Exosomes from Glioma-Associated Mesenchymal Stem Cells Increase the Tumorigenicity of Glioma Stem-like Cells via Transfer of miR-1587

Glioblastoma (GBM) is the most common and aggressive primary malignant adult brain tumor, and despite surgery, radiation, and chemotherapy, the median survival of patients with GBM is just over 1 year (1). This poor outcome is due in part to the cell-autonomous functions of therapeutically resistant tumor-initiating cells (TIC), also called glioma stem or stem-like cells (GSC). The surrounding stroma maintains a GSC-supportive niche that likely enhances GSC aggressiveness (2). Although classically, the stroma of GBMs was thought to be composed of reactive astrocytes, endothelial cells, and immune cells, each of which has been implicated in creating proglioma conditions (3–5), we and others have recently shown that GBMs harbor stromal cells resembling mesenchymal stem cells (MSC), which we called glioma-associated human MSCs (GA-hMSC; refs. 6, 7). GA-hMSCs are not merely bystanders in the tumor niche, but increase proliferation and self-renewal of GSCs in vitro, and enhance GSC tumorigenicity and mesenchymal features in in vivo intracranial models, confirming that GA-hMSCs are functionally important cells within the GSC niche (6). Secretion of IL6 by GA-hMSCs is one major intercellular communication mechanism that mediates the proliferation and stemness of GSCs (6). However, the mechanisms underlying the communication between the stroma and tumor cells are complex, and additional factors are likely involved.

Recently, evidence is building for a form of communication between neighboring cells that relies on nanosized (50–110 nm diameter), lipid bilayer vesicles, called exosomes, that are secreted by many cell types and taken up by neighboring or distant cells (8). This type of communication may be especially exploited in pathologic conditions such as cancer, where TICs and surrounding stromal cells may develop mutually supportive positive feedback loops of cellular communication (9, 10). Studies show that exosome-mediated transfer of oncogenic miRNA (oncomiR) from cancer cells can alter the biology of noncancer cells, whereas the transfer of tumor-suppressor miRNA can inhibit tumor growth (11–13).

Exosomal communication has not been extensively evaluated in glioma. Most recent reports focus on the impact of tumor cell–derived exosomes on stromal components. Glioma cells secrete exosomes that contain mRNA and miRNA and promote blood vessel formation (14). Extracellular vesicles (EV) from gliomas were shown to deliver miR-1 to recipient cells to modify glioma cell invasion and proliferation as well as impact stromal cells to promote endothelial tube formation (15). Another report demonstrated that GBM-derived exosomes could transfer miRNA to microglia (16).

Studies in other cancer types have provided some clues to the role of stromal-derived exosomes. Exosomes released from cancer-associated fibroblasts in breast cancer increased the invasion and motility of breast cancer cells (17). More recently, exosomes from gastric carcinoma–mesenchymal stem cells were shown to increase the growth and migration of human gastric carcinoma cells (18). Although no mechanism by which the exosomal contents mediate these tumor-enhancing effects was established, together these two studies demonstrate the tumor-promoting role of stroma-derived exosomes.

The role of tumor stroma-derived exosomes in the development and evolution of gliomas and the mechanisms by which exosomal communication impacts tumor cells are poorly studied, highlighting a major gap in knowledge. Thus, we sought to investigate exosomal communication within the glioma microenvironment, utilizing GA-hMSCs and GSCs isolated from patient tumors. We hypothesized that the transfer of specific cell-regulatory elements via GA-hMSC-derived exosomes could alter the biology of recipient GSCs, resulting in increased proliferation and clonogenicity.

GA-hMSCs were isolated from surgical specimens as described previously (6, 19). GA-hMSC lines met the criteria for MSCs as outlined by the International Society for Cellular Therapy (ISCT; ref. 19). GA-hMSCs have spindle shape morphology, are adherent in culture, and express the mesenchymal surface markers CD73, CD90, and CD105. All GA-hMSCs possessed the ability to tridifferentiate into adipocytes, chondrocytes, and osteocytes. GA-hMSCs were cultured in MSC medium: Eagle Minimum Essential Medium alpha (Sigma-Aldrich), 10% FBS, 1% penicillin–streptomycin, and 1% glutamine. Normal bone marrow-derived human MSCs (BM-hMSC) were purchased from Lonza, Inc. GSCs were isolated from human glioma surgical specimens as previously described and cultured in serum-free neural stem cell (NSC) medium: DMEM/Ham F-12 50/50 mix with l-glutamine (Corning CellGro), 2% B-27 supplement (Gibco), 20 ng/mL EGF, 20 ng/mL bFGF, and 1% penicillin–streptomycin (6, 20). GSCs and GA-hMSCs were validated with short tandem repeat (STR) fingerprinting in 2012. U251 cells were reauthenticated with STR fingerprinting in 2015. All cells tested mycoplasma-free prior to freezing and cells were thawed and used for experiments within five passages.

