The ERG-Regulated LINC00920 Promotes Prostate Cancer Cell Survival via the 14-3-3ϵ–FOXO Pathway

With almost 1,000,000 newly diagnosed cases annually, prostate cancer is the second most prevalent malignancy in men worldwide (1). While a large proportion of prostate tumors follow an indolent course, a considerable number of cases progress with highly heterogeneous clinical trajectories (2). Consequently, large-scale investigations into the prostate cancer genome, transcriptome, epigenome, and proteome (3–6) were undertaken with the intent of understanding the biology of prostate oncogenesis, and to establish molecular subtypes that can potentially guide the clinical management of the disease. In particular, The Cancer Genome Atlas Prostate Adenocarcinoma (TCGA-PRAD) study classified primary tumors into seven subtypes based on structural variations and somatic mutations (4). The International Cancer Genome Consortium-Early Onset Prostate Cancer (ICGC-EOPC) project, which surveyed the genomes, transcriptomes, and methylomes of tumors sampled from young patients (diagnosis at ≤55 years), identified molecular subgroups that can aid in risk stratification (6).

Chromosomal rearrangements involving ETS (E26 transformation-specific or E-twenty-six) transcription factor genes represent the most prominent somatic genetic alteration in prostate cancer. The fusion of the androgen-regulated TMPRSS2 with the ETS family member ERG (TMPRSS2:ERG) has been identified in approximately 50% of prostate tumors (4). TMPRSS2:ERG fusions result in the overexpression of the master transcription factor ERG, which normally plays a role in endothelial and hematopoietic cell differentiation (7). TMPRSS2:ERG fusions are considered early-stage events that can initiate prostatic intraepithelial neoplasia in adult murine prostate epithelial cells (8) and, with simultaneous loss of the tumor suppressor PTEN, adenocarcinoma (9). Specific TMPRSS2:ERG-induced transcriptional profile and cis-regulatory landscape in prostate cancer cells have also been reported (10). In addition, ERG has been observed to bind and retarget BAF chromatin remodeling complexes (11). Loss-of-function mutations in SPOP (12), whose protein product targets ERG for degradation, and ERF (13), an ERG repressor, were shown to be pathogenic lesions in prostate cancer. Yet, despite these molecular insights, the clinical relevance of ERG status and the reason for its apparent selection and enrichment in prostate tumors remain unclear.

In recent years, noncoding RNAs have been implicated in biological processes involved in normal cell physiology, and their altered expression has been associated with progression of multiple cancer entities (14). Long noncoding RNAs (lncRNA) in particular have been shown to be differentially expressed between tumors and their normal tissue counterparts, suggesting functional relevance. Currently, the number of annotated human lncRNAs has reached more than 150,000 (15). Despite this, only a few representative lncRNAs have been sufficiently characterized, revealing multiple modalities of function such as tethering chromatin modifiers to specific chromosomal loci (16), scaffolding protein complexes (17), regulating protein stability, and subcellular mobilization of RNA binding proteins (RBP; ref. 18). In prostate cancer, disease-associated lncRNAs that have been studied in depth include PCAT1 (19), SChLAP1 (20), and ARLNC1 (21).

Here, we explored the ICGC-EOPC transcriptome dataset for differentially expressed transcripts in tumors and normal tissue to identify as of yet undescribed lncRNAs related to prostate cancer progression. We found the previously uncharacterized lncRNA LINC00920 to display a significantly higher expression in prostate cancer tumors. We sought to characterize its role in prostate cancer cells through cellular assays and global gene expression analyses upon transcript depletion. Furthermore, we employed chromatin isolation by RNA purification coupled with mass spectrometry (ChIRP-MS; ref. 22) to identify the LINC00920 proteome and consequently dissect the molecular function of the transcript. Altogether, our results highlight a modulatory capacity of LINC00920 on 14-3-3/FOXO signaling. Given that FOXO transcription factors are downstream effectors of PTEN, which is commonly lost in prostate cancer (4), our work demonstrates a novel lncRNA-mediated cross-talk between the tumor suppressive PTEN signaling and oncogenic ERG in prostate cancer cells.

The prostate cell lines RWPE-1 (CRL-11609), LNCaP (CRL-1740), VCaP (CRL-2876), DU-145 (HTB-81), and PC-3 (CRL-1435) were purchased from the ATCC. PC-3 cells were propagated in F-12K Medium (ATCC) supplemented with 10% FBS (Life Technologies) at 37°C in a humidified incubator with 5% CO₂. VCaP cells were cultured in DMEM (Life Technologies) with 10% FBS. RWPE-1 cells were cultured in Keratinocyte Serum-Free Medium (Life Technologies) supplemented with 0.05 mg/mL bovine pituitary extract and 5 ng/mL EGF. DU-145 and parental LNCaP cells were cultured in RPMI1640 (Life Technologies) with 10% FBS. Tet-inducible LNCaP #126 clones (23) were cultured in RPMI1640 supplemented with 10% tet-free FBS (Clontech), and 80 μg/mL Hygromycin B (Invitrogen). Cells were tested for Mycoplasma contamination (24) and authenticated by SNP profiling (25), with the latest tests conducted on February 7, 2018. All experiments were performed on cells within 10 passages upon revival from cryopreservation.

