MicroRNA-138 suppresses glioblastoma proliferation through downregulation of CD44

Tumor suppressive microRNAs (miRNAs) are increasingly implicated in the development of anti-tumor therapy by reprogramming gene network that are aberrantly regulated in cancer cells. This study aimed to determine the therapeutic potential of putative tumor suppressive miRNA, miR-138, against glioblastoma (GBM). Whole transcriptome and miRNA expression profiling analyses on human GBM patient tissues identified miR-138 as one of the significantly downregulated miRNAs with an inverse correlation with CD44 expression. Transient overexpression of miR-138 in GBM cells inhibited cell proliferation, cell cycle, migration, and wound healing capability. We unveiled that miR-138 negatively regulates the expression of CD44 by directly binding to the 3′ UTR of CD44. CD44 inhibition by miR-138 resulted in an inhibition of glioblastoma cell proliferation in vitro through cell cycle arrest as evidenced by a significant induction of p27 and its translocation into nucleus. Ectopic expression of miR-138 also increased survival rates in mice that had an intracranial xenograft tumor derived from human patient-derived primary GBM cells. In conclusion, we demonstrated a therapeutic potential of tumor suppressive miR-138 through direct downregulation of CD44 for the treatment of primary GBM.

, further study was needed to clarify the role and druggability of miR-138 in GBM. For the first time, this study found from whole transcriptome and miRNA expression profiling analyses that miR-138 is downregulated in human primary GBM tissues and primary GBM cells derived from human patients with an inverse correlation with the expression of CD44. Ectopic restoration of miR-138 negatively modulated the expression of CD44 in GBM cells leading to the inhibition of cell proliferation and increased survival rate in orthotopic xenograft mouse model. We discovered that miR-138 directly regulates CD44 expression by binding to the 3′ UTR of CD44. For many years, activation of CD44 pathways has been implicated in GBM as responsible for various tumor supportive signaling pathways, such as cell motility, cell proliferation, invasion, angiogenesis and drug resistance 18 . Although specific inhibition of CD44 using antibody has been shown to decrease cancer cell proliferation, first humanized CD44-targeting monoclonal antibody developed by Roche (RO5429083) is still under clinical trials (ClinicalTrials.gov Identifiers: NCT01641250, NCT01358903) 19 . Our data strongly suggest that inhibition of CD44 signaling pathway through the restoration of miR-138 can benefit therapeutic outcomes in clinical settings for GBM patients.

miR-138 is downregulated in glioblastoma.
To determine whether miR-138 is tumor suppressive or tumor supportive in GBM, miRNA expression profiles were obtained by NanoString miRNA Expression Analysis from 13 frozen brain tissue samples (9 GBMs and 4 controls) (Supp Table 1). Total of 827 miR-NAs were detected and the miRNA expression profiles were then subjected to principal component analysis (PCA) for quality control, which clearly classified the samples into two groups (Fig. 1A). The high expression level of miR-21 was detected from the 9 GBM samples as previously known that miR-21 is one of the most www.nature.com/scientificreports/ highly overexpressed miRNAs in most types of human cancer including GBM [20][21][22] . A total of 89 differentially expressed miRNAs (DEmiRs) were detected by using DEseq2 R package with BH adjusted p-value < 0.01 and two fold changes (Supp Table 2) 23 . Among them, miR-138 was identified as one of the down-regulated miR-NAs in the GBM samples with