GA-hMSCs or GSCs were washed and incubated for 48 hours in serum-free and supplement-free NSC medium. GSC or GA-hMSC-derived exosomes were isolated by differential ultracentrifugation (21). Briefly, GA-hMSC–derived conditioned medium (CM) was filtered and centrifuged at 10,000 × g to remove large microvesicles (nonexosomal). The supernatant was ultracentrifuged at 100,000 × g. The raw exosome pellet was resuspended in PBS and ultracentrifuged at 100,000 × g, with or without a 30% sucrose cushion depending on the experimental usage. Exosome pellets were resuspended in PBS, NSC medium, or lysis buffer where indicated.

Protein from GA-hMSC–derived exosomes was extracted using membrane lysis buffer. Western blot analysis for the exosomal surface markers CD63 and GAPDH on MSC-derived exosomes, and the corresponding parental cell were performed using antibodies against CD63 (Santa Cruz Biotechnology, Inc.), GAPDH (Abcam #ab37168), CD16 (Abcam #ab94773), and CD32 (Abcam #ab45143).

BM-hMSC-derived and GA-hMSC–derived exosomes were prepared for electron microscopy according to Théry and colleagues (21). Exosomes were examined by electron microscopy on a JEM 1010 transmission electron microscope (JEOL) at an accelerating voltage of 80 kV. Digital images were obtained using the AMT Imaging System (Advanced Microscopy Techniques). Exosome grids were subjected to immuno-gold staining of the exosomal surface marker CD63 and the nonexosomal markers CD16 and CD32, utilizing gold-anti-rabbit (Sigma Aldrich #G7277), anti-CD63 (Santa Cruz Biotechnology, Inc., #15363), anti-CD16 (Abcam #ab94773), and anti-CD32 (Abcam #ab45143).

GA-hMSCs were transduced with a GFP-CD63 lentiviral construct (System Biosciences) and incubated for 48 hours in serum-free NSC medium. GFP-labeled GA-hMSC–derived exosomes were isolated and incubated with GSCs for 4 hours. Excess exosomes were removed by PBS wash, and internalization of GA-hMSC–derived exosomes by GSCs was analyzed by fluorescent confocal microscopy.

GSCs were dissociated and placed in a 96-well plate at 2.5 × 10³ cells/well. GA-hMSC–derived and GSC-derived exosomes quantified by CD63 ELISA were added to GSC cultures at doses of 5 × 10⁴ and 1 × 10⁵ exosomes/μL at time 0 and at 48 hours, and incubated for a total of 96 hours. GSCs were also treated with NSC medium and GA-hMSC–derived exosome-depleted conditioned medium (ED-CM). For miRNA experiments, GSCs were treated with either NSC medium, GA-hMSC-derived exosomes (1 × 10⁵ exosomes/μL), anti-miRNA (Qiagen), or GA-hMSC–derived exosomes (1 × 10⁵ exosomes/μL) plus anti-miRNA. Anti-miRNA (1 nmol/L) was delivered to GSCs by lipofection, and Lipofectamine was added equally across all treatment groups to control for lipofection effects. After 96 hours, GSCs were assessed for viability using a colorimetric assay [water-soluble tetrazolium (WST-1); Roche]. All experiments were performed in quadruplicate.

GA-hMSC–derived and GSC-derived exosomes were added to single-cell suspensions of GSCs (1 GSC/well verified prior to treatment) in 96-well plates at a dose of 1 × 10⁵ exosomes/μL at time 0, and at 1 and 2 weeks. GSCs were also treated with NSC medium and GA-hMSC–derived ED-CM. For miRNA experiments, GSCs were treated with NSC medium, GA-hMSC–derived exosomes, anti-miRNA (Qiagen), or GA-hMSC–derived exosomes plus anti-miRNA. Anti-miRNA was delivered to GSCs by lipofection at 1 nmol/L. Neurosphere formation was assessed after 3 weeks. All experiments were performed in quadruplicate.

