Extracellular Vesicles from High-Grade Glioma Exchange Diverse Pro-oncogenic Signals That Maintain Intratumoral Heterogeneity

Characterization of the genome (1, 2) and RNA transcriptome (3–6) provided an increasingly high-resolution picture of the heterogeneity of the HGG landscape. Single-cell RNA sequencing of HGG reveals that tumor cell subpopulations with different transcriptome subtypes coexist and have developmentally evolved over a period of time from a distantly related precursor, likely a cancer stem-like cell (7, 8). This intratumoral heterogeneity with redundant signaling network aberrancies underlies the ineffectiveness of conventional and targeted therapies (9–11).

Recently, there is increasing appreciation for the role that extracellular vesicles (EV) play in cell–cell communication in cancers. However, it is not known whether the profound genetic and phenotypic diversity of cells in a HGG is related to the intercellular transfer of functional molecules contained within EVs (12–14). Here, we show that EVs released by GSCs retain subtype characteristics and that their transfer between GSC subtypes leads to protumorigenic changes in the target GSC. Experimentally, in mouse brains, there is a hierarchical clustering of different GSC subtypes with visible transfer of EVs. Coupled with the finding that the EV proteome is predictive of HGG patient outcome, this study shows that EV communication may be a critical linchpin in HGG GSC subtype diversity.

Tumor samples were obtained as approved by Institutional Review Board at and The Harvard Medical School (HMS, Boston, MA). Surgery was performed by E.A. Chiocca and I. Nakano. GSCs were obtained by dissociation of tumor samples and cultivated in stem cell–enriching conditions using glutamine, B27, FGF, and EGF-supplemented Neurobasal medium (14, 15). The unique identity of cultured patient-derived cells was confirmed by short tandem repeat analysis (16).

The conditioned media were collected and EVs were isolated by differential centrifugation and analyzed using a NanoSight as described previously (14).

All mass spectra were acquired at the Bioproximity LLC and data were validated by Western blotting (14).

For EV transfer, 3 × 10⁵ GSCs/mL were maintained overnight in unsupplemented medium followed by 24-hour treatment with EV derived from mesenchymal (M) GSCs (EVM) or form proneural (P) GSCs (EVP; 5 μg protein/mL).

For spheroid formation, GSCs were dissociated to single cells and plated at 200 (M GSCs) or 1,000 (P GSCs) cells/well and in 96-well plate in unsupplemented medium and treated with EVs (5 μg of protein/mL) at 0 and 48 hours and analyzed after 96 hours.

For co-culture assay, single-cell suspensions of M GSCs, P GSCs, and M/P GSC coculture (at 1:4 ratio) were cultured for 48 hours in neurosphere medium.

Female athymic mice were purchased from Envigo. Mice were housed in HMS animal facility in accordance with NIH regulations. Protocols were approved by the HMS Institutional Animal Care and Use Committee. Intracranial tumor injection was performed as described previously (14). For GSC coimplantation experiments, either 1 × 10³ M GSCs or 5 × 10⁵ P GSCs, or both combined were used. For GSC/EV coimplantation, cells were pretreated with EVs for 24 hours, pelleted, and admixed with EVs (at 5 μg of protein/mL).

EV release was visualized by embedding either PALM-Tomato M GSC or PALM-GFP P GSC spheroids in collagen. All fluorescent and bright field microscopy assays were observed using a Nikon Eclipse Ti.

Experimental and clinical data were analyzed for molecular profiles of glioblastoma using the GBM-BioDP (17). Data are expressed as mean ± SD. The unpaired two-tailed t test was used to compare between two groups. One-way ANOVA, followed by Bonferroni test, was conducted to test for significance among multiple groups. P < 0.05 was considered significant.

