LINC00152 Promotes Invasion through a 3′-Hairpin Structure and Associates with Prognosis in Glioblastoma

Glioblastomas (GBM) are highly aggressive grade 4 gliomas and are the most common type of malignant glioma, with 10,000 new diagnoses each year (1). GBMs are a heterogeneous group of tumors that can be separated into four different subtypes, mesenchymal, classical, proneural, and neural, based on their transcriptional profile. Most of the focus on understanding glioma tumor biology has been on studying protein-coding genes and miRNAs (2). These efforts have identified commonly altered signaling pathways in GBMs, including mutations in EGFR, p53, and mTOR signaling (3, 4). Furthermore, miRNAs have been shown to play a role in many of the oncogenic phenotypes of GBMs, such as invasiveness and stemness of GBM stem cells (5, 6). Although there has been much effort on creating new targeted therapies for GBMs focusing on some of the aforementioned pathways, most have not been effective and the standard-of-care therapy, a combination of surgical resection, radiotherapy, and temozolomide, still leaves patients with a 5-year survival rate of roughly 10% (7).

High-throughput sequencing revealed that a majority of the human genome, long thought to be transcriptionally silent, is actually expressed. Indeed, when surveyed across many different cell types, it was found that nearly 80% of the human genome is actually transcribed (8). Many of these newly discovered transcripts are long noncoding RNAs (lncRNA). lncRNAs are a class of ncRNAs that are longer than 200 bases in length and can be further subdivided into subclasses based on chromosomal position relative to other genes, enhancers, or other genomic regulatory elements. lncRNAs have been shown to play many different functional roles in the cell, in part through regulation of transcription, mRNA stability, and mRNA translational efficiency (9, 10). Most of the research into the role of ncRNA in GBMs has been on miRNAs, with relatively few studies on lncRNAs. This leaves a crucial gap in our understanding of glioma pathogenesis. Indeed, lncRNAs have been shown to function in critical roles in a variety of tumor types, for example, HOTAIR in breast cancer, SChLAP1 in aggressive prostate cancer, MALAT1 in lung cancer, and DRAIC in prostate cancer (11–13).

LINC00152 is an lncRNA that was first identified as being hypomethylated during hepatocellular carcinoma tumorigenesis (14). It is also dysregulated in gastric cancer and esophageal squamous cell carcinoma (15, 16). However, there are conflicting reports on exactly how LINC00152 functions to promote the invasive phenotype. One study has argued that LINC00152 directly interacts with EGFR and affects AKT signaling, whereas others have suggested that LINC00152 acts as a competing endogenous RNA (ceRNA) through titrating miRNAs (5, 6, 17–20). Recently, we identified LINC00152 through an in-depth genomic analysis of gliomas as being highly expressed in GBMs (21). In this study, we characterize LINC00152's association with GBM clinical features and with tumor cell invasion and begin to functionally characterize LINC00152 structurally. Furthermore, we find that LINC00152 is overexpressed in 10 other tumor types compared with matched normal tissue and high LINC00152 expression is associated with a poor prognosis in 7 of these tumors.

U87 cells were maintained in MEM supplemented with 1% nonessential amino acids solution (catalog # 11140-050, Gibco), 1 mmol/L sodium pyruvate (catalog # 11360070, Gibco), 0.15% sodium bicarbonate (catalog # 25080094, Gibco), 10% FBS, and 1% penicillin/streptomycin.

For knockdown, U87 cells were transfected during two rounds of transfection. First, cells were reverse transfected with 40 nmol/L of siLINC00152_II (5′-UGACACACUUGAUCGAAUA-3′), siLINC00152_III (5′-CCGGAAUGCAGCUGAAAGA-3′), or a nonspecific siGL2 control siRNA (5′-CGUACGCGGAAUACUUCGA-3′) and 9 μL of Lipofectamine RNAiMAX transfection reagent (Thermo Fisher Scientific). Twenty-four hours later, a second round of transfection was performed using the same quantities of reagents. Twenty-four hours after the final transfection, cells were harvested and used for subsequent analysis.