Athymic nude mice (nu/nu) were purchased from the Department of Experimental Radiation Oncology, University of Texas MD Anderson Cancer Center (Houston, TX). All animal studies were performed under an Institutional Animal Care and Use Committee–approved protocol in accordance with federal, state, and institutional regulations.

GSCs (5 × 10⁶ cells) were pretreated with exosomes (1 × 10⁹ exosomes/mL), NSC medium, or GA-hMSC–derived ED-CM for 96 hours. Excess exosomes were eliminated by PBS wash, and GSC neurospheres were dissociated. Mice were anesthetized with ketamine (100 mg/kg)/xylazine (10 mg/kg). GSCs were injected into the right frontal lobe [n = 12/group, 5.0 × 10⁵ cells/mouse as previously described; 22, 23]. Mice (n = 9) were followed until moribund and then sacrificed. Another cohort (n = 3) were sacrificed 40 days postimplantation. Brains were removed, embedded in paraffin, and sectioned. Tumor volumes were calculated by adding multiple cross-sectional areas through the tumors after hematoxylin and eosin staining.

GA-hMSCs were expanded in MSC medium, washed, and cultured in NSC medium for 48 hours. GA-hMSC–derived CM was collected, ultracentrifuged, the ED-CM supernatant removed and saved, and the MSC-derived exosome pellets resuspended and lysed. Immunoblot was performed using human growth factor and cytokine antibody arrays (RayBiotech).

Isolated GA-hMSC–derived exosomes were exposed to RNase (1 nmol/L) to eliminate free-floating extra-vesicular RNA elements. Total RNA was extracted using the mirVana RNA Isolation Kit (Ambion). Total RNA was extracted from the parental cell line, and the miRNA profile for each sample was obtained using μParaflo microfluidic biochip technology, through LC Sciences.

To identify miRNA in the enriched subpopulation that were also the most highly expressed in GA-hMSC–derived exosomes, an expression-to-enrichment ratio (E:R ratio) was calculated, by multiplying the change in expression between the exosome and the parental cell, by the ratio of expression levels between the exosome and the parental cell.

GSCs were treated with GA-hMSC–derived exosomes (1 × 10⁹ exosomes/mL) or NSC medium for 48 hours. Excess exosomes were removed by PBS wash and total RNA was isolated. Gene expression profiling was performed using Illumina next-generation sequencing technology through LC Sciences.

miRNAs were overexpressed in GSCs by lentiviral (LV) transduction. Each LV-GFP-miRNA viral vector was transduced in GSCs (1 × 10⁶ cells) at a multiplicity of infection (MOI) of 3, and cultured under puromycin selection for 72 hours. Integration of the GFP-miRNA construct was assessed by fluorescent light microscopy.

The pEZX-MT06 luciferase reporter plasmid containing the NCOR1 3′UTR was purchased (Genecopoeia). A mutant 3′UTR in which the 3 miR-1587–binding sites identified by TargetScan were mutated was generated by synthetic gene synthesis (Thermo Fisher) and cloned into the same reporter plasmid. U251 glioma cells were transfected with the reporter plasmid and 10 nmol/L of either miR-1587 or control miRNA mimics (Dharmacon). Luciferase activity was measured 48 hours following transfection using the Dual-Glo Luciferase Kit (Promega).

Following antigen retrieval with citrate buffer (pH 6), immunostaining for NCOR1 was performed with the EnVision + System HRP (DAB) Kit (Dako) according to kit instructions, with anti-NCOR1 (Abcam #ab58396). Four fields from each tumor-bearing mouse (GSC-262 alone, n = 2; GSC-262 + GA-hMSC exosomes, n = 3) were scored for positive NCOR1 cells and the average percent of NCOR1-positive cells per mouse reported.

GSC-20 and GSC-262 cells were infected with lentiviral shRNA against NCOR1 and cultured under puromycin selection for 72 hours. To generate GSCs overexpressing NCOR1, an expression vector containing the full-length cDNA for NCOR1 was transfected into GSC-262 using Lipofectamine (Thermo Fisher) followed by G418 selection. RNA was isolated from GSCs using the RNeasy Kit (Qiagen), and NCOR1 expression was determined by quantitative RT-PCR (ABI, NCOR1 hs01094541_m1 and GAPDH hs02758991_g1). shRNA/ORF clones were purchased from GE Dharmacon through the MD Anderson shRNA and ORFeome Core and the clone numbers are: shRNA1, V3LHS_395515; shRNA2, V2LHS_91777; shRNA3, V2LHS_91779, shRNA4, V3LHS_395517; NCOR1 ORF, 100069131.