Recent data have shown that gene expression datasets may be used to classify GSCs from different subtypes (15). To investigate whether our collection of GSCs (Fig. 1A) display distinct expression profiles, an analysis of gene signature was performed (15) followed by principal component analysis (PCA). This defined two distinct subpopulations of GSCs, belonging to the P or the M subtype. As expected, there were also several GSCs that did not cluster into either subtype (Supplementary Fig. S1A). To validate whether the observed cellular transcriptional subtype diversity was reflected by the protein composition of EVs released by these GSCs, EVs were isolated and quantified followed by global mass spectrometry (MS)-based analysis of their proteome. Interestingly, there was significant physical heterogeneity of EVs released by both subtypes: larger and homogenously sized EVM, and smaller but with broader diversity in size EVP. There was no significant difference in the number of EVs shed by either subtype in culture (Supplementary Fig. S1B). Comprehensive MS analysis revealed over 1,400 proteins within the EV proteome. Over 90% of them were detected in both subtypes (Supplementary Fig. S1C). For subtype signature profiling, a differential expression analysis of EVP vs. EVM was carried out and 98 proteins were found to be significantly differentially released in a subtype-dependent manner (Fig. 1B, Supplementary Table S1), indicating that the proteome of GSC EVs recapitulated subtype clustering. In fact, when the GSC EV proteome data was queried with known molecular subtypes from TCGA (4), the EV signature clustered into the two same subgroups (P and M) as those from their respective GSCs (Fig. 1C; Supplementary Table S2). In silico analysis revealed that proteins identified in EVs from each subtype regulate different biologic processes and molecular functions. EVP proteins primarily control processes related to neurons/nervous system development, and EVM proteins function as regulators of receptor binding, protein synthesis and gene expression (Supplementary Fig. S1C). To provide validation of the proteome analysis by Western blots, we selected: (i) specific EV proteins that were most predictive of a subtype by TCGA classification (for example NCAM1, PLAU for P vs. M GSCs, see Supplementary Fig. S1D), (ii) proteins that were most differentially expressed between the two subtypes (for example, IGFPB2 or PNF1 for P vs. M GSCs, see Supplementary Fig. S1E), and (iii) proteins that are most commonly present in HGG EVs (ANXA2, FASN; Supplementary Fig. S1E; ref. 14). The analysis showed a cellular and EV subtype–specific pattern of distribution (Fig. 1D), supporting the global analysis (Fig. 1B).

To evaluate whether there were functional consequences after EVM and EVP exchange, GSCs of each subtype were exposed to EVs isolated from the other subtype, using membrane-tagged GSCs for visualization of EV internalization (18) (Fig. 2A). When P GSCs were exposed to EVM, there was a significant increase in GSC sphere frequency, number, viability, and volume (Fig. 2B and Supplementary Fig. S2A–D). Exposure of P GSCs to their own EVs had no effect. When M GSCs were exposed to EVP, there was no significant change in M GSC frequency (Fig. 2B) and viability (Supplementary Fig. S2C). However, the frequency of larger spheroids was moderately increased upon exposure to either EV (Supplementary Fig. S2B). Next, to model coexistence of P and M GSCs within tumors (6), single-cell GSCs from either M or P subtype were cocultured together to form spheroids (Fig. 2C, left). All of the spheroids ended up growing as mosaics instead of growing as homogenous cultures of M or P GSCs only (Fig. 2C, middle). This exclusivity of mosaic growth indicated a selective pressure for GSCs to grow in a heterogenous rather than a homogenous subtype environment (Supplementary movie). Interestingly, these mosaic spheroids formed a visibly distinct architecture with the M GSCs clustered together in the inner core of the spheroid, while P GSCs localized to the outer layers. Importantly, frequent exchange of EVs between cells from both subtypes was observed, suggesting dynamic and ongoing EV-mediated communication between cell types (Fig. 2C, right). To verify whether growth of each subtype benefited from the observed coexistence, growth of GSCs was assessed in coculture versus monoculture conditions. Similar to the EV treatment experiments, growth of M GSCs was not affected in coculture versus monoculture, but there was a significant increase in P GSCs growth when cocultured with M GSCs (Fig. 2D). To determine which signaling pathways may be involved in GSC subtype cross-talk, the expression of key pro-oncogenic kinases was analyzed: AKT was more phosphorylated in P GSCs, while ERK, SRC, and P65 were more phosphorylated in M GSCs (Supplementary Fig. S2E). Although these kinases were more phosphorylated in EVM-treated P GSCs, suggesting a possible mechanism for the observed increase in P GSC growth, the involvement of other signaling pathways cannot be excluded. The analysis of the GSC/EV proteome showed that EGFR was M GSC–enriched with little immunopositivity in P GSCs (Fig. 1D), which was further validated by in situ staining and qPCR (Supplementary Fig. S2F). When P GSCs were exposed to EVM, EGFR was detected in P GSCs, demonstrating that EGFR could be transferred from M GSCs to P GSCs by EVs. Such transfer, coupled with the previously observed kinase phosphorylation, could explain the significant growth enhancement observed in P GSCs exposed to EVM.