A total of 500 ηg of LINC00152 or LINC00152-mutant pCDNA3-flag vectors was transfected into U87 cells using 2 μL of Lipofectamine 2000 (Thermo Fisher Scientific). Cells were harvested after 48 hours for downstream analysis.

Total RNA and nuclear/cytoplasmic RNAs were extracted using TRIzol total RNA isolation reagent (Thermo Fisher Scientific), Protein and RNA Isolation System (Thermo Fisher Scientific), respectively. RNA samples were treated with RQ1 RNase-Free DNase (Promega) according to the manufacturer's instructions. cDNA was produced from 1 μg RNA using Superscript III Kit (Thermo Fisher Scientific) according to the manufacturer's instructions. qPCR and Western blotting were performed according to standard protocols.

LINC00152 subcellular fractionation was performed using Protein and RNA Isolation System (Thermo Fisher Scientific) according to the manufacturer's instructions.

For in situ hybridization, 3 × 10⁵ U87 and U251 cells were plated on the top of a cover glass in a 6-well plate. On the next day, cells were washed once with PBS and fixed for 10 minutes with 2% paraformaldehyde. Then, cells were washed 3 times with PBS and incubated with 1 mL of permeabilization buffer (1× PBS/0.5% Triton X-100) for 10 minutes at 4°C. Cells were again washed 3 times with PBS. Next, cells were blocked with 1 mL of prehybridization buffer (3% BSA in 4× SSC) for 20 minutes at 55°C. Ten nanograms of LINC00152 or negative control probe was added to 2 mL of hybridization buffer (10% dextran sulfate in 4× SSC), and cells were incubated overnight at 55°C. On the next day, cells were washed 3 times for 5 minutes using washing buffer I (4× SSC, 0.1% Tween-20), 3 times for 5 minutes using washing buffer II (2× SSC), and 3 times for 5 minutes using washing buffer III (1× SSC). Subsequently, cells were blocked for 15 minutes at room temperature using 2 mL of blocking solution (4% BSA/1× PBS). Next, cells were incubated with 300 μL of antibody solution [2% of BSA/1 × PBS and digoxigenin (1:250)] for 1 hour. Cells were washed with 0.1% Tween-20/PBS 3 times and with alkaline Tris buffer for 5 minutes at room temperature. Finally, the signal was developed by adding 400 μL of BCP/NBT solution until the signal was visible.

For measuring cell growth, 1,000 cells were plated in quadruplets in 96-well plates, and cell growth was measured using standard MTT reagent (Promega). To measure invasion, 2 × 10⁵ U87 cells in serum-free media were seeded into 24-well Matrigel Invasion Chambers (BD Biosciences), and the bottom was filled with media and 10% FBS as the chemoattractant. Cells were allowed to invade for 8 hours and then fixed and stained with crystal violet/methanol and invaded cells were counted.

The expression of LINC00152 in GBMs and low-grade glioma (LGG) compared with normal brain and tumor subtypes was performed as described previously (21). Expression of LINC00152 in all other The Cancer Genome Atlas (TCGA) tumors was determined by comparing expression data of only those tumors that had a matched normal tissue sample. Statistical significance was determined using a paired t test. TCGA patient survival data for GBMs and LGGs were retrieved from cBioPortal (www.cbioportal.org), and survival data for the remaining tumor types were retrieved from OncoLnc (www.oncolnc.org) in December 2016 (22–24). The expression threshold used to separate patients is outlined in the main text. Kaplan–Meier plots, HRs, and P values based on these separations were generated using the “survminer” package for R.