Survival analysis was performed using the log-rank (Mantel–Cox) test. All other comparisons were performed using t tests and all statistical tests are noted in the figure legends. Statistical analysis was performed using GraphPad Prisim v6.

We recently reported the isolation and characterization of GA-hMSCs from a large set of human glioma surgical specimens (6). To determine the role of exosomes in the interaction of stromal GA-hMSCs with tumor cells, specifically patient-derived GSCs, we selected four GA-hMSC lines, all of which met the ISCT criteria for MSCs (19). In addition, we chose four GSC lines isolated from surgical specimens. Two of these, GSC-262 and GSC-20, were isolated from the same tumor specimens as their respective GA-hMSCs (GA-hMSC-262 and GA-hMSC-20). The GA-hMSCs were not tumorigenic when implanted into the brains of athymic mice, consistent with their role as stromal cells. In contrast, all four GSC lines formed aggressive tumors when as few as 1,000 cells were implanted into mouse brains, consistent with their role as TICs. Three GSC lines and 3 GA-hMSC lines were previously characterized (6) and we characterized one additional line for both GSCs and GA-hMSCs (Supplementary Table S1). Low passage whole-genome sequencing confirmed that GA-hMSC-230 (6) and GA-hMSC-247 (Fig. 1A) harbor few genetic alterations, similar to BM-hMSCs, and are likely normal stromal cells recruited into the tumor (type 1 GA-hMSCs). We previously reported that GA-hMSC-262 and GA-hMSC-20 do have some block copy number variations, but these differ from the alterations found in the paired GSCs, and may represent examples of recruited normal MSC-like cells that acquired unique genomic alterations (type 3 GA-hMSCs). The cell lines described here provide two matched and two unmatched GA-hMCS/GSC pairs and represent two subtypes of GA-hMSCs.

To determine whether GA-hMSCs secrete exosomes, BM-hMSCs and each of the GA-hMSCs were cultured in serum-free, exosome-free medium, and exosomes were isolated from the supernatant. Western blot analysis of protein extracted from isolated vesicles from all four GA-hMSCs and from BM-hMSCs revealed the presence of the widely accepted exosome markers CD63 and GAPDH (Fig. 1B; ref. 24). CD63 was enriched in the exosome fraction compared with the whole-cell lysates of the parental cells. The nonexosomal markers CD16 and CD32 were absent from the vesicles isolated from all GA-hMSCs and BM-hMSCs, although they were present in the parental cells.

An electron microscope was used to further characterize the vesicles isolated from BM-hMSCs and GA-hMSCs. We identified abundant bilayer vesicles, 30–100 nm in diameter, consistent with the known appearance of exosomes (Fig. 1C). This size distribution was confirmed with NanoSight technology (Fig. 1D). Immunostaining and an electron microscope revealed positive CD63 gold labeling on the membrane of the vesicles from all GA-hMSCs and BM-hMSC (Fig. 1C). In contrast, CD16 and CD32 gold labeling was not observed, further supporting the classification of the isolated vesicles as exosomes.

To confirm the uptake of GA-hMSC–derived exosomes by GSCs, we labeled exosomes using a CD63-GFP. GSCs were cocultured with CD63-GFP–labeled GA-hMSC–derived exosomes, and fluorescent confocal microscopy revealed the presence of GFP-labeled GA-hMSC–derived exosomes exclusively in the cytoplasm of GSCs (Supplementary Fig. S1). These results indicate that GSCs can internalize GA-hMSC–derived exosomes.

To determine the growth effects of GA-hMSC–derived exosomes on GSCs, we assayed GSC proliferation and clonogenicity after exposure to GA-hMSC–derived exosomes. GSCs were treated with purified GA-hMSC–derived exosomes, GA-hMSC–derived ED-CM as a positive control, or purified GSC-derived exosomes (“self exosomes”). The purified GA-hMSC–derived exosomes significantly increased the proliferation of GSCs in a dose-dependent manner (Fig. 2A and B). Consistent with our previous observations (6), GA-hMSC–derived ED-CM also significantly increased GSC proliferation compared with untreated cells. At higher doses, the effect of purified exosomes significantly increased proliferation of all GSCs compared with ED-CM (Fig. 2A and B). Similarly, GA-hMSC–derived exosomes significantly increased GSC clonogenicity (Fig. 2C and D). Increased GSC proliferation and clonogenicity were observed upon treatment with exosomes derived from both matched and unmatched GA-hMSCs (Fig. 2; Supplementary Fig. S2A–S2H). In contrast, treatment of GSCs with GSC-derived self-exosomes did not have a significant effect on GSC proliferation or clonogenicity, confirming previous reports that exosomes do not exert any known autocrine effects (25).

