Comprehensive proteome profiling of glioblastoma-derived extracellular vesicles identifies markers for more aggressive disease

1The need for clinically useful biomarkers is becoming more apparent as the clinical management of glioblastoma (GBM) moves towards individualised therapy and adaptive trial designs. Extracellular vesicles (EVs) are stable, membrane-enclosed particles released from either the cell surface (microvesicles, 100–1000 nm) or from an endosomal route (exosomes, 40–100 nm). EVs are composed of an array of proteins, nucleic acids, lipids, and other metabolites that often reflect the cell of origin, meaning they are excellent reservoirs of biomarkers. Importantly, GBM-derived EVs can cross the brain–blood-barrier and are detectable in the peripheral circulation. Profiling the composition of GBM-derived EVs may, therefore, offer a non-invasive means of assessing tumours in situ, e.g., to identify molecular signatures indicative of tumour progression, recurrence and treatment failure. A ‘liquid biopsy’ would be especially valuable for patients with primary brain tumours, where radiological findings can be ambiguous, i.e., pseudoprogression and neurosurgery carries a very real risk of complication.

2Characterisations of cancer-derived EVs are gaining research momentum also to delineate the role of EVs in the tumour microenvironment. Interestingly, EVs offer an intercellular route to transfer oncogenic material that can change the genetic programme of non-malignant cells, with demonstrated functional consequences in transformed recipient cells related to proliferation, invasion, angiogenesis, chemoresistance and immune repression. Studies have described extensive RNA expression analyses of glioma-derived EVs, however, proteomic profiles are currently limited. Reported protein studies have identified small numbers of proteins (2D-gel electrophoretic or antibody-directed strategies), however relevant to GBM biology, or analysed EVs from limited sources. From other cancer-derived EV studies, we know that EVs contain a subset of cellular proteins, some of which depend on the cell of origin while other proteins are EV-enriched.

3We recently described a comprehensive GBM membrane proteome profile, including several invasion-related proteins that correlated with the cell’s ability to produce invadopodia (actin-rich cellular protrusions with proteolytic activity) under normal culture conditions. Interestingly, invadopodia act as multivesicular endosome (MVE) docking sites and are a site exosome release, meaning the ability to form invadopodia could determine the release of exosomes. Exosome secretion is an essential part of invadopodia biogenesis and maturation, including the release of key invadopodial metalloproteinase, MT1-MMP that degrades the extracellular matrix. Inhibition of two major regulators of invadopodia formation decreased exosome release from squamous cell carcinoma cells; in breast cancer, there are significant associations between cell invasion, invadopodia maturation and EV production. Together these findings not only indicate that EVs are genuine invasion structures of cancer cells, but also point to the potential benefit of profiling EVs as an indirect way to dissect molecular mechanisms of invadopodia biogenesis and function in tumour invasion.

4Here, we provide the most extensive GBM-derived EV protein profile, captured from six cell lines derived from GBM tumours using high-resolution mass spectrometry (MS). To identify candidate proteins associated with more aggressive disease, we performed correlation analyses between EV protein levels and the originating GBM cells’ ability to form invadopodia and then explored corresponding tumour gene expression levels in silico. The in vitro GBM EV proteome profile was then compared to glioma-derived EVs isolated from Cavitron Ultrasonic Surgical Aspirator (CUSA) fluid. The CUSA system is used to fragment and extract solid tumours from the central nervous system. CUSA washings contain tumour tissue fragments that are routinely used in diagnostic pathology; however, the fluid component of CUSA washings is typically discarded. Here we show that this surgical fluid represents a valuable and abundant source of brain tumour EVs. Comparative quantitative proteome analysis of EVs enriched from CUSA fluid collected during a high-grade (GBM) and a low-grade glioma surgical resection was also performed to substantiate the candidate invasion associated EV proteins identified in vitro.