a reverse correlation with miR-21 (Fig. 1B). Individual miRNA expression analysis by TaqMan quantitative real-time PCR (qRT-PCR) further confirmed that the relative fold change of miR-138 in the GBM patient samples was significantly down-regulated (0.009915 ± 0.003592, n = 9) when compared to negative control samples (0.3525 ± 0.08532, n = 4) (Fig. 1C, p < 0.001). We further examined the expression level of miR-138 in human patient-derived primary GBM cells, and found that the miR-138 expression was observed lower in 4 patient-derived primary GBM cells (GBM8 (0.1262 ± 0.004762), GBM12 (0.06386 ± 0.004912), GBM28 (0.07143 ± 0.004731) and GBM43 (0.08453 ± 0.002392)) than that in human neural stem cell (hNSC (1.537 ± 0.3098), a non-malignant negative control (Supp Fig. S1). The lower expression level of miR-138 was also observed in 6 glioma cell lines (LN229 (0.081 ± 0.005741), T98G (0.06172 ± 0.001572), U87 (0.07579 ± 0.002727), U373 (0.4155 ± 0.04372), LN18 (0.04435 ± 0.001073) and U251 (0.4743 ± 0.03747)) (Supp Fig. S1). From the public database, The Cancer Genome Atlas Program (TCGA) from the Betastasis (http:// www. betas tasis. com/ glioma/ tcga_ gbm/), the expression of miR-138 was also found to be lower in most brain tumor types than normal brain tissues (Fig. 1D, p < 0.001). Another public database presenting single cell RNA sequencing (scRNA-Seq) profiles from 3589 single cells of GBM patient summarized 2D-tSNE representation showed that miR-138 expression is barely detectable while the expression level of miR-21 is high in neoplastic cell population (Supp Fig. S2). Taken together, our studies indicated that miR-138 is down-regulated in GBM as coincided with public databases as well as previous reports 10,14,24 , implying a possible tumor suppressive role of miR-138 in GBM tumor.

miR-138 restoration inhibits GBM cell proliferation.
To assess a tumor suppressive potential of miR-138 in GBM, miR-138 was ectopically expressed in GBM cells to assess its impact on GBM cell proliferation. GBM patient-derived primary cells (GBM 12, 28 and 43) was transiently transfected with miR-138 mimics in comparison to scrambled negative control (miR-Ctrl). After 4 days from the transfection, all GBM cells showed a significant sensitivity to the miR-138 overexpression ( Fig. 2A). The inhibition of GBM cell proliferation was dependent on the concentration of miR-138 transfected, while the concentration dependency was not observed with miR-Ctrl. Same results were also observed from 5 GBM cell lines (LN229, U87, T98G, LN18 and U373) when transiently transfected with miR-138 (Supp Fig. S3). Using GBM12-RFP cell expressing red fluorescence protein (RFP) for live cell imaging, it was shown that the proliferation of GBM cells stopped growing after 48 h post transient transfection of miR-138 mimics (Fig. 2B). After 4 days from the miR-138 transfection, lower numbers of GBM12-RFP cells survived than the cells treated with miR-Ctrl (Fig. 2B,C). Again, the proliferation inhibition of GBM12-RFP cells was depending on the treated concentration of miR-138 (Supp Fig. S4). When the primary GBM cells (GBM12, 28 and 43) with transient overexpression of miR-138 were stained with Annexin V-PI, apoptotic cell populations were significantly increased (39.3 ± 0.4%, 31.1 ± 1.8%, 22.3 ± 0.4%, respectively) compared to the GBM cells transfected with miR-Ctrl (20.0 ± 2.4%, 13.2 ± 5.0%, 6.5 ± 0.2%, respectively) (Fig. 2D). Transient restoration of miR-138 also reduced the migration (Fig. 2E) and wound healing capability (Fig. 2F) of GBM cells. These data strongly indicated that miR-138 possesses a tumor suppressive role in GBM through an inhibition of cell proliferation.