RNA from cell lines was isolated using RNeasy Mini Kit (Qiagen) following the manufacturer's protocol. cDNA was synthesized from 2,000 ng RNA input using RevertAid H Minus First Strand cDNA Synthesis Kit (Thermo Fisher Scientific).

Quantitative PCR was performed in the LightCycler 480 II System using ABsolute qPCR (Thermo Fisher Scientific) or Premix Ex Taq (Takara Bio) SYBR Assays using primers listed in Supplementary Table S1. Crossing point-PCR-cycle (Cp) values were generated by the LightCycler 480 software (release 1.5.0) using the second derivative method.

Rapid amplification of cDNA ends (RACE) experiments was performed using the GeneRacer Kit (Life Technologies) and transcript-specific primers listed in Supplementary Table S2. Touchdown PCR made use of GSP1 primers to enrich for target DNA fragments and GSP2 primers were subsequently used. Finally, the 5′ and 3′ ends that were amplified were cloned into pCR4Blunt-TOPO (Life Technologies), and validated through Sanger sequencing.

Cells were seeded to reach at least 60% confluency 24 hours prior to transfection. On the day of transfection, maintenance medium was refreshed and cells were transfected with siRNAs (Supplementary Table S3) using Lipofectamine RNA iMAX (Thermo Fisher Scientific) according to the manufacturer's protocol. Cells were harvested for RNA and protein isolation 48 hours after transfection.

Forty-eight hours after siRNA transfection, PC-3 cells were treated with 300 nmol/L ipatasertib (GDC-0068, Selleckchem) or 5 μmol/L MK-2206 (Selleckchem). Cells were harvested after 24 hours for RNA isolation.

LINC00920 sequence as determined by RACE was used to survey the coding potential of the transcript. Sequences of control coding mRNAs, lncRNAs, and LINC00920 were loaded into Coding Potential Assessment Tool (version 2.0.0, http://lilab.research.bcm.edu/cpat/) and the hg19 assembly was selected as reference genome. Sequences were similarly loaded into Coding Potential Calculator 2 (CPC, version 2.0 beta, https://bio.tools/CPC2). Hg19 PhyloCSF tracks were extracted from the Track Data Hubs and visualized in the UCSC genome browser spanning the chromosomal locus of LINC00290.

A fractionation protocol (26) was adapted to prepare cytoplasmic, nucleoplasmic, and chromatin lysates from PC-3 cells for RNA extraction. A total of 30 × 10⁶ cells were used for subcellular fractionation. RNA from all fractions was extracted using RNeasy Mini Kit (Qiagen), followed by cDNA synthesis. Relative enrichments of HPRT1, GAPDH, LINC00920, NEAT1, and MALAT1 transcripts in each fraction were measured by qPCR.

Cell proliferation, cell migration and invasion, and colony formation assays were performed on PC-3 and DU-145 cells. For proliferation assays, 5 × 10³ transfected cells were seeded into a 96-well plate in triplicate. Cell proliferation was monitored 48-, 72-, and 96-hour posttransfection using the colorimetric WST-1 Reagent (Roche). To quantify cell migration, 1 × 10⁵ cells suspended in serum-free medium were seeded in duplicate into 8 μm transwell inserts (Greiner Bio-One) placed in a 24-well plate. Serum-supplemented medium was added to the bottom of each well to facilitate cell migration. Forty-eight hours after incubation, migrated cells were trypsinized from the bottom of the transwell insert and quantified using the WST-1 reagent. Cell invasion was assayed in transwell inserts precoated with 50 μL 1:6 dilution of Matrigel (BD Biosciences). For colony formation assays, 1.3 × 10³ cells were seeded in duplicate into 6-well plates and cultivated for 9 days. Colonies were stained with crystal violet and counted using the OpenCFU software.

Three biological replicates of LINC00920 knockdown in PC-3 cells using siRNA-Q2, siRNA-Q3, and a scrambled control (Supplementary Table S3), respectively, were performed. Forty-eight hours posttransfection, RNA was extracted using the RNeasy Mini Kit (Qiagen) following the manufacturer's protocol. Gene expression profiling was performed using the Human HT-12 v4 Expression Bead Chip from Illumina. Normalized expression data were processed using the ingenuity pathway analysis (IPA) and gene set enrichment analysis (GSEA; ref. 27). For IPA, top 1,000 upregulated and top 1,000 downregulated genes upon LINC00920 knockdown were overlapped for two independent siRNA experiments. For GSEA, expression fold change values of all genes generated from the biological replicates of microarray experiments were used to prepare .rnk files for each knockdown condition. Using .rnk files as input, enrichment analysis of canonical pathways and gene ontology gene sets was performed using GSEAPreranked (desktop application version 3.0) with permutations value set to 1,000.

Promoter sequence 1,000 bp upstream of the annotated LINC00920 transcription start site was extracted from the UCSC genome browser. ERG binding motifs along the promoter sequence were scanned and scored using JASPAR CORE (28) at a threshold of 85%.