To assess the effects of GA-hMSC–derived exosomes on GSCs in vivo, GSCs were pretreated with GA-hMSC–derived exosomes, and subsequently implanted into the right frontal lobe of nude mice. Pretreatment of GSC-262 with GA-hMSC-262–derived exosomes significantly decreased (P < 0.05) median survival from 56 to 45 days (Fig. 3A). Furthermore, pretreatment of GSC-20 with GA-hMSC-20–derived exosomes significantly decreased (P < 0.05) median survival from 49 to 37 days (Fig. 3B). In agreement with in vitro results, pretreatment of GSC xenografts with GA-hMSC–derived ED-CM also significantly decreased (P < 0.05) median survival, whereas pretreatment of GSCs with GSC-derived self-exosomes did not have a significant effect on median survival.

Histologic analysis of brain specimens from mice implanted with GSCs pretreated with GA-hMSC–derived exosomes demonstrated a significant increase in tumor volume at 40 days postimplantation (Fig. 3C and D). As expected, pretreatment of GSCs with GA-hMSC–derived ED-CM also significantly increased tumor volume. In contrast, and in agreement with in vitro results, treatment of GSCs with GSC-derived self-exosomes did not have a significant effect on tumor volume. These histologic results corroborate with survival data and findings from in vitro experiments. Together, these results indicate that GA-hMSC–derived exosomes significantly increase the tumorigenicity of GSCs.

We previously demonstrated that GA-hMSCs impact GSC proliferation via secretion of IL6 (6). To determine whether GA-hMSC–derived exosomes also contain growth-promoting proteins or cytokines, we utilized protein array technology to examine 41 growth factors and receptors and 59 cytokines. Consistent with our previous study, GA-hMSC conditioned media and ED-CM contained several growth promoting factors (e.g., IL6). However, cytokines and growth factors were below detection in the purified exosome fraction. (Supplementary Fig. S3A and S3B). These data indicate that, whereas GA-hMSC–derived CM and ED-CM contain growth factors secreted by the cell, these growth factors, receptors, and cytokines are not detected within purified GA-hMSC–derived exosomes.

We next utilized miRNA microarray technology to analyze 2,019 miRNAs (miRBase 19.0) for their level of expression in MSCs and MSC-derived exosomes. miRNA profiles for GA-hMSC–derived exosomes and BM-hMSC–derived exosomes contained numerous miRNAs, which were significantly different (P < 0.001, paired t test) from that of the parental cell line (Fig. 4A). Cluster analysis of the top 20 differentially expressed miRNAs indicated that exosomal miRNA cluster separately from cellular miRNA (Fig. 4B). GA-hMSC exosomes clustered together with the exception of GA-hMSC 247, which was more closely related to exosomes from BM-hMSCs. We identified 37 exosomal miRNAs with a ≥5,000 hybridization intensity (top 0.2% of the most highly expressed miRNA), among exosomes derived from all 4 GA-hMSCs, which we refer to as highly expressed miRNA (Supplementary Table S2).

We then identified a subpopulation of miRNA that were highly enriched in MSC-derived exosomes when compared with the parental cell. Exosomal miRNA that had significantly different (P < 0.05, paired t test) average levels of expression between the GA-hMSC–derived exosomes and the parental GA-hMSCs are referred to as enriched miRNA. miRNA that had expression changes greater than 2 SDs from the mean (top 2.5%) were termed highly enriched miRNA (Supplementary Table S2). We calculated the expression-to-enrichment ratio (E:R ratio) for each miRNA, and identified a group of 8 miRNAs that were the most highly expressed and highly enriched (>3 SD from the mean) in MSC-derived exosomes when compared with the parental cells (Fig. 4C; Supplementary Table S2). This cutoff correlated with average miRNA expression levels >5,000 hybridization intensity. The expression of each of these 8 highly expressed and highly enriched miRNAs was significantly increased in GSCs after treatment with GA-hMSC–derived exosomes (Supplementary Fig. S4A). Conversely, we also identified a group of 8 miRNAs that were the most highly expressed and highly enriched (>1 SD from the mean) in the parental cells when compared with MSC-derived exosomes (Fig. 4C; Supplementary Table S2). This group of miRNAs is thus relatively depleted in MSC-derived exosomes.