CD44 expression is inversely correlated with miR-138 expression in GBM. Determination of direct targets of a miRNA is a key information to understand an underlying mechanism by which miR-138 plays a tumor suppressive role in GBM. To identify candidate target genes of miR-138, total RNA sequencing (RNA-seq) was performed on the 13 patient samples. Through bioinformatics analysis on the RNA-seq obtained from human gBM samples (n = 9) in comparison to those from normal (n = 4), total of 420 genes were identified as differentially expressed genes (DEGs) (Supp Table 3). Among the top 5 out of 420 DEGs, we noticed that CD44 is predicted as a significant target of miR-138 by prediction software, mirDIP (microRNA Data Integration Portal) (Supp Table 4). Individual gene expression analysis by qRT-PCR confirmed that the expression level of CD44 is higher in GBM samples (11.06 ± 0.1995, n = 9) than that in control samples (7.168 ± 0.7913, n = 4) (Fig. 3A, p < 0.001). TCGA database also showed the CD44 expression in GBM samples was at substantially higher level (2329 ± 76.09, n = 200) than that of control samples (299.9 ± 50.01, n = 9) (Fig. 3B, p < 0.001). Single cell-RNA sequencing (scRNA-Seq) database also showed the overexpression of CD44 in neoplastic cell population in GBM patients (Supp Fig. S5). When plotted the expression levels of miR-138 and CD44 in the human GBM and normal samples (n = 13), it clearly showed an inverse correlation between miR-138 and CD44 (Fig. 3C, Slope = − 0.08235, R 2 = 0.8863, p < 0.001). CD44 has been known to be overexpressed in GBM as well as many cancers 18 . As a cellular receptor of hyaluronan, the overexpression of CD44 involves in cell proliferation signaling pathway in neoplastic cells 25 . Gene Set Enrichment Analysis (GSEA) on the RNA-seq data revealed that genes involving in hyaluronan metabolic process including CD44 are highly enriched in the GBM samples compared to control samples with the normalized enrichment score (NES) of 2.13 and false discovery rate (FDR) of 0.0012 (Fig. 3D). These results indicate the inverse correlation between miR-138 and CD44. The specific regulation between the two molecules during GBM progression requires further investigation.
CD44 is a direct target of miR-138. The inverse expression correlation between miR-138 and CD44 suggests that CD44 is a possible target gene of miR-138. CD44 overexpression and the activation of CD44 signaling pathway have been implicated in GBM with pro-survival functions 18 Fig. S6), further supporting a negative regulation of CD44 by miR-138. Since miRNAs regulate the target gene expression through a direct binding of miRNA to the 3′ UTR of target gene mRNA, dual luciferase reporter assay was performed with a luciferase vector plasmid containing the 3′ UTR sequence of human CD44. Bioinformatics search revealed that the 3′ UTR region of human CD44 contains two putative binding sites (CD44 364-370 and CD44 753-759) matched with the seed sequence of miR-138-5p (Fig. 3G). When GBM cells were co-transfected with miR-138 mimics and the luciferase reporter vector plasmids, the luciferase activity was significantly decreased by miR-138 indicating a direct interaction between miR-138 and the 3′ UTR region of CD44 (Fig. 3H). The repression was rescued by mutations in the putative binding sites, since miR-138 did not decrease the luciferase activity of GBM cells transfected with mutated reporter vector plasmids (Fig. 3H). These data clearly demonstrated that miR-138 down-regulates the expression of CD44 through direct binding to the 3′ UTR region of CD44.

CD44 inhibition by miR-138 induces cell cycle arrest through p27 activation.
To understand the working mechanism behind the miR-138 mediated inhibition of GBM cell proliferation, CD44 signaling pathway was further scrutinized. Transfection of miR-138 in GBM cells reduced cell viability, while CD44 restoration . Among the downstream effectors of CD44 signaling pathway, the protein expression level of cell cycle inhibitor p27 was found to be seriously induced after miR-138 transfection in all tested GBM cells (GBM12, GBM28 and GBM43) (Fig. 4C). It was previously known that CD44 negatively regulates the expression and activation of p27 through AKT pathway 26 . However, restoration of CD44 expression level by co-transfection of GBM cells with cCD44 and miR-138 reversed the increase of p27 expression level through the gain of AKT activation (Fig. 4C). Since the activated p27 translocates into nucleus to inhibit cell cycles, the expression level of p27 was measured in both nuclear and cytoplasmic fractions from GBM cells. Treatment of miR-138 to GBM cells significantly increased the protein expression level of p27 in nuclear fraction, while the increase of p27 was barely observed in cytoplasmic fraction (Fig. 4D). Immunofluorescence imaging on the GBM cells clearly www.nature.com/scientificreports/ showed the strong accumulation of p27 after miR-138 treatment compared to miR-Ctrl (Fig. 4E). These data suggest that miR-138 can arrest GBM cell cycle by p27 activation through inhibition of CD44/AKT pathway.