A total of 4 × 10⁶ formaldehyde-fixed PC-3 cells and 30 μL of ChIP-grade Protein G Magnetic Beads (Cell Signaling Technology) were used for each immunoprecipitation reaction. Two micrograms of α-ERG (Abcam, ab92513) or rabbit IgG (Abcam, ab172730) was added to the cleared chromatin lysate. Antibody hybridization was facilitated overnight with rotation at 4°C. Next day, 30 μL of prewashed magnetic beads were added to the reaction tube and hybridized at 4°C for 2 hours. Chromatin elution was performed by adding 150 μL elution buffer (50 mmol/L Tris pH 7.9, 1 mmol/L EDTA pH 8.0, 1% SDS, 50 mmol/L NaHCO₃, and 300 mmol/L NaCl) and 2 μL RNase A to the input and beads. Proteinase K was used to reverse chemical cross-links. DNA was isolated from the eluate using UltraPure phenol:chloroform:isoamyl Alcohol (25:24:1, v/v; Invitrogen). The precipitated DNA was used in subsequent qPCR assays.

LINC00920 promoter fragments and a nongenic location in chromosome 12 were amplified from PC-3 genomic DNA using Phusion High-Fidelity Polymerase (New England Biolabs). Mutant promoter fragments were generated by overlap extension-PCR. In particular, transversions within the ERG binding motif were introduced (GGAA→CCAA). The blunt-ended PCR products were A-tailed and cloned into pCR2.1-TOPO (Invitrogen). Inserts were sequence verified (Eurofins Scientific) and subcloned into pGL4.10[luc2]. For luciferase assays, LNCaP #126 clones were seeded in triplicate in 96-well plates and ERG overexpression was induced with 50 ng/mL Doxycycline (Sigma-Aldrich) after 48 hours. Forty-eight hours postinduction, the cells were cotransfected with a 10:1 ratio of pGL4.10[luc2] construct and pAAVpsi2 (29) Renilla vector using Lipofectamine2000. Luciferase activity was measured after 48 hours using the Dual-Glo Luciferase Assay System (Promega).

ChIRP-MS was performed as described previously (22). For each pulldown reaction, cell lysate equivalent to 2 × 10⁸ formaldehyde-fixed cells was hybridized with 600 pmol of pooled biotinylated oligos (siTOOLs Biotech; Supplementary Table S4) overnight at 37°C. Biotinylated complexes were captured on Dynabeads MyOne Streptavidin C1 (Invitrogen) magnetic beads. Captured proteins were recovered from the beads by benzonase elution and precipitated overnight at 4°C using 25% trichloroacetic acid. Precipitated proteins were washed with 100% ice-cold acetone. Protein samples were boiled in reducing Laemmli buffer and loaded into SDS-PAGE. Coomassie-stained gel pieces were cut out, cysteines were reduced by DTT and carbamidomethylated using iodoacetamide, and proteins were digested with trypsin overnight. Peptides were loaded in a cartridge trap column packed with Acclaim PepMap300 C18 (5 μm, 300 Å wide pore, Thermo Fisher Scientific) and separated in a 60-minute gradient from 3% to 40% acetonitrile on a nanoEase MZ Peptide analytic column. Eluted peptides were analyzed by an online coupled Q-Exactive-HF-X mass spectrometer. Data analysis was carried out by MaxQuant (version 1.6.3.3). An enrichment ratio of 1.2 (LINC00920 ChIRP:lacZ ChIRP) was set to identify candidate LINC00920-interacting proteins. Enriched proteins common to three biological replicates were investigated by computing gene set overlaps in the molecular signatures database (MSigDB) using the BioCarta, Kyoto Encyclopedia of Genes and Genomes, Reactome, and gene ontology gene sets.

A total of 7.5 × 10⁶ glutaraldehyde-fixed VCaP cells and 45 μL of ChIP-grade Protein G Magnetic Beads (Cell Signaling Technology) were used for each immunoprecipitation reaction. Three micrograms of α-14-3-3ϵ (Cell Signaling Technology #9635), α-14-3-3ζ (Cell Signaling Technology #7413), or rabbit IgG (Cell Signaling Technology #2729), respectively, was added to the cleared lysate. Magnetic beads were prewashed and added into the hybridization reaction. Samples were incubated overnight with rotation at 4°C. Beads were washed with RIP wash buffer (3× SSC buffer, 1 mmol/L EDTA pH 8.0, 0.1% Tween-20, and 1× PMSF) and cross-link reversal of input and beads was performed by proteinase K digestion. Immunoprecipitated RNA was isolated using phenol/chloroform/isoamyl alcohol (125:24:1 mixture pH 4.3, Thermo Fisher Scientific). cDNA was synthesized as described, and used for qPCR assays.

Sequence-verified pcDNA3.1(+)_LINC00920 was digested with XbaI to obtain a linearized DNA template for in vitro transcription. Biotinylated LINC00920 RNA was generated using the T7 RNA Polymerase (New England Biolabs) with a biotin RNA Labeling Mix (Roche). The RNA product was purified using the RNeasy Mini Kit (Qiagen).