5The mean sizes of the U87MG and LN229 EV were estimated as 92.6 ± 1.2 and 109.9 ± 2.9 nm, respectively (Fig. 1a). Vesicles with diameters of approximately 100 nm were observed using TEM (Fig. 1b, c). Overall, 844 proteins were identified (≥2 peptides, 95 % confidence) of which 145 proteins (17.2 %) were common to EVs secreted by all cell lines (Supplementary Table 1; Vesiclepedia, dataset_599). We identified 15 of the top 20 previously reported exosomal proteins, eight of which were detected in all EV preparations.1 Cytochrome c, a marker for mitochondrial membrane contamination found in apoptotic blebs (i.e. much larger vesicles), was not observed in any EVs. A schematic of the GBM EV proteome is provided (Fig. 2a) and describes a diverse set of proteins associated with MVBs (i.e., PDCD6IP and clathrin), cell adhesion, cytoskeleton, metabolism, membrane trafficking and chaperones. Primary sub-cellular localizations included significant enrichments of exosomal proteins (88.2 %; Fig. 2b). Identities of proteins novel to GBM EVs are annotated in Supplementary Table 1. This is the first account of osteonectin (SPARC; Vesiclepedia ID, VP_6678) and laminin subunit alpha-4 (LAMA4, VP_3910) proteins in cancer EVs, although corresponding mRNA species were documented in GBM EVs. Gene names (145), corresponding to proteins common to all GBM EVs, were mapped in the IPA environment. Prominent up-stream regulators included NFE2L2 (p = 3.53E−20) and TP53 (p = 9.28E−20), with associations to 29 and 75 target molecules, respectively. Significant biological associations included cell growth/proliferation (81 molecules), cell fate (80) and cell-to-cell signalling (53); significant canonical pathways included the protein ubiquitination pathway (23/255), glycolysis I (7/25) and actin cytoskeleton signalling (11/217). Top scoring interaction networks showed functional association to cellular movement, cell fate, cellular growth and proliferation and cell-to-cell signalling (score 132, 78 molecules) and infectious disease, metabolic disease and amino acid metabolism (92, 61).

6Abundance levels of 14 EV proteins significantly correlated to cell invasiveness (r2 > 0.5, p < 0.05, n ≥ 5; Table 1). Significantly associated diseases and cellular functions included cancer (14 molecules), neurological disease (8) cell-to-cell signalling and interaction (8) and cellular movement (8), with significant upstream regulation from TP53, DYSF, PRL, CTNNB1 and RAB7B (6.33E−08 < p value <8.79E− 06). An interaction network was generated using the Path Explorer tool and included links to 12 membrane proteins2 previously detected at higher levels on more invasive GBM cells; several genes corresponding to significant membrane proteins were also predicted to be activated in the generated network (Fig. 3a). Abundance changes of ANXA1, ITGB1 and PDCD6IP in EV lysates from the most (U87MG) and least invasive (LN229) cell lines were confirmed by Western blot (Fig. 3b). ITGB1 levels were also significantly higher in WC lysates of U87MG cells; an inverse relationship was observed for PDCD6IP. Higher ANXA1 levels were also indicated by increased fluorescence during nanosight particle tracking (Fig. 3c).

7Relative gene expression levels corresponding to 14 invasion-related EV proteins were analysed in silico to indicate whether these proteins might be clinically relevant. Transcript levels of ANXA1, IGF2R, ITGB1, PDCD6IP and ACTR3 were significantly higher in GBM specimens, compared with normal brain across all three datasets (Fig. 4a–e). Significant increases in IGF2R were observed in diffuse and anaplastic astrocytomas and oligodendrogliomas, increased PDCD6IP in diffuse and anaplastic astrocytomas, and increased ANXA1 in anaplastic astrocytomas relative to normal brain. Interestingly, significant differences in ANXA1 were observed across the four TCGA GBM transcriptional subtypes, with significantly higher levels in classical and mesenchymal subtypes relative to neural and proneural tumours; proneural tumours displayed significantly lower gene levels compared to other subtypes (Fig. 4f).