U87 cells were treated with a combination of the two siRNAs as mentioned earlier and total cell RNA was isolated using TRIzol and subsequently purified using RNeasy Isolation Kit (Qiagen). Sequencing libraries were generated using NEB NEXT Ultra Directional RNA Library Prep Kit and samplers were barcoded with NEBNext Multiplexing oligos per standard manufacturer protocols. Libraries were sequenced with 75-bp paired-end reads NextSeq500 instrument, in the Biomolecular Analysis Facility, University of Virginia School of Medicine. Sequencing reads were aligned to the hg38 reference genome using HISAT (25). Gene abundances and identification of differentially expressed genes were performed using HTSeq and DESeq2 (26, 27). An adjusted P value (obtained by DESeq2) cutoff of 0.05 and log₂ fold change of 2 was used to define differentially expressed genes. GSEA analysis was performed on preranked gene list based on fold change (siLINC00152/siGL2) against 50 hallmark gene sets (28). For plotting enrichment score obtained by GSEA analysis (as shown in Fig. 4B), we have included only the genes that are either induced or repressed 1.5-folds upon siLINC00152. The raw and processed data were deposited in Gene Expression Omnibus (GEO) under accession number GSE111652.

Secondary structure predictions of LINC00152 were determined using mfold (29). The two structures with the lowest predicted free energies were selected for comparisons with psoralen analysis of RNA interactions and structures (PARIS) and Ribo-seq. For PARIS data analysis of LINC00152, raw sequencing data from Lu and colleagues were aligned to the hg19 genome using STAR (spliced transcripts alignment to a reference) with the alignment parameters outlines in Lu and colleagues' work (30, 31). Aligned reads were then processed to identify gapped mapping to LINC00152 and visualized with IGV (32). We used ribosome profiling data from Gonzalez and colleagues and aligned reads to the hg19 genome using HISAT2 (33). We then examined reads that mapped to LINC00152 for their distribution along the message to ensure that they were not legitimate ribosome footprints using IGV (32). The predicted secondary structure elements and protein-bound region were then compared with the in silico secondary structure predictions.

We first identified LINC00152 from a comprehensive analysis of lncRNAs in gliomas (21). LINC00152 was one of the most differentially expressed lncRNAs in GBMs compared with normal brain tissue; however, it is not upregulated in grade 2 and 3 gliomas (Fig. 1A; Supplementary Fig. S1A). We have validated the upregulation of LINC00152 in an independent set of patients with GBM compared with normal FFPE brain tissue (21).

We tested whether LINC00152 is preferentially expressed in a particular GBM subtype, but that did not appear to be the case. The differences in LINC00152 expression between the subtypes were not statistically significant, although the median expression of LINC00152 is lowest in the proneural GBM subtype (P < 0.1; Supplementary Fig. S1B). Even though LINC00152 is not upregulated in LGGs as a whole, the IDHwt LGG subtype expresses 4 times as much LINC00152 as normal brains (P < 0.00001; Fig. 1B). This is interesting, because IDHwt LGGs are far more aggressive than the other LGG subtypes and display clinical properties similar to GBMs (34).

Because LINC00152 is upregulated in brain tumors compared with normal brain tissue, we next elucidated the association of LINC00152 association with survival of patients with GBM and LGG. To do this, we assessed the survival difference of patients expressing high (top 33% highest expressing LINC00152 cohort) and low level of LINC00152 expression from the TCGA for both GBM and LGG. In GBMs, patients who had high expression of LINC00152 had a poor prognosis (P = 0.02) compared with the patients expressing low level of LINC00152, with a median survival of 11.9 and 15.4 months, respectively (Fig. 2A). Furthermore, LINC00152 expression was also able to separate patients into two distinct prognostic groups in LGGs. LGG patients with high expression of LINC00152 had a median survival of 62.1 months, whereas the low expressing group had a median survival of 98.2 months (P < 0.0001; Fig. 2B). These results demonstrate that not only is LINC00152 overexpressed in gliomas, but that this overexpression is associated with poor patient outcome.

It was intriguing to examine LINC00152 expression in other cancers compared with their respective normal tissues. We compared the expression of LINC00152 in all TCGA tumor samples with paired normal and tumor RNA sequencing (RNA-seq) data. Surprisingly, LINC00152 is upregulated in nearly every tumor type we analyzed, including head and neck squamous carcinoma, renal papillary tumor, hepatocellular carcinoma, colorectal carcinoma, renal clear cell carcinoma, breast-invasive carcinoma, stomach adenocarcinoma, uterine carcinoma, thyroid carcinoma, and lung adenocarcinoma (Fig. 1C–L).