Ectopic expression of miR-138 improves survival rates in intracranial tumor bearing mice.
To evaluate the preclinical impact of miR-138 in GBM tumors, miR-138 was overexpressed in orthotopic GBM tumor xenograft. The intracranial GBM tumor was induced by injecting 2 × 10 5 human primary GBM cells into brain of NSG mice. Kaplan-Meier survival curves showed that overall mice survival rate (n = 8) was significantly prolonged by miR-138 overexpression with a median survival compared to miR-Ctrl treated mice group (n = 8) in all tested GBM cells: GBM12 (49 vs. 39 days, p < 0.05), GBM28 (35 vs. 28 days, p < 0.001) and GBM43 (28 vs. 24 days, p < 0.01) (Fig. 5A). Consistent with in vitro observations as described above, the decreased CD44 and the increased p27 expressions were cleared observed from the harvested GBM tumors in western blotting (Fig. 5B). Immunohistochemistry also showed the inverse correlation between CD44 and p27 expressions in the harvested GBM tumors relative to miR-138 overexpression along with the decreased Ki67+ cells, indicating the inhibition of cell proliferation by miR-138 (Fig. 5C). Taken together, our statistically significant results suggest the conceptual therapeutic advantage of miR-138 in mice bearing an intracranial tumor.

Discussion
Including GBM, miR-138 has been known to be downregulated from many cancers, such as melanoma, breast, neuroblastoma and pancreatic tumor 7,8 . Many studies showed that an overexpression of miR-138 can suppress cell proliferation, metastatic ability and drug resistance of cancer cells, suggesting a tumor suppressive role of miR-138. In normal brain tissues, miR-138 is highly expressed in dendrites to regulate dendritic spine morphogenesis www.nature.com/scientificreports/ by targeting a depalmitoylation enzyme acyl protein thioesterase 1 (APT1) 27 . Coincided with our data, low expression of miR-138 was previously observed in brain tumors and the low expression levels of miR-138 were associated with poor survival of GBM patients [10][11][12][13] . As shown in our data, a series of functional assays on the impact of miR-138 on GBM cells clearly demonstrated that the restoration of miR-138 abolishes GBM cell proliferation and migration and significantly increase survival rates of intracranial GBM tumor bearing mice. These data indicate a tumor suppressive role of miR-138 plays in primary GBM. Concurrent transcriptome analysis on GBM patient samples has identified CD44 for the first time as a potential target of miR-138, since its expression level was also inversely correlated with that of miR-138. Further functional study data confirmed that miR-138 directly downregulates the expression of CD44 protein by binding to the 3′ UTR of CD44. Transmembrane receptor CD44 is highly overexpressed and activated not only in GBM, but also many other cancers 18 . Activation of CD44 can support tumor progression by increasing many aspects of tumor supporting signaling pathways inside the tumor cells, such as cell proliferation, cell motility, tumor growth, angiogenesis and drug resistance, and even through communication with surrounding cells within tumor microenvironment (TME) 18 . Our finding strongly indicates the therapeutic potential of miR-138 as an attractive druggable target to modulate aberrantly deregulated pro-survival gene networks in GBM cells. Clinical trials with humanized CD44-targeting monoclonal antibody are being conducted in an attempt to neutralize the activated CD44 19 . Interestingly, inhibition of CD44 by overexpressing miR-138 in GMB cells significantly increased nuclear translocation of cell cycle inhibitor p27. As a result, GBM cell population arrested at G0/G1 phase was increased in a response to the miR-138 expression. Similar observation was previously reported in acute myeloid leukemia (AML), where CD44-triggered activation of PI3K/Akt signaling pathway downregulates p27 and cytoplasmic relocation 26 . In the future studies, single cell-RNA sequencing on mouse GBM tumor after miR-138 treatment will enable to dissect the molecular changes in various type of cells within the tumor microenvironment.