8CUSA washings were collected during resections of a low-grade glioma (LGG) and a high-grade GBM (HGG) and tissue fragments processed for diagnostic histopathology (Fig. 5a, b). The HGG was confirmed as a primary WHO2007 Grade IV GBM tumour (IDH1wildtype) and the LGG, a WHO2007 Grade II diffuse astrocytoma (IDH1mutant immunopositive; Fig. 5b2). Crude EVs isolated from HGG (1.0 × 1013 particles/mL) and LGG (9.59 × 1012 particles/mL) CUSA fluid showed particle sizes with mean diameters of 107.9 ± 5.6 and 130.0 ± 1.8 nm, with large populations of 85 and 95 nm sized particles, respectively (Fig. 5c). As it was likely that the crude CUSA EV preparations contain tissue and cellular debris, including contaminating intracellular organelles, EVs were enriched further by density gradient ultracentrifugation and isolated from fractions with densities reflecting the reported range for EVs. Fractions 7–9, corresponding to densities of 1.09–1.11 g/mL, contained particles with combined mean and mode diameters of 122.8 ± 5.6 and 95.2 ± 9.9 nm for HGG EVs and 134.3 ± 5.8 and 103.1 ± 8.7 nm for LGG EVs (Fig. 5d). TEM confirmed a vesicular morphology in the combined density fractions 7–9 from HGG (Fig. 5e1) and LGG (Fig. 5e2). Enriched EV fractions were then subject to quantitative MS analysis. We identified 1559 and 1133 proteins at 95 % confidence levels in at least two of three MS replicates in HGG and LGG EV (Fr7–9), respectively. Of these, 971 proteins were confidently identified in both samples (Fig. 5f). There was considerable overlap with the 145 in vitro GBM EV signature proteins, of which 115 were identified in HGG EVs and 90 in LGG EVs, including SPARC, however LAMA4 was not sequenced. Twenty-five EV proteins were identified in the HGG CUSA EVs alone and may be related to more advanced disease (indicated in Supplementary Table 1). While contamination of mitochondrial and endoplasmic reticulum proteins was observed, our analyses included the identities of 18 (LGG) and 19 (HGG) of the top 20 exosomal proteins (Supplementary Table 1), and approximately half of identified proteins had ‘exosomes’ as a sub-cellular compartment annotation (Fig. 5g). Of the 14 putative invasion proteins identified in the in vitro correlation analysis above, nine proteins (ANXA1, IGF2R, ITGB1, PDCD6IP, ACTR3, CALR, IPO5, MVP, PSMD2) were significantly higher in HGG compared to LGG CUSA enriched-EVs (p < 0.05; Benjamin Hochberg adjusted p value significance threshold p < 0.033; Fig. 5h).

9EV size distributions, morphologies and protein compositions indicate that exosomes are a predominant population in our preparations. Several identified proteins are involved in EV biosynthesis, including members of the ubiquitin-dependent complex ESCRT, i.e., vacuolar protein sorting-associated protein 35 and ubiquitin-like modifier-activating enzyme 1, suggesting that EVs analysed here originate from MVBs. Protein ubiquitination, the top scoring canonical pathway, is important for exosome formation especially during the recruitment of ESCRT machinery into MVB. Although 0.2 µm filtration would theoretically remove microvesicles larger than 200 nm, their presence cannot be discounted for NTA measurements are less precise for larger vesicles (Fig. 1) and our in vitro EV preparations were not purified by density gradient ultracentrifugation.

10To our knowledge, this is the first account of SPARC and LAMA4 proteins in EVs secreted by cancer cells, with previous observations restricted to normal saliva, bloods from healthy donors or from patients with stable coronary artery disease, as well as cultured endothelial cells and embryonic stem cell-derived mesenchymal stem cells.3 TGBF1, also observed in EVs isolated from high-grade glioma patient sera, could be a candidate GBM EV marker. Despite the documented roles of TGFB1, SPARC and LAMA4 in GBM progression and invasion, only the SPARC protein was identified in the two clinical glioma EV preparations profiled here.

11We previously reported significantly higher ANXA1 protein levels in the membrane proteomes of more invasive GBM cells. Increased ANXA1 transcript levels were also observed in GBM and anaplastic astrocytoma tumours compared to normal tissue (Fig. 4) and high ANXA1 expression identified a group of astrocytoma and GBM patients with reduced survival. Interestingly, ANXA1 expression levels change significantly across the TCGA transcriptional GBM subtypes with mesenchymal and classical tumours displaying the highest levels and proneural tumours (frequently IDH1mutated) the lowest. These differences may reflect the molecular disparities between the different tumour strata or perhaps simply, the reported differences in overall survival. ANXA1 protein levels were also significantly higher in the HGG EVs compared with LGG EVs, which again may be associated with differences in tumour invasiveness between low- and high- grade tumours as well as the observed difference in IDH1 mutational status (Fig. 5b2). Understanding the role of ANXA1 in GBM-derived EV is important in determining the influence of EV on the brain tumour microenvironment and role in tumour cell invasion. Further study of ANXA1 as a prognostic biomarker and anti-invasion target is warranted.