Recombinant human 14-3-3ϵ protein was obtained from Abcam (ab54317). For each affinity pulldown, 300 ng of recombinant protein and 0.5 pmol purified biotinylated RNA were hybridized. Dynabeads M-270 Streptavidin Beads (Invitrogen) were washed and blocked with 0.1% BSA overnight at 4°C. Next day, 0.5 pmol RNA was diluted to 50 μL with RNA structure buffer (10 mmol/L Tris-HCl pH 7.5, 0.1 mmol/L KCl, and 10 mmol/L MgCl₂) and denatured at 75°C for 2 minutes. The recombinant protein was diluted to 1 mL with EMSA buffer (20 mmol/L HEPES pH 7.6, 50 mmol/L KCl, 3 mmol/L MgCl₂, 1 mmol/L EDTA pH 8.0, and 5 mmol/L DTT) supplemented with 1 × cOmplete, Mini protein inhibitor, 1× PMSF, and 0.05 U/μL Superase In RNase inhibitor. RNase A treatment was performed at a final concentration of 10 μg/mL. Denatured RNA was hybridized with the protein solution at room temperature for 2 hours. Blocked and washed beads were added to the hybridization solution. Afterwards, the beads were washed five times with ice-cold RIP wash buffer. Proteins were eluted in 30 μL 1× Roti-Load 1 protein loading buffer followed by incubation at 95°C for 5 minutes. The supernatant was used for Western blot analysis.

Reduced protein samples were electrophoresed in 4%–20% Mini-PROTEAN TGX Precast Protein Gels (Bio-Rad). Protein transfer onto polyvinylidene difluoride membrane was facilitated using the Trans-Blot Turbo System (Bio-Rad). The membrane was blocked with 5% BSA in 1× PBS-T, transferred into the primary antibody solution, and incubated with rotation at 4°C overnight. The blot was incubated with the appropriate secondary antibody solution at room temperature for 1 hour. Chemiluminescence immunoblot signals were developed using the SuperSignal West Dura Extended Duration Substrate (Thermo Fisher Scientific) and imaged by ChemiDoc XRS+ with Image Lab (Bio-Rad) Software. Band intensities were quantified using Fiji (30). Primary antibodies against 14-3-3ϵ (1:1,000, Cell Signaling Technology #9635), FOXO1 (1:1,000, Cell Signaling Technology #2880), FOXO3a (1:1,000, Cell Signaling Technology #2497), GAPDH (1:1,000, Cell Signaling Technology #2118), and HPRT1 (1:5,000, Abcam, ab109021), and secondary antibody conjugated to horseradish peroxidase against rabbit (1:10,000, Cell Signaling Technology #7074) were used in this study.

PC-3 cells were seeded into 8-well Chamber Slides (ibidi GmbH) and transfected as described above. Fluorescence immunocytochemistry, as described previously (31), was performed on cells 48 hours posttransfection. Rabbit anti-FOXO1 antibody (1:250, Abcam, ab70382) and donkey anti-rabbit IgG H&L conjugated with Alexa Fluor 647 (1:500, Abcam, ab150075) were used to stain FOXO1. Afterwards, cells were mounted using VECTASHIELD HardSet Antifade Mounting Medium with DAPI, and cured at 4°C overnight. Micrographs were obtained using the Axio Observer 7 Inverted Fluorescence Microscope (Zeiss).

Cells were seeded into 8-well chamber slides, transfected, and treated accordingly. Cells were fixed with 4% formaldehyde for 15 minutes at room temperature. The cells were washed thrice with 1× PBS, and covered with blocking solution (5% BSA, 0.3% Triton X-100 in 1× PBS) for 1 hour at room temperature. Cells were incubated overnight at 4°C with diluted primary antibodies targeting FOXO1 (1:1,000, Cell Signaling Technology #2880) and 14-3-3ϵ (1:250, Santa Cruz Biotechnology, sc-23957). Development of fluorescence signals was performed using Duolink In Situ Red Starter Kit Mouse/Rabbit (Sigma-Aldrich). Visualization and z-stack images were obtained using the Axio Observer 7 inverted fluorescence microscope. Maximum stack projections and proximity ligation assay (PLA) signal quantification were performed in Fiji as described previously (32).

All experiments were reproduced in three biological replicates. All quantitative data are presented as mean ± SE. Unpaired t test was used to accept or reject the null hypothesis that there was no significant difference between the control and treated conditions with respect to cellular phenotypes (i.e., cell proliferation, colony formation, and cell migration), expression levels (i.e., relative gene expression and luciferase assays), and enrichment values [i.e., for ChIP/RNA immunoprecipitation (RIP)-qPCR]. P ≤ 0.05 was considered statistically significant.

The microarray data supporting the findings of this study has been deposited in Gene Expression Omnibus under accession number GSE130978.

The ICGC-EOPC cohort (6) used for the selection of LINC00920 consisted of 125 prostate tumor and 10 normal tissue specimens. The selection of differentially expressed transcripts was guided by the following criteria: (i) the transcript must be of the long intergenic noncoding RNA biotype; (ii) the lncRNA must have at least two exons; (iii) the average transcript count (i.e., baseMean value) must be at least 500 for either the tumor or normal sample group; and (iv) a significant (P < 0.05) up- or downregulation must be observed between the tumor and normal sample group (log₂ fold change >|1|). LINC00920 met all these criteria (Fig. 1A and B).

LNCaP, VCaP, DU-145, and PC-3 cells were used as prostate cancer models to validate the lncRNA expression in vitro by qPCR (Fig. 1C). LINC00920 expression levels from the metastatic lines were compared with the benign prostate epithelial cell line RWPE-1. Significant upregulation was observed in PC-3 and VCaP, but not in LNCaP and DU-145 cells, underscoring the cell type–specific nature of lncRNA expression and suggesting a specific lncRNA regulation mechanism shared by VCaP and PC-3 cells.