Using the miRTarget 2.0 database, we identified 251 unique predicted gene targets (>90 target score) for 7 of the 8 highly expressed and highly enriched miRNA (no predicted gene targets for miR-4508) in MSC-derived exosomes (Supplementary Table S3; ref. 26). Conversely, we also identified 245 unique predicted gene targets for all 8 of the miRNAs that are depleted from MSC-derived exosomes (Supplementary Table S3). Expression levels for the predicted gene targets in the two groups were obtained from gene expression profiling performed on GSC-262 and GSC-20. The expression levels of the predicted gene targets in untreated GSCs were compared with that of GSCs treated with GA-hMSC–derived exosomes for 48 hours. The average fold change in expression of the 251 predicted gene targets for the 7 enriched miRNAs, was significantly greater (P < 0.01) than the average fold change in expression of the 245 predicted gene targets for the 8 depleted miRNAs (Fig. 4D). These results indicate that highly expressed and highly enriched miRNA in GA-hMSC–derived exosomes are capable of downregulating their predicted gene targets in GSCs.

To determine the influence of the enriched miRNA in GA-hMSC–derived exosomes on GSC proliferation, we utilized lentiviral transduction to overexpress each of the 7 enriched exosomal miRNAs in GSC-262 and GSC-20. Of the 7 miRNAs evaluated, only the overexpression of miR-1587 and miR-3620-5p produced significant increases in the proliferation of GSCs, which were similar to that of GSCs after treatment with GA-hMSC–derived exosomes (Fig. 5A). miR-1587 also promoted a significant increase in the clonogenicity of GSCs, similar to that of GSCs after treatment with GA-hMSC–derived exosomes (Fig. 5B).

To evaluate the role of miR-1587 in exosome-mediated GSC proliferation and self-renewal, we treated GSCs with GA-hMSC–derived exosomes in the presence of anti-miR-1587. The effects of GA-hMSC–derived exosomes on GSC proliferation (Fig. 5C) and self-renewal (Fig. 5D) were significantly impaired with the addition of anti-miR-1587. This indicates that miR-1587 contained within GA-hMSC–derived exosomes has a causal role in increasing the proliferation and self-renewal of GSCs.

We assessed the functionality of miR-1587 by directly analyzing one of its predicted gene targets, the nuclear hormone receptor corepressor-1 (NCOR1). Overexpression of miR-1587 decreased GSC expression of NCOR1 when compared with untreated GSC controls, similar to the effects of purified exosomes on GSCs (Fig. 5E). Importantly, treating GSCs with anti-miR-1587 partially reversed the exosome-mediated downregulation of NCOR1. NCOR1 protein levels were also decreased in tumor sections from mice bearing GSC-262 tumors after treatment with GA-hMSC-262 exosomes (Supplementary Fig. S4B and S4C). To demonstrate a direct impact of NCOR1 levels on GSC proliferation, we used shRNA to knockdown NCOR1. NCOR1 inhibition was verified by quantitative RT-PCR (Supplementary Fig. S5A and S5B). Specific NCOR1 knockdown led to increased GSC proliferation (Fig. 5F; Supplementary Fig. S5C). In contrast, overexpression of NCOR1 in GSC-262 led to a decrease in proliferation (Supplementary Fig. S5C). A luciferase reporter assay demonstrates direct regulation of the NCOR1 3′UTR by miR-1587 (Supplementary Fig. S5D). These results indicate that regulation of NCOR1 in GSCs by miR-1587 in GA-hMSC–derived exosomes plays a role in promoting the proliferation and clonogenicity of GSCs.

We took a unique approach to explore stroma–tumor communication by examining the interactions between matched and unmatched pairs of GA-hMSCs and GSCs. Here we show that GA-hMSCs release exosomes that are internalized by GSCs, resulting in a significant increase in GSC proliferation and clonogenicity in vitro, as well as a significant increase in GSC tumorigenicity in vivo. Moreover, we show that GA-hMSC–derived exosomes contain highly expressed and highly enriched miRNA, one of which, miR-1587, was able to significantly increase both GSC proliferation and clonogenicity in in vitro experiments. The delivery of miR-1587 by GA-hMSC–derived exosomes resulted in the downregulation of the tumor suppressor NCOR1 in the recipient GSCs (Fig. 6).