Because LINC00152 is overexpressed in the majority of tumors that we have analyzed, we next wanted to determine whether LINC00152 expression is associated with patient survival in the TCGA tumors that had higher levels of LINC00152 compared with the paired normal samples. To do this, we performed Kaplan–Meier analysis for each tumor type by separating patients into two groups, the top quartile LINC00152-expressing tumors and the lowest quartile LINC00152-expressing tumors. From the original list of tumors, LINC00152 expression was associated with poor patient outcome in head and neck squamous cell carcinoma, lung adenocarcinoma, renal clear cell carcinoma, and hepatocellular carcinoma (Fig. 2C–F). The poor outcome of patients with renal papillary carcinoma was not statistically significant comparing the top and bottom quartiles of LINC00152 expression (P = 0.1), but the poor outcome was statistically significant (P = 0.015) when we compared patients in the top third and bottom third based on LINC00152 expression (Supplementary Fig. S1C).

Although LINC00152 was not overexpressed in LGGs relative to normal brain, it was upregulated in an aggressive subpopulation of LGGs (those with IDH wild type) and was associated with poor patient outcome. This made us realize that even if a tumor type does not overexpress LINC00152 globally relative to normal tissue, overexpression of the lncRNA in specific tumors may still be associated with poor outcome. We therefore examined other TCGA tumors that did not show a global increase of LINC00152 expression in the cancers relative to normal tissue for the predictive value of the expression of this lncRNA. Interestingly, even among these tumors, LINC00152 expression was associated with poor patient outcome in pancreatic adenocarcinoma when we compare the tumors in the top third and bottom third (Supplementary Fig. S1D), and acute myeloid leukemia, with the top quartile and bottom quartile for LINC00152 expression (Supplementary Fig. S1E). These results highlight the fact that in nine tumor types (GBMs, LGGs, head and neck squamous cell carcinoma, renal clear cell carcinoma, hepatocellular carcinoma, lung adenocarcinoma, renal papillary carcinoma, pancreatic adenocarcinoma, and acute myeloid leukemia), LINC00152 appears to function as unfavorable gene whose expression is associated with a poor patient outcome.

Subcellular fractionation (Fig. 3A and B) and in situ hybridization (Fig. 3C) revealed that LINC00152 is primarily localized in the cytoplasm of U87 cells. We next sought to determine whether the upregulation of LINC00152 seen in GBMs is associated with any cancer phenotypes in GBM cell lines. LINC00152 has previously been shown to affect multiple cellular phenotypes, including cell growth, migration, invasion, and epithelial-to-mesenchymal transition (EMT; refs. 35, 36). We knocked down LINC00152 expression using two separate siRNAs or overexpressed the lncRNA and found that LINC00152 knockdown or overexpression did not affect cell proliferation for a period of 10 days (Supplementary Fig. S2B and S2D). We next assayed whether LINC00152 expression was associated with tumor cell invasion using a transwell migration assay. Knockdown of LINC00152 in U87 cell lines led to a statistically significant reduction in cell invasion with both siRNAs targeting LINC00152 (Fig. 3D and E). Conversely, overexpression of LINC00152 led to an increase of over 2-fold in the number of invaded cells (Fig. 3F and G). These findings suggest that LINC00152 knockdown decreases invasion of GBM cells, whereas upregulation in GBMs promotes the invasive phenotype that is commonly seen in patient tumors.

To better understand how LINC00152 affects cellular invasion, we performed RNA-seq on U87 following knockdown of LINC00152 using a combination of two different siRNAs. Knockdown of LINC00152 leads to large changes in gene expression, with 259 genes significantly upregulated and 295 downregulated at least 2-fold (Fig. 4A). Thus, to determine the most significant molecular pathways regulated by LINC00152, we performed gene set enrichment analysis (GSEA), a method that can identify pathway enrichment from fold change–based preranked gene list from RNA-seq (28). This analysis showed a significant enrichment of upregulated genes upon siLINC00152 involved in EMT (Fig. 4B). Among the differentially expressed genes involved in EMT, the changes were validated by qPCR on 12 of 13 genes after siLIN00152 treatment (Supplementary Table S1). More interestingly, six of the genes that were downregulated by LINC00152 knockdown were conversely upregulated by overexpression of the lncRNA: TPM2 (Tropomyosin 2), PTX3 (Pentraxin 3), IGFBP4 (Insulin growth factor binding protein 4), TGM2 (Transglutaminase 2), SPP1 (Secreted phosphoprotein 1), and LUM (Lumican; Supplementary Table S1). Moreover, overexpression of the siRNA-resistant M8 was sufficient to upregulate these genes even after knockdown of endogenous LINC00152 (Supplementary Fig. S3). These results indicate that LINC00152 may induce U87 cell invasion by regulating the expression of at least these six genes.