Because lncRNA annotations are subject to frequent revisions, we performed RACE to determine the length, and validate the presence of polyadenylation at the 3′ terminal end of prostate-specific LINC00920 transcripts. Using RNA isolated from normal human prostate, we confirmed the transcription of LINC00920 from a biexonic gene as annotated in RefSeq and GENCODE (Fig. 1D). The 3′ end that emerged from our RACE analysis was longer than the transcript annotated in RefSeq but shorter compared with the GENCODE annotation. Detailed inspection of the LINC00920 exon 2 sequence revealed eight putative polyadenylation (i.e., AATAAA or ATTAAA) signals, two of which could potentially enable transcript processing that would result in the 3′-end determined by RACE (Supplementary Fig. S1). Reference chromatin immunoprecipitation (ChIP) tracks of H3K27ac and H3K4me3, which are histone marks present at accessible and active promoters, further verify the transcription of LINC00920 from the annotated locus.

Coding-Potential Assessment Tool (CPAT; ref. 33), CPC(34), and Phylogenetic Codon Substitution Frequencies (PhyloCSF; ref. 35) were utilized to evaluate the noncoding potential of LINC00920 (Fig. 1E). Using the transcript sequence verified by RACE, the CPAT and CPC scores were computed for LINC00920, ACTB (ENST00000331789.5), and GAPDH (ENST00000396861.1) mRNAs as coding transcript controls, as well as MALAT1 (ENST00000534336.1) and NEAT1 (ENST00000501122.2) as noncoding controls. As expected, CPAT scores for ACTB and GAPDH mRNA were 0.999, and thus above the coding threshold score of 0.364 (33) for human genes, while values for MALAT1, NEAT1, and LINC00920 transcripts were all below 0.025. Similarly, CPC values were positive for the protein coding transcripts and negative for all three lncRNAs. Finally, projection of the LINC00920 sequence onto PhyloCSF tracks for three forward reading frames revealed negative codon scores for regions where evaluation was possible. Altogether, these results indicate that LINC00920 does not code for a protein.

The functional role of LINC00920 in prostate cancer cells was analyzed by RNAi followed by cellular assays. Upon LINC00920 depletion (Fig. 2A), we observed a significant reduction of cell proliferation for both PC-3 and DU-145 cells at 72 and 96 hours posttransfection timepoints (Fig. 2B). VCaP cells similarly showed increased doubling times upon LINC00920 depletion (Supplementary Fig. S2A and S2B). Transfected PC-3 and DU-145 cells also formed fewer colonies (Fig. 2C and D), and exhibited decreased migratory potential across a Boyden chamber (Fig. 2E and F). However, the cells did not show significant changes in invasiveness through a Matrigel matrix (Supplementary Fig. S2C and S2D).

Gene expression arrays were performed in PC-3 cells to identify cellular and biological processes associated with LINC00920 silencing. For both siRNAs, GSEA showed negative enrichments of pathways associated with cell division, cell cycle, microtubule-based movement, and apoptosis (Fig. 3A). These perturbed processes likely reflect the observed phenotypes upon lncRNA knockdown. IPA using 315 genes (Supplementary Table S5) shared among the top 1,000 upregulated and downregulated genes upon knockdown with the two siRNAs (Fig. 3B) revealed the molecular and cellular functions “Cellular Development,” “Cellular Growth and Proliferation,” “Cell Death and Survival,” “Molecular Movement,” and “Gene Expression” as the most deregulated categories (Fig. 3C). In parallel, prostate cancer–associated processes such as PTEN signaling and 14-3-3–mediated signaling pathways were found to be activated and deactivated, respectively, upon LINC00920 knockdown (Fig. 3D). Because PC-3 cells are PTEN deficient, these observations suggested the possibility of induced effector pathways downstream of PTEN. Interestingly, in both knockdown microarray datasets, FOXO signaling was predicted to be activated as represented by positive z-scores for the FOXO3, FOXO1, and FOXO4 transcription factors (Fig. 3E). These results suggest that the oncogenic properties of LINC00920 in prostate cancer cells are likely due to cellular alterations converging on the PTEN/14-3-3/FOXO signaling axis.

