-pre-miR-363 compared with control (XRN1 alone; Fig. 5E; Supplementary Fig. S5D). However, knocking down XRN1 in IFIT5-overexpressing LAPC4-Con cells diminished the in vitro degradation rate of pre-SS⁶Mut-miR-363 after incubation with IFIT5 (Fig. 5F; Supplementary Fig. S5E). These findings provide further evidence for the specific function of IFIT5–XRN1 complex in miR-363 turnover. In addition, using recombinant IFIT5 protein with or without XRN1 enzyme, the result (Fig. 5G; Supplementary Fig. S5F) clearly indicated that both IFIT5 and XRN1 proteins are required to degrade pre-miR-363 transcript in vitro. Similarly, the pre-SS⁶Mut-miR-363 is more sensitive to rIFIT5–XRN1 complex-mediated degradation than pre-DSMut-miR-363 (Fig. 5H; Supplementary Fig. S5G). Overall, these data demonstrate that the IFIT5–XRN1 complex is responsible for the degradation of pre-miR-363.

To survey additional miRNAs subjected to IFIT5-mediated precursor miRNA degradation, we performed miRNA microarray screening in IFIT5-overexpressing LAPC4-Con and IFIT5-siRNA KD LAPC4-KD cells (Supplementary Table S3). In particular, among IFIT5-regulated miRNA candidates, both miR-101 and miR-128 appear to have 5′-end single nucleotide overhang structure similar to the pre-miR-363 (Supplementary Table S3) and exhibit tumor suppressor function. We further confirmed that the presence of IFIT5 reduced the expression of mature miR-101 and miR-128 as well as miR-363 in PC3 cell line (Fig. 6A). In contrast, IFIT5-KD in LAPC4-KD cells increased the expression of all three miRNAs (Supplementary Fig. S6A). Also, XRN1 KD in IFIT5-expressing cells could rescue the expression levels of mature miR-363, miR-101 and miR-128 (Supplementary Fig. S6B), indicating the requirement of XRN1 in IFIT5 complex in degrading these miRNAs. Similarly, IFNγ treatment resulted in reducing the expression of miR-101, miR-128, and miR-363 (Fig. 6B). This inhibitory effect of IFNγ can be reversed or diminished by knocking down IFIT5, STAT1, or XRN1 (Fig. 6C). Similarly, overexpression of DAB2IP in PC3 cells also diminished the inhibitory effect of IFNγ on the suppression of miR-101, miR-128, and miR-363 level (Supplementary Fig. S6C), supporting the key role of IFIT5 in IFNγ-elicited precursor miRNAs processing. Based on the 3′UTR sequence, ZEB1 mRNA was predicted as a common target for both miR-101 and miR-128 (Supplementary Fig. S6D), and the results indeed indicated that both miR-101 and miR-128 could suppress ZEB1 mRNA levels (Supplementary Fig. S6D).

By comparing the precursor structures of miR-101 and miR-128, it appeared that both pre-miR-101 and pre-miR-128 have similar 5′-end structure with pre-miR-363 (Supplementary Table S3), we therefore generated two mutant constructs: one with 5′-end single stranded overhang (SSMut) and the other with double-stranded blunt end (DSMut; Fig. 6D and E) to test their expression in IFIT5-expressing PC3 and LAPC4-KD cell lines. As we expected, DSMuts were resistant to IFIT5-elicited miRNA degradation and resulted in elevated expression of mature miRNA in PC3 (Fig. 6F and G) and LAPC4-KD cells (Supplementary Fig. S6E and S6F). Again, DSMuts appeared to degrade ZEB1 more efficiently in PC3 (Fig. 6F and G) and LAPC4-KD cells (Supplementary Fig. S6E and S6F), which are correlated with the suppression of cell invasion in PC3 cells (Fig. 6H and I) and cell migration in LAPC4-KD cells (Supplementary Fig. S6E and S6F). Overall, the effect of IFIT5–XRN1 complex on pre-miR-101/128/363 processing is unique with respect to the similar 5′-end overhang structure.

Based on the mechanism of action of IFIT5–XRN1 complex in the degradation of miRNAs that can target EMT factors, we further examined whether IFNγ could elicit EMT by suppressing these miRNAs via STAT1 signal axis and its downstream effector–IFIT5/XRN1 complex. Indeed, IFNγ treatment increased the PC3 cell invasion (Fig. 7A) and migration (Supplementary Fig. S7A) that was diminished in the absence of STAT1 or IFIT5 (Fig. 7A; Supplementary Fig. S7A), which is consistent with the expression of EMT factors (Slug and ZEB1) or decrease in the mesenchymal marker (vimentin) or increase in the epithelial marker (E-cadherin; Fig. 7B and C). As we expected, the expression of all these three miRNAs was inhibited by IFNγ in a dose-dependent manner (Fig. 6C) and IFNγ failed to suppress the expression of these miRNAs in the absence of XRN1, STAT1, or IFIT5 (Fig. 6C) in which no induction of Slug and ZEB1 mRNA was detected (Fig. 7D). Similarly, the effect of IFNγ on Slug and ZEB1 mRNA induction can be diminished in cells transiently transfected with miR-101, miR-128, or miR-363 (Supplementary Fig. S7B).