Several previous studies have suggested that LINC00152 acts as an miRNA sponge by titrating different miRNAs (miR-376c-3p, miR-4775, miR-4767, miR-138-5p, miR-103, and miR-205) in different types of tumors, including GBMs (5, 6, 17–20). However, suggestions that an lncRNA acts as an miRNA sponge are sometimes questioned because the abundance of the lncRNA is often far less than that of the targets of the miRNAs and of the miRNAs themselves. If LINC00152 acts as an miRNA sponge in U87 cells, we would expect that the targets of these miRNAs would be repressed upon knockdown of the lncRNA and the subsequent release of the miRNAs from interaction with the lncRNA. However, we find that there is a statistically significant upregulation of the targets of these six miRNAs compared with nontargets when LINC00152 is knocked down, ruling out the possibility of LINC00152 acting as a ceRNA for these miRNAs (Fig. 4C).

Over the past decade, several new technologies have been developed to examine the secondary structures of lncRNAs on a global basis, one such technique is PARIS (30). PARIS is based on reversibly crosslinking RNA duplexes (stems of stem-loops) and gentle digestion with a single-strand RNase, S1 nuclease, to cut looped single-stranded portions of an RNA's secondary structure. The surviving RNA duplexes from the stems are then ligated to each other and subjected to high-throughput sequencing. RNAs containing stem-loops will have sequencing reads corresponding to the stems with gaps (corresponding to the loops) that do not overlap with a splice site. We analyzed publicly available PARIS data from HeLa cells to determine whether LINC00152 contains any secondary structure elements that could be detected by PARIS. Following alignment, we identified reads with a 2-nt gap that were present in the PARIS libraries (Supplementary Fig. S4C). These reads are positioned from position 285 to 373 of the 496 nt long LINC00152, with a small 2 base gap starting at position 342 (Fig. 5A). Sequence analysis of this region revealed some complementarity, suggesting that this region might in fact form a stem-loop structure (Fig. 5A).

To get a better understanding of overall LINC00152 secondary structure, we used publicly available RNA secondary structure prediction tool, mfold, to identify secondary structure predictions for LINC00152 that are consistent with a stem-loop being present from 285 to 373 (29). The top 2 secondary structures with the lowest free energy differed in their exact base-pairing, but the overall stem-loop structure was largely the same. Importantly, both structures were consistent with a stem-loop being present from position 285 to 373 (Fig. 5A; Supplementary Fig. S4A and S4B). Furthermore, the resulting loop from the stem formation is rather small, 4 nt, which is consistent with the small 2-nt gap seen by PARIS.

We next asked whether we could use a separate method to independently validate the hairpin formation in LINC00152. Ribo-seq (ribosome profiling) is a technique that has been used to identify RNAs that interact with the ribosome and how the ribosome is distributed across those RNAs (37). This information has also been used to ascertain that some lncRNAs are associated with ribosomes, but not translating ribosomes (38). Recently, it was determined that the polysomes isolated for Ribo-seq are contaminated with other ribonucleoprotein (RBP) complexes. As a result, RNA footprints from RBPs that are not ribosome proteins can be detected in Ribo-seq data (33). To determine whether we could identify RBP–RNA footprints from LINC00152, we analyzed publicly available Ribo-seq data from normal brain samples (39). In two of the three normal brain Ribo-seq samples, we detected an RBP footprint at positions 303–330 of LINC00152. In addition, in one of the samples, there was an RBP footprint from 354 to 382 (Fig. 5A; Supplementary Fig. S4D). These two footprinted areas are located on opposite strands of the same stem-loop that was detected by PARIS, providing additional evidence of the existence of this stem-loop and suggesting that this stem is bound by a protein in an RBP (Fig. 5A).