Although there are several studies exploring the intercellular communication between stromal cells and cancer stem cells (CSC), few studies have investigated the role of stromal cell–derived exosomes in interactions with CSCs. Previous studies have indicated that stromal cell–derived exosomes can enhance the growth of CSCs in other cancer types, thereby supporting the results of our study (17, 18). More recently, astrocyte-derived exosomes were shown to support brain metastasis via miRNA delivery (27). We demonstrate that GA-hMSC–derived exosomes not only enhance the growth of GSCs, but also increase their clonogenicity, ultimately resulting in increased tumor burden and decreased survival in vivo. To our knowledge, this is the first study to examine the function of exosomes from tumor-derived MSCs on CSC growth and self-renewal in gliomas.

A recent report demonstrated that GSC-derived EVs support intratumoral heterogeneity and that EVs derived from GSCs of the mesenchymal subtype had an increased proliferative effect on proneural subtype GSCs, whereas treatment with self-EVs had no effect (28). Consistent with this report, we observed minimal effects on GSC growth when treated with self-derived exosomes. However, we demonstrate increased GSC proliferation and clonogenicity after exposure to exosomes derived from both matched and unmatched GA-hMSCs. Also in agreement with this report, the greatest increases in GSC proliferation were observed when GSCs were treated with exosomes from GA-hMSC-20, which was derived from a mesenchymal tumor. Notably, treatment with exosomes from GA-hMSC-20 (mesenchymal) produced the most significant increases in proliferation of GSC-262, a proneural GSC. We also observed that GSCs derived from classical GBM (GSC-7-2 and GSC-11) had a lower response to the GA-hMSC exosomes of all subtypes (Supplementary Fig. S2C, S2D, S2G, and S2H). Clustering analysis indicated that GA-hMSC-247 and the corresponding exosomes were more closely related to BM-hMSCs than the other GA-hMSCs in terms of miRNA profiles (Fig. 4B). Further work with additional GSCs and GA-hMSCs from each GBM subtype is needed to confirm the impact of GBM subtype on the effects of GA-hMSC–derived exosomes on tumor growth, but our current studies suggest that exosomal communication is impacted by the complex heterogeneity of GBM.

Our previous study demonstrated that GA-hMSCs support GSC growth through secretion of IL6 (6). This effect is verified in the current study by the effects of ED-CM on GSC proliferation and clonogenicity (Fig. 2). GA-hMSC–derived exosomes lacked IL6 and other cytokines and growth factors (Supplementary Fig. S3), prompting us to explore the miRNA content of exosomes. miRNAs have the potential to produce longer lasting effects in recipient cells due their slow decay rate and their ability to regulate multiple genes. Consistent with other cell types, including BM-hMSCs (29–31), we showed that the GA-hMSC–derived exosomes contained unique miRNA profiles as compared with the parental cell, including a specific group of miRNA that were highly expressed.

Unlike the group of depleted miRNA from GA-hMSC–derived exosomes that are tumor suppressive, the group of highly expressed and highly enriched miRNA in GA-hMSC–derived exosomes consists of newly identified miRNA whose targets and functions have not been clarified. The potential for these unstudied miRNA to be oncomiRs is suggested by our experimental results in which GA-hMSC–derived exosomes increased the tumorigenicity of GSCs. Given the wide range of genes that miRNAs target, and the wide range of miRNAs that can target the same gene, the effects of GA-hMSC–derived exosomal miRNA on GSCs are likely not attributable to a single miRNA. However, our results indicate that miR-1587 is at least one exosomal miRNA that appears to mediate the increase in proliferation and clonogenicity of GSCs. Although we focused our report on the miRNA that were both highly expressed and highly enriched in exosomes, it is possible that miRNA highly expressed in GA-hMSCs that were not overly enriched in exosomes may also be important for exosomal effects on tumor growth. Consistent with this concept, we observed high expression (without enrichment) of miR-21, a well-known oncomiR (Supplementary Table S2). Our observations that the highly expressed/exosomal depleted miRNA contained several tumor-suppressive miRNAs and that the highly expressed/highly enriched miRNA have growth-promoting roles suggests that there may be a specificity to exosomal miRNA packaging as has been reported for other cell types (32).