Apparently, IFNγ is capable of inducing EMT at low concentrations that are not antitumorigenic or antiproliferation (Supplementary Fig. S7C); its direct antitumor activity is known at much higher concentration (>1,000 ng/mL; ref. 21). These data provide new evidence that IFNγ is a potent inducer of EMT via STAT1–IFIT5/XRN1 signal axis of miRNA regulation.

To examine the in vivo effect of IFNγ on prostate cancer metastasis and the role of IFIT5 in this event, we treated control and IFIT5-KD PC3 cells with IFNγ for 48 hours then cells were injected intravenously into SCID animal via tail vein. IFNγ treatment significantly increases the number and size of metastatic nodules at lung parenchyma, in contrast, loss of IFIT5 dramatically reduces metastasis of prostate cancer with or without IFNγ (Fig. 7E and F; Supplementary Table S4). Furthermore, we demonstrated the effect of IFNγ on EMT clinically, we used an ex vivo culture system (16) using human prostate cancer specimens and data indicated that IFNγ was able to induce the expression of IFIT5, ZEB1, Slug (Fig. 7G), and vimentin (Supplementary Fig. S7D) genes whereas miR-363, miR-101, and miR-128 levels were significantly suppressed by IFNγ treatment (Supplementary Fig. S7E). Meanwhile, the level of miR-363, miR-101, and miR-128 in these ex vivo specimens is inversely correlated with the clinic-pathologic stage of prostate cancer patient donors (Supplementary Table S5; Supplementary Fig. S7F). We also surveyed the expression status of IFIT5 from different grades of prostate cancer specimens and data (Fig. 7H) indicate that IFIT5 mRNA levels were significantly elevated in the high-grade prostate cancer. As expected, the expression pattern of miR-363, miR-101, and miR-128 levels was opposite to that of IFIT5 (Fig. 7H), which is consistent with our observation from tissue culture cell lines. In contrast, miR-92a-2 and miR-19b-2 known as oncomirs, exhibited an elevated expression pattern in prostate cancer tissues compared with normal tissues (Supplementary Fig. S7G), supporting the specificity of IFIT5 on miRNA degradation. Meanwhile, data from prostate cancer specimens also demonstrated a similar correlation between IFIT5 mRNA and miR-363 or miR-101 (Fig. 7I). In addition, analyses of EMT factors or markers in a The Cancer Genome Atlas (TCGA) prostate cancer dataset demonstrated a positive correlation between IFIT5 and ZEB1 (or Slug; Fig. 7J), and vimentin (Supplementary Fig. S7H).

A recent study using whole exon and whole transcriptome sequencing of patients with metastatic tumors demonstrated a strong correlation between cancer metastasis and the expression of interferon-induced genes or EMT. Prostate cancer is often found to have many different kinds of infiltrated immune cells such as macrophages, dendritic cells, and tumor-infiltrating lymphocytes. Instead of eliciting tumor immunity, these immune cells with secreting cytokines are capable of facilitating prostate cancer development. For example, a study has demonstrated that fibroblast growth factor 11 (FGF11) released by the recruited CD4⁺ T cells can induce cell invasion by increasing matrix metalloproteinase 9 (MMP9) in prostate cancer cells (22). In addition, IL4 produced from CD4⁺ T cells has shown to increase prostate cancer cell survival and proliferation by activating the JNK signaling pathway in cancer cells (23). Moreover, IL17 secreted from T helper cells is capable of facilitating prostate cancer invasiveness by increasing several EMT transcription factors and MMP7 (24).

However, IFNγ, a type II IFN derived predominantly from CD4⁺/CD8⁺ lymphocytes and NK cell, is shown to have antitumor activities during innate immune response. Also, IFNγ has been used as a therapeutic agent exhibiting antiproliferative (25), antimetastatic (26), pro-apoptotic (27–30), and anti-angiogenesis (31–34) effects in various cancer types. However, several reports indicate that IFNγ could also facilitate tumor progression. For example, IFNγ can elicit CD4⁺ T-cell loss and impair secondary antitumor immune responses after initial immunotherapy using tumor-bearing mouse model (35). In colorectal carcinoma, IFNγ has been shown to facilitate the induction of indoleamine 2,3-dioxygenase (IDO) that induces the production of kynurenines metabolites and impairs the function of surrounding T cells (36). In addition to its role in immune modulation, blockade of IFNγ receptor (IFNGR) can inhibit peritoneal disseminated tumor growth of ovarian cancer (37). Noticeably, serum IFNγ levels become elevated after radiotherapy in patients with prostate cancer (38). Nevertheless the effect of IFNγ on the overall survival of patients with prostrate cancer remains controversial (39).