To determine whether this newly identified, potentially protein-bound, stem-loop plays a role in LINC00152 function, we created a series of LINC00152 deletion mutants (Fig. 5B; Supplementary Fig. S5A). The sites of the deletions were chosen based on PARIS and Ribo-seq analysis as well as two in silico predicted structures of LINC00152 (Fig. 5A; Supplementary Fig. S4C and S4D). We assessed whether independent overexpression of the mutants was able to stimulate U87 cell invasion. Overexpression of M2 (which removed the minimal amount of the protein bound stem-loop, nucleotides 280–401) or M3 (which removed the stem-loop and the remaining 3′ end) led to a decreased cell invasion significantly compared with full-length LINC00152 (P < 0.05; Fig. 5D). On the other hand, the mutant M4 (which removed the 3′ end but preserved the stem-loop) or M7 (which removed the extreme 3′ end, and also preserved the stem-loop) increased U87 cell invasion. Other deletion mutants that removed regions of LINC00152 5′ to the stem loop (M5 or M6) stimulated cellular invasion to a similar extent as full-length LINC00152. Finally, overexpression of M8, containing only the protein-bound stem-loop (nucleotides 280–401) was sufficient to stimulate invasion of U87 cells (Fig. 5D). These results suggest that M8 stem-loop is necessary and sufficient for stimulation of cell invasion.

Consistent with this conclusion, overexpression of the stem-loop also induced the six genes involved in EMT to the same extent as the full-length LINC00152 (Supplementary Table S1). In addition, knockdown of LINC00152 by siLINC00152_II (a siRNA that targets a region on LINC00152 outside of M8) decreased cell invasion, while the siRNA-resistant M8 was sufficient to rescue cell invasion (Fig. 5E).

We next tested with point mutations whether the ability to stimulate invasion of U87 cells depends on the LINC00152 stem-loop structure. Two mutants on opposite side of the stem disrupt the stem-loop (mutA: changes bases 333–336 and mutB: changes bases 349–352; Fig. 6D). Neither mutA nor mutB stimulated the invasion of U87 cells as well as full-length LINC00152 or M8 (Fig. 6A). In contrast, when the two mutations were combined in mutAB, the stem-loop structure was reconstructed and this promoted invasion to the same extent as full-length LINC00152 or M8 (Fig. 6A). Therefore, we can conclude that the stem-loop structure itself is essential for LINC00152 to stimulate cellular invasion.

As reported previously (40), LINC00152 is a close homolog of another lncRNA on chromosome 2, MIR4435-2HG; both have nearly identical sequences (with only 6 base mismatches) and both contain M8 sequence (Fig. 7A). LINC00152 and MIR4435-2HG are both transcribed from chromosome 2, LINC00152 is located at chr2: 87455476-87606739, and MIR4435-2HG is transcribed from chr2:111196350-111495115. In order to estimate the expression level of these two RNAs, we considered RNA-seq reads that uniquely mapped without any mismatch to either LINC00152 or MIR4435-2HG. This analysis showed that LINC00152 and MIR4435-2HG are expressed at the same level in U87 cells, and both of the RNAs are knocked down upon treatment of siLINC00152 to a similar extent (Fig. 7B and C). Thus, the phenotype that we observe with siRNA directed toward LINC00152 is also likely through knocking down the highly similar MIR4435-2HG. Moreover, given the high similarity between the two transcripts, and the fact that, from nucleotides 382 to 478, MIR4435-2HG forms a 97 nucleotides long stem-loop in the same position as LINC00152, it is likely that MIR4435-2HG overexpression phenocopies the effects of LINC00152 on cell invasion. However, we see an increase in cell invasion when we exogenously express LINC00152, saying that LINC00152 by itself can promote cell invasion. Again, upon considering uniquely mapped reads in TCGA RNA-seq data, we found that both LINC00152 and MIR4435-2HG are equally expressed in LGG and GBM (Fig. 7D). Analysis of TCGA RNA-seq data also revealed a positive correlation between the expression of the two RNAs in GBM and LGG (Fig. 7E and G), suggesting that these two RNAs may be coregulated. Moreover, the Kaplan–Meier plot to estimate survival showed that the expression of either RNA is associated with poor patient survival (Fig. 7G and H).