Analyses performed on both the TCGA-PRAD (n = 568) and the ICGC-EOPC (n = 135) datasets revealed positive correlations between ERG and LINC00920 expression levels. Pearson correlation values of 0.4561 and 0.5637 were obtained from the TCGA-PRAD and ICGC-EOPC cohorts, respectively (Fig. 4A). To determine whether this correlation is causal, LINC00920 expression was monitored upon reciprocal knockdown and overexpression of ERG in prostate cancer cells. VCaP cells harbor an allele of the TMPRSS2:ERG fusion (36). ERG knockdown in VCaP cells resulted in significant decrease of LINC00920 expression (Fig. 4B). In contrast, knockdown of LINC00920 in VCaP cells did not result in significant decrease of ERG levels (Supplementary Fig. S3). Moreover, in a tet-inducible LNCaP model (23), induction of ERG overexpression resulted in a concomitant upregulation of LINC00920 (Fig. 4C). To investigate whether ERG can directly activate the transcription of LINC00920, we performed ChIP and luciferase assays using LINC00920 promoter DNA fragments. Two putative ETS domains (Fig. 4D) harboring the core GGA(A/T) motif were predicted within the 1,000 bp window upstream the LINC00920 transcription start site using JASPAR (28). To determine whether these regions are bona fide binding sites for ERG and are relevant for LINC00920 transcription, we performed ChIP using an ERG-targeting antibody. ChIP primers were designed around the two binding domains (Fig. 4E). In comparison to the IgG control, ERG ChIP resulted in significant enrichment of regions amplified by all three primer pairs (Fig. 4F). Thus, ERG directly interacts with the promoter region of LINC00920. In addition, promoter luciferase assays in the tet-inducible ERG-overexpressing LNCaP cells were performed to query whether ERG exhibits preferential binding to either ETS domain. Double transversion (GG > CC) mutations were introduced separately into the identified ETS domains as controls. Upon induction of ERG overexpression, cells transfected with the wild-type promoter fragment showed significantly increased luciferase activity compared with the empty vector and nongenic controls (Fig. 4G). Mutations in either ETS binding region decreased the promoter activity, with ETS domain 2 appearing to be a more relevant binding site than ETS domain 1. Together with the ChIP-qPCR data, these results demonstrate the direct regulation of LINC00920 by ERG, primarily at the ETS binding domain located −60 bp relative to the transcription start site (ETS domain 2). Because PC-3 cells do not harbor the TMPRSS2:ERG allele, we hypothesized that another ETS family member most likely mediates LINC00920 overexpression in this cell line. ETV4 has previously been described to be highly expressed in PC-3 when compared with other prostate cell lines (37). This observation was recapitulated at the transcript level (Fig. 4H). SiRNA-mediated knockdown of ETV4 was performed in PC-3 cells to determine the regulatory effect of ETV4 on LINC00920. Notably, LINC00920 expression was reduced to about 50% upon ETV4 knockdown, suggesting that ETV4 regulates its expression in PC-3 cells (Fig. 4I). Thus, the different endogenous expression levels of LINC00920 in prostate cancer cell lines (Fig. 1C) can be explained through regulation by specific ETS factors.

Similar to proteins, the subcellular enrichment of a transcript can give insight into its function. For example, nuclear lncRNAs have been implicated in transcriptional control of specific genes by recruiting and acting as tethers and scaffolds for chromatin remodeling proteins (16). On the other hand, cytoplasmic lncRNAs have been reported to form complexes with RBPs and mediate processes important for various cellular functions such as protein localization and turnover (18), scaffolding (17), and mRNA translation and stability (38). Consequently, we quantified the relative enrichment of mature LINC00920 transcripts in the chromatin, nucleoplasmic, and cytoplasmic fractions of PC-3 cells (Fig. 5A). As expected, the mRNA controls HPRT1 and GAPDH were most abundant in the cytoplasmic fraction, while the nuclear lncRNAs NEAT1 and MALAT1 were predominantly present in the chromatin fraction. LINC00920 showed high nuclear enrichment, but also, unlike NEAT1 and MALAT1, considerable (ca. 15%) cytoplasmic localization. This finding suggested a molecular function that is not restricted to a particular subcellular compartment.

Because RNAs typically exert their function as parts of ribonucleoprotein complexes, we performed ChIRP-MS to determine protein interaction partners of LINC00920. Thirty 20-nt biotinylated single-stranded antisense DNA oligonucleotides were used to enrich for LINC00920 and its associated proteins (Fig. 5B). Triplicate pulldown experiments of endogenous LINC00920 identified 21 enriched proteins compared with the lacZ control (Fig. 5C; Supplementary Table S6). Gene ontology analysis of these proteins revealed RNA binding and transcript splicing functions (Fig. 5D). Among the identified proteins were two 14-3-3 protein isoforms: 14-3-3ϵ (YWHAE) and 14-3-3ζ (YWHAZ; Fig. 5E). 14-3-3 proteins are small chaperone proteins that bind to phosphorylated ligands. Such binding provides steric hindrance or elicits conformational changes altering the biochemical properties of the proteins bound to 14-3-3 (39). RIP using 14-3-3–specific antibodies was performed in VCaP cells to validate the association of LINC00920 and the 14-3-3 proteins as identified from the ChIRP-MS data. As our RIP protocol included chemical cross-linking and RNA shearing steps, precipitating the full intact transcript would be unlikely. Thus, we used multiple primers targeting the span of the lncRNA transcript to validate 14-3-3 binding and at the same time identify the putative protein binding region of the lncRNA. Significant LINC00920 enrichment over the IgG control was observed for the primer pair amplifying the intronic junction of the transcript upon 14-3-3ϵ, but not 14-3-3ζ, precipitation (Fig. 5F). These results point to the specific interaction of LINC00920 with the 14-3-3ϵ protein isoform. Complementary to RIP, affinity purification on streptavidin beads using in vitro transcribed biotinylated LINC00920 (bi-LINC00920) was performed to determine whether recombinant 14-3-3ϵ (r14-3-3ϵ) can be precipitated by LINC00920. RNase A treatment was used as control to verify that the precipitation of the protein is LINC00920 dependent. In the absence of RNAse A, r14-3-3ϵ could be probed via immunoblotting, while RNA digestion abrogated the band signal (Fig. 5G). These observations suggested a direct interaction between the LINC00920 and proteins in solution.