A recent study has reported another aspect of miR-138 as a tumor suppressive miRNA 10 . They showed that miR-138 can educate CD4+ T cells by downregulating two immune checkpoints, programmed cell death 1 (PD-1) and cytotoxic T-lymphocyte-associated molecule 4 (CTLA-4), resulting in 43% increase in median survival time in GBM tumor bearing mice 10 . This study implied that miR-138 can induce a conversion of cold tumor to immunologically hot tumor. Future co-delivery of miR-138 into GBM cells and surrounding immunosuppressed TME cells will be an intriguing approach to demonstrate a therapeutic potential of miR-138 for GBM treatment.
In contrast, Sampath's laboratory found that miR-138 was overexpressed in tumor-initiating glioma stem cell (GSC) 9 . They demonstrated that an ectopic expression of miR-138 promoted self-renewal potential of GSCs resulting their growth and survival. Their finding indicates that miR-138 may play a different role in a small subset of GBM stem cells, therefore further study will need to clarify whose GBM patients will benefit from the www.nature.com/scientificreports/ restoration of miR-138. These contradictory observations may have arisen from the heterogeneous nature of GBM 8,[15][16][17] . Our analysis was focused on primary GBM samples without history of recurrence or previous drug treatments. In the future study, it will be intriguing to study the correlation between miR-138 and GBM subtypes (IDH WT vs IDH mutants) by DNA sequencing. Nevertheless, our results strongly demonstrated the therapeutic potential of miR-138 for primary GBM tumor, which needs to be further assessed by clinical trials. To translate our findings useful for future clinical trials, we will need to develop therapeutic strategies to deliver miR-138 efficiently into GBM cells, or even surrounding tumor-supportive immune cells. For an example, recently developed RNA-driven small RNA delivery system may be a promising candidate to achieve the challenging task since it has successfully shown its preclinical potential to specifically target brain tumor cells across BBB via folate (FA)-mediated folate receptor (FR) recognition on tumor cells [28][29][30][31] .

Materials and methods
Cells and cell cultures. Patient-derived de-identified primary GBM cells was kindly provided by Dr. Jann  Table 1). RNA sequencing tags were only considered when they mapped to the same DNA strand as indicated by GRCh38. p11 annotation using HISAT2 32 . After assembly, 96.53% of the total clean reads were mapped to the reference transcriptome. The fragments per kilobase of transcript per million mapped reads (FPKM) value of the transcripts was calculated based on StringTie with default parameters 32 . In the preprocessing step of mRNA analysis, non-expressed gene, which defined as having a FPKM expression level of < 1 in > 80% of samples, were removed. The remaining transcripts were then used to perform the downstream analysis. Differentially expressed genes (DEGs) between GBM and control samples were detected by Ballgown based on the expression levels obtained from StringTie with threshold of adjusted p-value less than 0.01. In addition, the differentially expressed miR-NAs (DEmiRs) based on the matched mRNA-seq dataset were also detected by DEseq2 with adjusted p-value < 0.01 33 . We further required the DEGs or DEmiRs to have more than two-fold changes 34 . The functional enrichment analysis of the DEGs were conducted using the pathway databases Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) through DAVID tool (version 6.8, http:// david. ncifc rf. gov). To report reliable results, the pathways and GO terms with adjusted p-value < 0.05 were selected (adjusted by Benjamin-Hochberg Procedure, false discovery rate (FDR)). The mRNA-seq data were submitted to NCBI GEO database (GSE165286).