The human genome was once thought to be mainly dormant and most of the transcription was devoted in producing protein-coding genes. We now know that the genome is transcriptionally vibrant and only a small fraction of the expressed genome, roughly 2%, encodes for protein-coding genes. GWAS and high-throughput sequencing studies have found that many of the genomic lesions and expression alterations seen in cancer and other pathologies fall within nonprotein-coding regions of the genome and may lead to dysregulation of ncRNAs (41–43). Furthermore, there is a growing body of evidence implicating lncRNAs in playing a direct role in normal cellular physiology, as well as driving pathogenesis in a variety of disorders, including cancer (43–46). Indeed, recent work has illustrated the critical role that lncRNAs play in cancer, including iconic examples such as HOTAIR in breast cancer and HULC in hepatocellular carcinoma and DRAIC in prostate cancer (11, 14, 47).

In this study, we have shown that LINC00152 is an lncRNA that is upregulated in many different cancer types and is highly upregulated in GBMs. Although LINC00152 is not upregulated in all LGGs relative to normal brain tissue, it is upregulated in the highly malignant IDHwt LGG subtype, further supporting LINC00152's association with aggressive tumors. This raised the interesting possibility that in tumors where LINC00152 is not differentially overexpressed or is moderately upregulated in the tumor population relative to normal tissue, LINC00152 could still be highly upregulated in a more aggressive subgroup of the tumors. This was indeed found to be true in pancreatic adenocarcinomas and acute myeloid leukemias. LINC00152 expression is associated with patient survival in nine different cancer types, including GBMs and LGGs. LINC00152 expression promotes cell invasion, which is consistent with its association with poor patient outcomes.

To assess the coding probability of LINC00152, we used the Coding Potential Assessment Tool (CPAT) and found a score of 0.0289, suggesting that this lncRNA has no coding potential, because a CPAT score of <0.5242 is considered noncoding (48). In addition, ExPasy (49) predicts the first LINC00152 ORF of 42 amino acids (126 nucleotides) from nucleotides 64 to 189. This ORF is not similar to any ORF predicted for the mouse transcript Gm14005. Moreover, blastx of the translated 126 nucleotide ORF did not show any hits in mouse protein database. In a different frame, there is a longer ORF of 92 amino acids (which also does not have a homolog in the mouse transcript) that is preceded by a short ORF that has three stop codons. So, the longer ORF is also unlikely to be translated.

Previous studies have shown that LINC00152 is an oncogenic lncRNA involved in regulating invasion in different types of tumors (5, 6, 18–20), including gliomas (5, 6). Yu and colleagues (5) have reported an in vivo tumor xenograft study; downregulation of LINC00152 produced smaller tumors and increased survival rates when compared with control. Thus, our findings reinforce the idea that LINC00152 is an oncogenic lncRNA that is associated with aggressive tumors by promoting cell invasion. Moreover, through analysis of global RNA structure mapping and RNA–protein interaction data, we identified a protein-bound stem-loop in the 3′ region of LINC00152. The structure–function analysis demonstrated that this stem-loop is necessary and sufficient for stimulating invasion of U87 cells, that it can rescue the loss of invasion seen after knockdown of LINC00152, and that the base-pairing of the opposite strands of the stem-loop, rather than the sequence at the mutated sites, is more important for stimulating invasion of U87 cells. However, it is likely that there are specific sequences along the M8 stem-loop that are important for LINC00152 function that will be examined in a future study.