The capability of exosomal miRNA to regulate gene expression in recipient cells has been established in various tissues, and in the brain, this process has been described in both physiologic and pathologic conditions. The delivery of miR-124a via neuron-derived exosomes increased the expression of glutamate transporter-1 in recipient astrocytes. Furthermore, the disruption of this communication pathway is linked to key changes that occur during the progression of amyotrophic lateral sclerosis (33). miRNA in exosomes derived from the U251 GBM cell line were able to alter the transcriptome of recipient human brain microvascular endothelial cells (34). The communication of exosomal miRNA significantly downregulated the expression of 19 genes involved in the maintenance of the blood–brain barrier (34). Our study also demonstrated that exosomal miRNA cargo can result in downregulation of target genes in GSCs (Figs. 4 and 5), further confirming that cells can utilize exosome-mediated regulation of gene expression to modify their surrounding microenvironment. Our in vitro studies indicate that the GA-hMSC exosomes' effects on GSC proliferation and clonogenicity were at least in part via exosomal miRNA targeting gene expression in GSCs. Furthermore, levels of the miR-1587 target NCOR1 were decreased in the tumors of mice when GSCs were pretreated with exosomes from GA-hMSCs, suggesting that similar exosome-mediated gene regulation mechanisms occur in vivo.

This study is also the first to describe the targeting and downregulation of NCOR1 in gliomas by miR-1587. Since the identification of NCOR1 by Horlein and colleagues in 1995, several other NCORs have been identified, and have been shown to function in both normal physiologic and pathologic conditions, including cancer (35–39). NCOR was initially established as an oncogene (40). However, this study evaluated the NCOR family and did not distinguish between effects caused by NCOR1 and NCOR2. In tumor cells from patients with astrocytic gliomas, low expression of NCOR1 in both cytoplasmic and nuclear fractions, and high nuclear expression of NCOR2, were observed. This study suggested that NCOR2 may have an oncogenic function, whereas NCOR1 has a tumor suppressor function (41). Consistent with this concept, knockdown of NCOR1 resulted in glioma growth and invasiveness both in vitro and in vivo, demonstrating a tumor-suppressive function (42). Furthermore, decreased expression of NCOR1 correlated with epithelial-to-mesenchymal transition in GBM (42). Together, these studies indicate that downregulation of the tumor suppressor NCOR1 in gliomas promotes tumor growth.

This study describes exosome-mediated miRNA delivery as an additional stroma–tumor communication mechanism in glioma (Fig. 6). GA-hMSCs therefore provide a tumor supportive environment both by the secretion of soluble, growth-promoting factors, such as IL6, and via exosomal delivery of specific oncomiRs. We identified one specific miRNA and characterized one potential target; however, further study will be required to clarify the specificity of miRNAs loading into exosomes and to fully characterize these novel miRNAs and their targets. As both MSCs and exosomes are currently being explored as delivery modes for therapeutics, this and future work will be important considerations in the design and implementation of such delivery methods.

R. Kalluri reports receiving a commercial research grant from Codiak Biosciences, has ownership interest (including patents) in Codiak Biosciences, and is a consultant/advisory board member for Codiak Biosciences. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J. Figueroa, A. Hossain, A.J. Bean, E.T. Walters, F.F. Lang

Development of methodology: J. Figueroa, T. Shahar, A. Hossain, J. Gumin, F.F. Lang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Figueroa, L.M. Phillips, T. Shahar, A. Hossain

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Figueroa, L.M. Phillips, A. Hossain, H. Kim, R.G. Verhaak, F.F. Lang

Writing, review, and/or revision of the manuscript: J. Figueroa, L.M. Phillips, G.A. Calin, R. Kalluri, F.F. Lang

Study supervision: J. Fueyo, F.F. Lang

We thank David M. Wildrick, PhD, for editing the manuscript.

This work was supported by grants from the NIH (5R01 CA115729-05), National Cancer Institute, SPORE in Brain Cancer (Project 1, Core B, and Core C; 1P50 CA127001-06), The Broach Foundation for Brain Cancer Research, The Elias Family Fund, The Gene Pennebaker Brain Cancer Fund, The Ben and Catherine Ivy Foundation, The Anthony Bullock III Foundation; The Brian McCulloch Fund, The Uncle Kory Foundation, The Jason & Priscilla Hiley Fund (to F.L. Lang); the NIH/NCI grants 1UH2TR00943-01 and 1 R01 CA182905-01; and the UT MD Anderson Cancer Center Brain SPORE (2P50CA127001 to G.A. Calin).

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.