Because global gene expression analysis of LINC00920-depleted PC-3 cells had revealed concomitant activation of FOXO transcription factors, we further validated this predicted activation by measuring expression levels of the canonical FOXO target genes BCL2L11, GADD45A, and PMAIP1 upon lncRNA knockdown. Because of the PTEN-null status of PC-3 cells, it was necessary to uncouple the influence of the hyperactive PI3K pathway on FOXO signaling that could potentially mask the effect of LINC00920 depletion. Consequently, we attenuated AKT activity in PC-3 cells using two pharmacologic AKT inhibitors, ipatasertib and MK-2206. Ipatasertib is a highly selective pan-AKT inhibitor proven to have low off-target activity, and MK-2206 selectively targets AKT1/2. As expected, while LINC00920 silencing alone yielded minor changes in the expression of FOXO targets, simultaneous LINC00920 knockdown and AKT inhibition through ipatasertib or MK-2206 treatment significantly upregulated all genes in comparison with the siRNA control (Fig. 6). The convergent effects brought about by ipatasertib and MK-2206 indicate that hyperactive AKT due to PTEN loss can curtail the derepressive effect of LINC00920 on FOXO targets. In support of this, we observed similar significant activation of FOXO genes through LINC00920 knockdown alone in PTEN-proficient VCaP cells (Supplementary Fig. S4).

Because the observed upregulation of FOXO gene targets upon silencing of LINC00920 could not be attributed to FOXO expression deregulation at protein levels (Supplementary Fig. S5A), we hypothesized that the binding of LINC00920 to 14-3-3ϵ might influence the nucleocytoplasmic shuttling dynamics of FOXO proteins, subsequently perturbing the functional FOXO pool. Canonically, 14-3-3 proteins are chaperones that bind to phosphorylated ligands, including FOXO transcription factors. For FOXO proteins in particular, 14-3-3 binding licenses FOXO for nuclear exclusion (40). AKT-mediated phosphorylation of FOXO induces 14-3-3 binding, preventing reentry of FOXO into the nucleus. In agreement with this hypothesis, immunofluorescence detection of FOXO1 in PC-3 cells showed that LINC00920 knockdown promoted nuclear retention of FOXO1 compared with cells transfected with nontargeting siRNA (Fig. 7A). These results suggest that LINC00920 plays a role in regulating FOXO1 activity by influencing its subcellular localization.

To determine whether LINC00920 affects FOXO/14-3-3 complex formation, we performed PLAs using primary antibodies against FOXO1, the most abundant FOXO isoform in PC-3 cells (Supplementary Fig. S5B), and 14-3-3ϵ. Cells depleted of LINC00920 by RNAi exhibited attenuated 14-3-3ϵ/FOXO1 interaction as visualized by significantly diminished PLA punctae (Fig. 7B and C). This trend was observed in both untreated and ipatasertib-treated cells, suggesting that the ability of LINC00920 to affect 14-3-3ϵ/FOXO1 binding is independent of AKT activity. Complementing these results, LINC00920 overexpression led to an increase in cellular punctae counts, indicating higher frequency of 14-3-3ϵ/FOXO1 interaction. Altogether, these findings are in-line with the proposed activity of LINC00920 in supporting the formation of the 14-3-3ϵ/FOXO1 complex.

A variety of roles for lncRNAs in cancer have emerged in recent years. In prostate cancer, multiple lncRNAs have been identified to correlate with the disease (19, 21). However, the number of differentially expressed, well-characterized lncRNAs remains disproportionate. Here, we identify LINC00920 to be upregulated in the ICGC-EOPC cohort. We show that ERG drives the expression of LINC00920 via direct transcriptional activation, resulting in the modulation of FOXO activity. Although, gene fusion events leading to ERG overexpression are observed in almost 50% of primary prostate tumors, the full consequences of this overexpression remain unclear (10, 11). Previous studies have implicated ERG expression with elevated DNA damage, increased cellular migration, and metastatic potential (8), which are in-line with the enrichment for ERG alterations in prostate cancer cells. These points, together with the results of our cellular assays, indicate that LINC00920 acts downstream of ERG to mediate tumorigenic ERG-associated effects. In particular, LINC00920 knockdown decreased cell proliferation, migration, and colony formation in PC-3 and DU-145 cells. Cell invasiveness appeared to be indifferent to LINC00920 levels, and this suggests that while LINC00920 can promote epithelial-to-mesenchymal transition (EMT)-associated properties (i.e., migration), the lncRNA does not trigger complete EMT that would allow increased invasion through a basement membrane matrix.