GSEA of RNA-seq from LINC00152 knocked down cells also supports the idea that LINC00152 is involved in promoting invasion. More specifically, TPM2, PTX3, IGFBP4, TGM2, SPP1, and LUM were downregulated by siLINC00152 and upregulated after LINC00152 overexpression. Because we did not compare the global gene expression changes with LINC00152 overexpression and knockdown by RNA-seq, there are likely to be many other genes that will be regulated similarly to the six genes we tested in this study. Because siLINC00152 decreased invasion, we focused on genes in the RNA-seq data whose change (up or down) will decrease invasion. Of these, 13 transcripts qRT-PCR after siLINC00152 validated the changes in 12. Out of these 12, 6 genes were changed in the opposite direction when LINC00152 full length or M8 was overexpressed (Supplementary Table S1).

In addition, despite previous suggestions, analysis of RNA-seq showed us that LINC00152 is not acting as a ceRNA that sponges miRNAs.

LINC00152 is expressed from a syntenic location from mouse transcript. Mouse has a single gene, MIR4435-2HG (Gm14005 or MORRBID), but humans have two closely related genes, LINC00152 and MIR4435-2HG.

MIR4435-2HG is a host gene for an miRNA, as miR-4435 is transcribed from an intron of MIR4435-2HG. However, none of the effects observed by overexpression of LINC00152 are due to miR-4435, because the functional assays were done using the cDNA of LINC00152. Furthermore, miR-4435 is not detected in any of the short RNA-seq libraries (such as miRGator v3.0 and miRmine) from brain, glial cell lines, and gliomas.

Mouse MIR4435-2HG has been proposed to be a prosurvival lncRNA that represses a gene in cis, the proapoptic gene BCL2L11 (BIM) by recruiting the polycomb repressive complex, PRC2, to the BCL2L11 promoter (50). We considered the possibility that LINC00152, although cytoplasmic, is acting as a prooncogenic RNA by similarly suppressing BCL2L11. siLINC00152 increases BCL2L11 RNA (Supplementary Fig. S6B–S6D), but this is not expected to decrease cell invasion. Second, overexpression of LINC00152 from a heterologous site or the M8 hairpin of the lncRNA did not decrease BCL2L11 (Supplementary Fig. S6E–S6G) and yet increased cell invasion. Third, analyzing previously published mouse PAR-CLIP data, we determined that EZH2 from the polycomb complex associates with an intronic region of MORRBID (mouse MIR4435-2HG/MORRBID), which is not near the M8 region. Fourth mouse MIR4435-2HG encodes only the first third of the M8 hairpin that we have found to be functions in human LINC00152 or human MIR4435-2HG. Finally, LINC00152 is predominantly cytoplasmic, arguing against any role in recruiting any factors to the genome. Collectively, these results suggest that interaction with PRC2 is not necessary for the stimulation of invasion seen upon overexpression of the LINC00152 or the M8 hairpin RNA.

In conclusion, LINC00152/CYTOR and its homolog MIR4435-2HG functions as an oncogenic lncRNA in GBMs through the action of a protein-bound stem-loop and potentially plays a critical oncogenic role in a wide variety of cancer types. The results rule out a mechanism of action involving the sponging of miRNAs as proposed in the literature, or interaction with the Polycomb complex proposed for the mouse MIR4435-2HG/MORRBID RNA. LINC00152 could also serve as a tumor biomarker or a target for future cancer therapeutics.

No potential conflicts of interest were disclosed.

Conception and design: B.J. Reon, A. Dutta

Development of methodology: B.J. Reon, A. Dutta

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): B.J. Reon, B. Takao Real Karia, A. Dutta

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): B.J. Reon, B. Takao Real Karia, M. Kiran, A. Dutta

Writing, review, and/or revision of the manuscript: B.J. Reon, B. Takao Real Karia, M. Kiran, A. Dutta

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Kiran, A. Dutta

Study supervision: A. Dutta

The authors thank Dutta laboratory members for advice and helpful discussions. This work was supported by grants from the NIH R01 CA166054, AR067712, and a V foundation award D2018-002 to A. Dutta. B. Takao Real Karia was partly supported by CAPES BEX 0320-13-7. M. Kiran was partly supported by a DOD award PC151085.

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.