12Integrin β1 (ITGB1), a known EV protein, is important for invadopodia formation and ITGB1 gene levels are elevated in GBM tumours (Fig. 4). ITGB1 has multiple direct and indirect interactions with other invasion-associated proteins, particularly ITGA5, which dimerizes with ITGB1 to form the fibronectin (FN1) receptor, α5β1. We recently showed that ITGA5 levels are significantly associated with GBM invasion and reduced patient survival. Following FN1-mediated ubiquitination of ITGA5, α5β1 is sorted into MVEs via ESCRT machinery and destined for lysosomal degradation. Although the orchestrated recycling of FN1-α5β1 (cycles of cell adhesion and detachment) is a requirement for migration, the fate of α5β1 is unknown. It is feasible that more motile cells endocytose more ITGB1, which is then sorted into MVEs that are shed as EVs. As more invasive cells express more ITGA5 on their surface and secrete EVs with more ITGB1, the role of FN1-α5β1 in the GBM tumour microenvironment should be further delineated as it may offer an attractive therapeutic target.

13There was a significant association between high levels of PDCD6IP (also known as ALIX) and ACTR3 and increased invadopodia formation. The interaction network (Fig. 3) included links to several molecules with key regulatory roles in invadopodia formation, i.e., SRC, actin regulatory complex Arp2/3, WASF1. ACTR3 is an ATP-binding component of Arp2/3 and together with an activating nucleation-promoting factor such as Wiskott-Aldrich syndrome protein (WASP), WASF1 (WAVE) or WASH, it mediates actin polymerization and invadopodia formation. PDCD6IP and ACTR3 are both involved in endosomal sorting, which is important for exosome biogenesis. As invadopodia are proposed as sites for exosome secretion, this could imply that more invasive or invadopodia-producing cells secrete more exosomes. This is supported by observations that tumours cells produce more exosomes per cell than normal cells and PDCD6IP and ACTR3 levels are higher in GBM tumours compared to normal tissue (Fig. 4).

14Calreticulin (CALR) levels were increased in EVs secreted by more invasive GBM cells. CALR is a critical regulator of Ca2+ homeostasis, its overexpression increases intracellular Ca2+. Increased cellular Ca2+ stimulates exosome secretion, which again supports the notion that more invasive GBM cells secrete more exosomes. Major vault protein (MVP) mRNA was previously observed in GBM-derived EVs and significantly higher protein levels were identified in more invasive GBM EVs here. MVP facilitates the nuclear tumour-suppressing function of PTEN in a Ca2+ dependent manner; nuclear PTEN is unable to inhibit PI3K signalling, leading to a more malignant phenotype. Interestingly, EVs are highly enriched in vault RNAs that complex with MVP to form the vault organelle that plays important roles in transport mechanisms, signalling and immune responses. MVP is upregulated during malignant transformation and tumour progression and has been linked to chemoresistance. CALR was shown to promote invasion by increasing MMP-2 and MMP-9 and is implicated in regulating radiosensitivity and radiation-induced apoptosis in GBM. CALR is also a critical component of antigen processing and loading into MHC I. Higher CALR levels in more invasive GBM EVs might be important for local and distant intercellular communication and have immunogenic modulatory effects.

15CALR functions as a chaperone for amyloid beta A4 protein (APP), also identified at significantly higher levels in EVs from more invasive cells. Increased APP protein levels were observed in GBM tumours and APP metabolites are enriched in exosomes purified from brain tissues. Increased APP was shown to up-regulate leucine-rich glioma inactivated-3 in rat astrocytes, which interacts with flotillin-1 to mediate APP trafficking, endocytosis and exosome formation in neuronal cells. Elevated APP expression is also associated with gliosis and is the main component of the senile plaques; Alzheimer’s pathology is present in about half of all cases of GBM. While APP seems to be a part of a poorly understood cell-contact signalling pathway, elevated APP levels in invasive GBM-derived EVs suggests that this communication occurs via EV delivery.