In our study, we provide additional evidence that IFNγ is capable of inducing EMT, leading to cancer invasiveness via IFIT5-mediated turnover of tumor suppressor miRNAs (Fig. 6). We also noticed that low concentration of IFNγ without cytotoxicity is capable of inducing EMT of prostate cancer (Fig. 7). To strengthen the clinical evidence of IFN-induced EMT, we treated ex vivo prostate cancer explants with IFNγ and demonstrated that IFNγ could induce similar elevations of IFIT5 and EMT transcriptional factors and suppression of miR-101, miR-128, and miR-363 (Fig. 7G; Supplementary Fig. S7E). Taken together, these data show that IFNγ has a biphasic effect on cancer development. Nevertheless, the protumorigenic effect of IFNγ at low concentration is expected to raise a concern for its application as an antitumor or immunotherapeutic agent.

Unlike other IFIT family proteins, IFIT5 is characterized as a monomeric protein that is capable of binding to viral RNA with 5′-triphosphate group (4) and a broad spectrum of cellular RNA with either 5′-monophosphate or 5′-triphosphate group, including tRNA and other RNA polymerase III transcripts (6). However, the interaction of IFIT5 with miRNA is largely unknown. Knowing that precursor miRNA shares a similar stem loop structure with tRNA and a precursor miRNA still retains 5′-monophosphate group after processing from its primary transcript, we are able to show that IFIT5 is capable of interacting with 5′-end of pre-miRNA molecules. After binding to pre-miRNA, IFIT5 recruits XRN1 to form unique miRNA turnover complex (Fig. 5). For the first time, we demonstrated that the specificity of miRNA recognition by IFIT5 is mainly determined by the 5′-end overhang structure of pre-miRNAs (Figs. 4 and 6). Interestingly, these three tumor suppressor miRNAs (i.e., miR-101, miR-128, and miR-363) share similar 5′-end structure in their pre-miRNA and function in suppressing EMT despite of targeting different EMT transcriptional factors such as ZEB1 and Slug.

To conclude, our study provides a new functional role of IFIT5 in miRNA biogenesis (Fig. 7K), particularly, a new understanding of differential regulation of cluster miRNAs.

Until now, the clinical correlation of IFIT5 in prostate cancer is largely undetermined. In this study, we were able to demonstrate that the expression of IFIT5 is elevated in high-grade prostatic tumor and inverse correlation between IFIT5 and miR-101, -128, and -363 in prostate cancer tumor specimens as well as from prostate cancer TCGA database; this correlative relationship was not observed in other members of the miR-106a-363 cluster. In addition, a significant clinical correlation between IFIT5 and EMT transcription factors (ZEB1 or Slug) was observed from prostate cancer TCGA dataset, which supports the regulatory network of IFIT5-miRNAs-EMT in prostate cancer. Also, data from explants provide additional evidence for the promoting effect of IFNγ on prostate cancer progression, Taking together, we have unveiled new function of IFNγ related with prostate cancer progression and potential therapeutic target(s) from its underlying mechanism.

G.V. Raj reports receiving other commercial research support from Bayer, has received speakers bureau honoraria from Astellas and Pfizer, and has ownership interest (including stock, patents, etc.) in EtiraRx, GaudiumRx, and C-diagnostics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: U-G. Lo, G.V. Raj

Development of methodology: U-G. Lo, D. Yang, C.-J. Lin, R. Sonavane, P. Kapur, G.V. Raj

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Gandee, E. Hernandez, J. Santoyo, S. Ma, R. Sonavane, J. Huang, P. Kapur, G.V. Raj, C.-H. Lai

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): U-G. Lo, J. Santoyo, L. Moro, A.A. Arbini, P. Kapur, G.V. Raj

Writing, review, and/or revision of the manuscript: U-G. Lo, A.A. Arbini, G.V. Raj, D. He, J.-T. Hsieh

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): R.-C. Pong, L. Gandee, A. Dang, J. Santoyo, S.-F. Tseng, H. Lin

Study supervision: J.-T. Hsieh

We thank Dr. Collins (University of California, Berkeley, CA) for providing IFIT5 cDNA constructs, Dr. Dong (Emory University, Atlanta, GA) for providing the psiCHECK2-Slug3′UTR plasmid. Drs. Kou-Juey Wu (China Medical University, Taichung, Taiwan) and Dr. Vimal Selvaraj (Cornell University, Ithaca, NY) for the helpful discussion. We also acknowledge the assistance of the Southwestern Small Animal Imaging Resource, which is supported in part by the Harold C. Simmons Cancer Center through an NCI Cancer Center Support Grant (1P30 CA142543), and the Department of Radiology (NIH 1S10RR024757). This work was supported by grants from the United States Army (W81XWH-11-1-0491 and W81XWH-16-1-0474 to J.-T. Hsieh) and (W81XWH-14-1-0249 to U.-G. Lo), and the Ministry of Science and Technology in Taiwan (MOST103-2911-I-005-507 to H. Lin).

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.