To determine whether the in vivo microenvironment influenced the EV-mediated cross-talk between GSCs observed in vitro, we performed set of experiments using an intracranial GSC xenograft model. First, P GSCs were coinjected with either EVP or EVM. Second, P GSCs were coinjected with M GSCs (Fig. 3A). For the first experiment, animals were sacrificed four days after GSC implantation: an approximately 3-fold higher percentage of Ki-67–positive P GSCs upon coinjection with EVM was observed when compared with P GSCs alone. However, this increase was not observed upon coinjection with EVP (Fig. 3B). Because M GSCs grow more aggressively than do P GSCs (15), P GSCs and M GSCs were admixed in a 500:1 ratio in the second coimplantation experiment. Nine days after implantation, tumors formed by mixed GSCs were significantly larger than those comprised of M GSCs only (Fig. 3C, top), perhaps unsurprisingly as the latter were initially formed by much fewer cells. However, volume of tumors originated by M GSCs implanted with P GSCs was approximately 4-fold greater than that of M GSCs alone (Supplementary Fig. S3A), suggesting that the heterogeneous environment provided a mitotic advantage to M GSCs. This differed somewhat from the in vitro data where EVPs did not provide a growth advantage to M GSCs, possibly suggesting additional influences to growth in the in vivo model. Analysis of tumor architecture revealed that M GSCs tended to form the noninvasive tumor core, while P GSCs located preferentially to the peripheral zone with single cells invading into surrounding tissue, similarly to in vitro observation of spheroids' architecture. Both cell types were found with numerous internalized EVs from the other cell type (Fig. 3C, bottom and Supplementary Fig. S3B), suggesting that EV-mediated communication is in fact frequent between these cell types in vivo. Finally, a cohort of animals from the coimplantation experiment was studied for survival, showing a significant survival disadvantage for tumors composed of two cell subtypes (Fig. 3D), when compared with those formed by M GSCs or by P GSCs alone. Analysis of tumor composition revealed that despite the initial dominance of P GSCs, by day 9 most of the tumor was composed of M GSCs and by the terminal phase almost the entire tumor was composed of M GSCs (Supplementary Fig. S3C–D). Likely, M GSCs benefited from the P GSCs coimplantation at the beginning and overgrew to take over the entire neoplasm, suggesting that heterogeneity coexisting within one tumor ecosystem drives tumor progression.

Finally, we examined whether the subtype-specific heterogeneity of the GSC EV proteome was relevant to HGG patients' outcome according to TCGA (4, 19). Genes encoding for all GSC EV proteins were associated with poorer outcome in a full cohort of HGG patients (Fig. 4A and Supplementary Fig. S4A). However, when we examined the EV protein profiles separately, the EVM signature was associated with a significantly worse outcome in P but not M tumors (Fig. 4B and Supplementary Fig. S4B). Conversely, the EVP signature was associated with a significantly worse outcome in M but not P tumor subclasses (Fig. 4C and Supplementary Fig. S4C). These datasets indicated that heterogeneity within tumors, which may be propagated throughout the tumor via EV communication, was associated with decreased survival. This is in agreement with the scenario proposed by Patel and colleagues (6) where the clinical outcome of a glioblastoma subclass is influenced by the proportion of tumor cells of alternate subtypes and emphasizes the clinical importance of intratumoral heterogeneity.

The combination of tumor EVs and cell-specific secreted molecules that can be taken up by neighboring or distant recipient cells, leading to changes in gene expression, suggests a cell-specialized role in physiologic and pathologic conditions (20). We have leveraged the analysis of GSC EV proteomics to characterize heterogeneous extracellular programs within HGG tumors interrelating their release, uptake, and function, which have fundamental implications for biology and therapeutic strategies of HGG, as EVs provide two-way influence through release/uptake between tumors' cells in vivo (Fig. 4D). Analysis of single-cell transcriptome demonstrated transcriptional diversity within an individual tumor (6) and our analysis reveals that transfer of EV protein cargo between tumor cells subpopulations offers a means and opportunity for dynamic transitions.

E.A. Chiocca is a consultant/advisory board member for Tocagen, Nanotx, Alcyone, Merck, and Stemgen. No potential conflicts of interest were disclosed by the other authors.

Conception and design: F. Ricklefs, X.O. Breakefield, E.A. Chiocca, J. Godlewski, A. Bronisz

Development of methodology: F. Ricklefs, M. Mineo, I. Nakano, R. Weissleder, X.O. Breakefield, A. Bronisz

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): F. Ricklefs, M. Mineo, A.K. Rooj

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): F. Ricklefs, M. Mineo, R. Weissleder, E.A. Chiocca, J. Godlewski, A. Bronisz

Writing, review, and/or revision of the manuscript: F. Ricklefs, A.K. Rooj, A. Charest, R. Weissleder, X.O. Breakefield, E.A. Chiocca, J. Godlewski, A. Bronisz

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): F. Ricklefs, A. Charest, A. Bronisz

Study supervision: F. Ricklefs, R. Weissleder, E.A. Chiocca, J. Godlewski, A. Bronisz

The authors thank Hannes Ricklefs for helping in the data analysis by writing a python program.

This work was supported by grants NCI P01 CA69246 (E.A. Chiocca, R. Weissleder, X.O. Breakefield) and by NCI P01 CA163205 (E.A. Chiocca), NCI 1R01 CA176203-01A1 (J. Godlewski), and DFG RI 2616/1-1 (F. Ricklefs).