16Insulin-like growth factor 2-receptor (IGF2R) was measured at higher levels in more invasive GBM EVs and tumour mRNA levels were higher in gliomas than normal brain (Fig. 4). Insulin-like growth factor-binding protein 2, a glioma marker linked to poor prognosis, binds to and modulates IGF2R. IGF2R also bind cathepsins that are typically localised to lysosomes; EV biogenesis is now understood to involve pathways common to lysosome degradation. Cathepsin D (CTSD) was also increased in EVs from more invasive cells, its release and activity is linked to glioma invasion, and may act directly by degrading local ECM structures or indirectly through activation of cysteine proteinases. Interestingly, elevated CTSD serum levels correlate with glioma grade and high CTSD transcript levels in GBM tumours is associated with reduced survival. CTSD levels in circulating EVs might offer valuable, non-invasive prognostic information. ECM1 overexpression is associated with poor prognoses in breast, gastric and laryngeal cancer. ECM1 mRNA is enriched in GBM-EVs compared to cells. Higher EV protein levels detected here as well as previous links to more aggressive cancer phenotypes suggest that ECM1 is an interesting target for further study.

17While the phenotype of a cell or tissue correlates directly with protein expression, they may not correlate with mRNA levels, therefore the expression levels of the nine invasion-associated proteins that did not show significance in silico may still be useful protein biomarkers. Along with the five proteins that did show concordance with mRNA levels in silico, CALR, IPO5, MVP and PSMD2 protein levels were significantly higher in HGG compared to LGG enriched EVs; ECM1 levels were also higher, however with borderline significance (p = 0.060).

18Although the ability to detect appropriate biomarkers in the peripheral circulation is the sine qua non of a liquid biopsy, EVs isolated from peripheral blood pose two key problems during the initial discovery phase of biomarker development and should be considered for translation of the in vitro GBM EV protein signature described here. Firstly, the presence of high abundance proteins (albumin, immunoglobulins, transferrin and lipoproteins etc.) comprise ~99 % of the protein content of blood, masking the presence of low abundance proteins that are of major interest for biomarker discovery and make high throughput proteomic analysis of serum or plasma-derived EVs problematic. Secondly, EVs are secreted by all bodily organs with a significant proportion in the blood being platelet-derived. Tumour-derived EVs exist at relatively low concentrations within the blood compared to the total EV population; high enough for targeted detection, but not sufficient for the bottom-up, high throughput analytical approaches for biomarker discovery. This necessitates enrichment steps, which are still in the process of being standardised, precluding comprehensive and collaborative analysis in many cases; the establishment of a ‘gold-standard’ for EV sourcing remains elusive. An ideal EV source would need to deliver a relatively homogeneous mixture at high enough concentration that even low abundance particles could be detected and quantified. While the limitations of using immortalized cultured cells as models of complex heterogeneous disease are patent, there are no pan-GBM specific EV surface proteins described that would enable positive isolation methods from the blood. Thus, obtaining enriched sources of brain tumour EVs, i.e., from homogenized tumours, surgical aspirates or cerebrospinal fluid, presents the most plausible approach to translate the proposed GBM EV signature here as well as further test EV-associated biomarkers. Our preliminary analyses here indicate that CUSA fluid represents a valuable and abundant source of brain tumour EVs. Once confirmed, candidate biomarker proteins would also require further assessment to determine whether they are exploitable as biomarkers, i.e., that levels are above that of background soluble or normal EV levels in the peripheral circulation.

19A common set of 145 proteins was identified in EVs secreted by six GBM cell lines and may be useful for distinguishing GBM-specific EVs in the circulation. Many of the invasion-related EV proteins resolved are associated with key molecules involved in regulating invadopodia formation. Gene levels corresponding to five invasion-related EV proteins (ANXA1, ACTR3, ITGB1, IGF2R and PDCD6IP) were significantly higher in GBM lesions, with common functions relating to actin polymerisation and endosomal sorting. Several targets identified here warrant further testing as potential biomarkers, including the putative prognostic marker ANXA1. The role of FN1 and it’s integrin-α5β1 receptor in the GBM microenvironment should be further delineated and the inhibition of this association, e.g., by volociximab treatment, should be assessed. On the whole, these data indicate that more invasive GBM cells secrete more exosomes, a strategy that perhaps allows tumours to hijack their microenvironment and modulate anti-tumour immunity. Finally, we have identified CUSA washings as a novel source of brain tumour-derived EVs. The analysis of which could expedite the translation of clinically relevant blood-based biomarkers for GBM patient management.

20Below is the link to the electronic supplementary material.