Through microarray analysis, we found that LINC00920 knockdown results in upregulation of PTEN signaling in PTEN-null PC-3 cells, suggesting that processes downstream of PTEN are affected by LINC00920. Indeed, the PTEN effectors and tumor suppressive FOXO transcription factors were activated, specifically their canonical gene targets GADD45A, BCL2L11, and PMAIP1 upon depletion of LINC00920. We used AKT inhibitors and tempered AKT hyperactivation brought about by PTEN deficiency in PC-3 cells to clarify and validate the derepressive effect of LINC00920 depletion on these FOXO targets. GADD45A encodes a tumor-suppressive protein implicated in DNA repair, maintenance of genomic stability, cell-cycle control, and apoptosis (41). GADD45A has been shown to be downregulated in primary prostate tumors compared with nonmalignant tissue (42). BCL2L11 (or BIM) and PMAIP1 (or NOXA) are essential proapoptotic proteins belonging to the BH3-only protein family. BH3-only proteins initiate the mitochondrial apoptotic pathway by activating Bax-like proteins or by binding and sequestering antiapoptotic Bcl-2 proteins (43). Upregulation of these genes could partly rationalize the cellular phenotypes observed upon LINC00920 knockdown. How these FOXO1 targets are activated can perhaps be attributed to enhanced nuclear accumulation of FOXO1 upon LINC00920 depletion, as we have observed in immunocytochemistry assays. Particularly, LINC00920 appears to promote the nuclear export of FOXO1, consequently attenuating its function as a transcription factor.

Because lncRNAs work in conjunction with proteins or other nucleic acids, identification of the lncRNA interactome is central to understanding their roles in tumor-related processes. To this end, we performed ChIRP-MS. We found that the majority of the captured proteins identified by MS are well known RBPs involved in RNA splicing and maturation. However, the consistent enrichment of 14-3-3 proteins in the ChIRP-MS experiments was notable in light of the connection between LINC00920 and FOXO activity, as well as the enrichment of 14-3-3 signaling signature in the microarray data. 14-3-3 proteins play important roles in the functional regulation of FOXO proteins. FOXOs harbor both a nuclear export signal at the C-terminus, and a nuclear localization signal proximal to the forkhead (FH) domain, permitting nucleocytoplasmic shuttling. This shuttling mechanism is dependent on FOXO phosphorylation, which is primarily mediated by AKT (44). The phosphorylated residues on FOXO act as docking points for 14-3-3 proteins whose binding initiates the formation of the nuclear export complex (40). Upon PI3K pathway activation, dual phosphorylation of FOXO by AKT at the C-terminal and FH domains triggers 14-3-3 binding (45). The nonexclusive localization of LINC00920 transcripts in the nuclear and cytoplasmic compartments is also consistent with this model.

We performed proximity ligation experiments to validate the supportive role of LINC00920 in the assembly of 14-3-3ϵ and a FOXO protein in PC-3 cells. On the basis of its abundance in these cells, we selected FOXO1 as the representative FOXO protein in the PLA experiments. Our results showed that low LINC00920 levels decreased the interaction between 14-3-3ϵ and FOXO1, while transcript overexpression had a concomitant positive influence on protein binding.

Several lncRNAs have been reported to mediate RBP interactions, and this capability is attributed to potential binding sites presented by secondary structures of RNA molecules. Indeed, RNA scaffolding was asserted to be an important regulatory mechanism and there is evidence that this process occurs frequently for most cellular processes (46). Interestingly, for established RBP-binding lncRNAs such as HOTAIR (47) and MEG3 (48), functional RNA–complex interactions can be facilitated by binding to only one component of the RBP. Moreover, the lncRNAs NRON (49) and RMRP (18) have been shown to play roles in subcellular mobilization of lncRNA–protein complexes. These lncRNA paradigms set the precedent and echo our proposed functional model of LINC00920 in prostate cancer cells.

The molecular function of LINC00920 provides a novel insight on how ERG facilitates its downstream effects through an lncRNA-mediated attenuation of FOXO1. Because FOXO1 signaling is increased upon LINC00920 depletion and LINC00920 binds to 14-3-3ϵ, we reason that LINC00920 enhances the formation of the 14-3-3ϵ/FOXO1 shuttling complex, ultimately leading to the nuclear exclusion of the tumor-suppressive FOXO1 (Fig. 7D). This model is consistent with the decline of FOXO signaling through the clinical course of prostate cancer, particularly in ERG-overexpressing cancer cells (50). Thus, LINC00920 overexpression provides a way for ETS-fusion–positive cells to circumvent the tumor suppressive influence of PTEN signaling by attenuating one of its downstream effectors.

No potential conflicts of interest were disclosed.

A.K. Angeles: Conceptualization, formal analysis, validation, investigation, visualization, methodology, writing-original draft, writing-review and editing. D. Heckmann: Conceptualization, supervision, investigation. N. Flosdorf: Validation, investigation. S. Duensing: Conceptualization, supervision, writing-review and editing. H. Sültmann: Conceptualization, resources, formal analysis, supervision, investigation, writing-review and editing.

We would like to thank the Microarray- and Mass Spectrometry–based Protein Analysis Units of the DKFZ Genomics and Proteomics Core Facility for providing the Illumina Whole-Genome Expression Bead chips, mass spectrometry analysis, and related services. We are likewise grateful to Sven Diederichs and Minakshi Gandhi for their technical support during the establishment of ChIRP experiments, and to Sabine M. Klauck for the critical reading of the article. We thank Sabrina Gerhardt for excellent technical support. A.K. Angeles was supported by the Helmholtz International Graduate School fellowship at the German Cancer Research Center.

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.