Ectopic expression of miRNAs. GBM  Flow cytometry. Detection of CD44 positive cells was determined by flow cytometry. GBM cells transfected with miRNAs as described above in six well plates were washed with PBS 4 days after the transfection and harvested with mild trypsiniation. The harvesed cells were stained with human CD44 antibody labeled with FITC, then subjected to flow cytometry by Beckman CytoFLEX Flow Cytometer (Beckman Coulter, USA). The obtained flow cytometry data was analyzed to determine the population of CD44 positive GBM cells against non-FITC cell population using FlowJo version 10 software (BD Biosciences, USA).
Cell proliferation assay. GBM cells plated in 96 well plates were transfected with miRNAs as described above. Four days later, each wells were treated with equal volume of CellTiter-Glo Luminescent Cell Viability Assay. After 10 min of incubation, the luminescence intensity was measured by Synergy H1 multi-mode microplate reader (BioTek Instruments, USA). When GBM12-RFP cell was used, the end-point cell viability was imaged by visualizing RFP-positive cells with excitation 532 nm and emission 588 nm using EVOS FL Auto2 Cell Imaging System (Thermo Fisher Scientific, USA).
Cell migration assay. GBM cells were transiently transfected with miRNAs as described above in a 8 µm pore size Boyden chamber (Transwell from Corning Costar, USA), which was then inserted into 24 well plates. Four days later, the migrated cells in the bottom chamber were fixed with 1% glutaraldehyde followed by staining with 0.5% crystal violet. Images of stained cells from each well were obtained by bright field microscope (EVOS FL Auto2, Thermo Fisher Scientific, USA).
Wound healing assay. GBM cells were plated in 12 well plates. Next day, the cell layer was scratched vertically using a sterile micropipette tip. Detached cells were washed off with PBS, and the wells from each group were imaged under a bright field microscope (EVOS FL Auto2, Thermo Fisher Scientific, USA) as Day 0. The adhered cells were then transiently transfected with miRNAs as decribed above. Four days later, each well was imaged.
Dual luciferase reporter assay. Firefly/Renilla Duo-Luciferase reporter vector containing the 3′ UTR clone of human CD44 was purchased from GeneCopoeia, USA. The 3′ UTR of human CD44 was predicted to include two binding sites of miR-138 at 364 and 753 bases, respectively. Two mutations (CD44-mut1 starting at base 364 and CD44-mut2 starting at base 753) were introduced into the miR-138 binding sites by replacing adenines and guanosines within the seed sequences to cytosines using site-directed mutagenesis (GenScript, USA). The plasimd DNAs amplified from E.coli were prepared by Maxi kit (Qiagen, USA). GBM cells were plated in 12 well plates 1 day before transfection of 1 μg CD44 3′ UTR DNA, CD44-mut1 or CD44-mut2 with Lipofectamine 3000 (Invitrogen, Thermo Fisher, USA). Next day, the cells were transfected with miRNAs as described above. After further 48 h, cells were lysed with Passive Lysis Buffer (Promega, USA) and assayed with Dual Luciferase Assay kit (Promega, USA) according to the manufacturer's instruction.
In vivo studies. Six to eight weeks old NSG mice (NOD-scid IL2Rgamma null ) purchased from the Jackson Laboratory, USA were housed and handled in accordance with the guideline of UTHealth Center for Laboratory Animal Medicine and Care (CLAMC) and the animal protocols approved by the UTHealth Animal Welfare Committee (AWC). All animal studies were carried out in compliance with the ARRIVE guidelines (http:// www. nc3rs. org. uk/ page. asp? id= 1357). Intracranial glioblastoma xenograft was generated in the NSG mice by implanting GBM cells as previously described 28,29 . Survival studies used eight mice per each groups. Tumor tissues were harvested from the mice brain, then either snap frozen in liquid nitrogen for western blotting analysis or fixed in 4% PBS-saturated formaldehyde. The fixed tissues were embedded in paraffin and sectioned at 5 µm thickness. Representative sections from each group were stained with Hematoxylin and Eosin (H&E) to be examined under bright field microscope (EVOS FL