Liquid biopsy has great potential to improve the management of brain tumor patients at high risk of surgery-associated complications. Here, the aim was to explore plasma extracellular vesicle (plEV) immunoprofiling as a tool for noninvasive diagnosis of glioma.
PlEV isolation and analysis were optimized using advanced mass spectrometry, nanoparticle tracking analysis, and electron microscopy. We then established a new procedure that combines size exclusion chromatography isolation and proximity extension assay–based ultrasensitive immunoprofiling of plEV proteins that was applied on a well-defined glioma study cohort (n = 82).
Among potential candidates, we for the first time identify syndecan-1 (SDC1) as a plEV constituent that can discriminate between high-grade glioblastoma multiforme (GBM, WHO grade IV) and low-grade glioma [LGG, WHO grade II; area under the ROC curve (AUC): 0.81; sensitivity: 71%; specificity: 91%]. These findings were independently validated by ELISA. Tumor SDC1 mRNA expression similarly discriminated between GBM and LGG in an independent glioma patient population from The Cancer Genome Atlas cohort (AUC: 0.91; sensitivity: 79%; specificity: 91%). In experimental studies with GBM cells, we show that SDC1 is efficiently sorted to secreted EVs. Importantly, we found strong support of plEVSDC1 originating from GBM tumors, as plEVSDC1 correlated with SDC1 protein expression in matched patient tumors, and plEVSDC1 was decreased postoperatively depending on the extent of surgery.
Our studies support the concept of circulating plEVs as a tool for noninvasive diagnosis and monitoring of gliomas and should move this field closer to the goal of improving the management of cancer patients.
The development of noninvasive strategies for brain tumor diagnosis remains a challenge of high clinical relevance. Extracellular vesicles (EV) have received considerable attention as a potential liquid biopsy biomarker as they represent “a miniature of its cell of origin.” Here, using an interdisciplinary approach, we have developed a new procedure for the isolation and analysis of patient plasma EVs. In a well-defined glioma patient study cohort, we for the first time identify syndecan-1 (SDC1) as an EV constituent that noninvasively can discriminate between glioblastoma multiforme (GBM; WHO grade IV) and low-grade glioma (WHO grade II). Importantly, we found strong support for EV-SDC1 originating directly from GBM tumors. We conclude that tumor-derived EVs may serve as a potential tool to facilitate minimally invasive diagnosis and monitoring of gliomas, which should stimulate future efforts to move this field closer to the goal of improving the management of cancer patients.
Primary brain tumors remain among the most challenging forms of cancer to diagnose and treat in adults and children. The World Health Organization uses histologic criteria to discriminate between low-grade (I and II) glial tumors with a relatively good prognosis and high-grade tumors [III and IV, or glioblastoma multiforme (GBM)], which are the most common and aggressive primary brain tumors in adults (1, 2). These molecular classifications are currently dependent on tissue samples obtained after biopsy or tumor resection. However, this is not always feasible, as intracranial tumors are among the most inaccessible for a diagnostic biopsy due to considerable risks of postsurgical complications (e.g., neurologic damage, bleeding, infections), resulting in the delayed onset of oncological treatment and ultimately worse patient outcome. The development of noninvasive strategies for the diagnosis of glioma tumors thus remains a challenge of high clinical relevance, especially with regard to tumor spatiotemporal heterogeneity (3).
Extracellular vesicles (EV) are small (∼40–1,000 nm), lipid bilayer-enclosed vesicles secreted into a variety of biological fluids, including plasma. EVs have emerged as critical components in intercellular communication during the development and progression of cancer (4–7). The comprehensive molecular content of EVs, including proteins, RNA, DNA, and lipids, largely mimics their cell or tissue of origin, and tumor cells, in general, secrete excessive amounts of EVs. This has stimulated considerable efforts to develop EVs as a tool for minimally invasive diagnosis of cancer (8, 9). In GBM, a pilot study from our group suggested that the plasma EV (plEV) proteome reflects the tumor oxygenation status (10), whereas others have shown that serum and cerebrospinal fluid derived EVs correlate with EGFR expression and mutation status, tumor size, and treatment response (11–18). Although these studies are based on relatively small patient cohorts, they provide important support to the concept that EVs reflect the molecular profile of GBM tumors and can differentiate between GBM patients and healthy control subjects; however, the need of discriminating between benign and malignant gliomas of different grades, which would have more obvious clinical impact, remains unmet.
To reach the full potential of EVs as a source of biomarkers and to distinguish plEVs from contaminating plasma proteins (19, 20), it is essential to identify specific tumor-derived EV proteins to facilitate downstream, multiomics profiling studies. Motivated by clinical needs, we have used size exclusion chromatography (SEC), advanced mass spectrometry, and an ultrasensitive proximity extension immunoassay in a well-defined population cohort of brain tumor patients in the search for tumor-derived plEV proteins.
Materials and Methods
A detailed description of LC-MS/MS experiments is given in Supplementary Materials and Methods.
Study design and sample collection
The patient material was from a population-based trial cohort (“MRI study”) encompassing patients referred to the Neurosurgery Department at Lund University Hospital, Lund, Sweden, with a suspicion of an intracranial tumor. Inclusion criteria were age 18 years or above, WHO performance status 0 to 3, and ability to give written informed consent before study entry. The study was carried out according to the ICH/GCP guidelines and in agreement with the Helsinki declaration, and was approved by the local ethics committee, Lund University (Dnr. 2011/814 and 2012/188). Patients were diagnosed by routine magnetic resonance imaging (MRI) of the brain (3T Magnetom Skyra, Siemens AG), surgical and pathologic procedures, received standard oncological treatment and were followed up according to local and national recommendations as well as by repeated MRI examinations according to the study protocol including sagittal T1-w, axial T2-w, axial diffusion-weighted imaging, axial and coronal T1-w ± gadolinium. Blood samples were collected in EDTA tubes, centrifuged at 2,000 × g for 10 minutes at room temperature (RT) and stored in a −80°C freezer. Longitudinal plasma samples were collected at baseline (preoperative) and 3 weeks after surgery (postoperative) prior to start of the oncological treatment. The present biomarker cohort was established at the cutoff date of September 1, 2016, consisting of the first consecutive 136 patients. Before processing the clinical cohort, we performed extensive optimization of plEV isolation by SEC from control subjects at the Department of Oncology, Lund University. We validated the suitability of the plEV isolation techniques and its downstream processing in identifying potential EV protein biomarkers by conducting in-depth proteomic analyses using advanced LC-MS/MS (Fig. 1). Unblinding of clinicopathologic parameters and corresponding experimental data was done after finishing all experiments.
U87-MG GBM cells were purchased from ATCC and routinely cultured in DMEM growth medium, supplemented with 10% fetal bovine serum (FBS), 2 mmol/L L-glutamine, 100 U/mL penicillin, and 100 μg/mL streptomycin (PEST). The U3043 cells are part of the Uppsala University Human Glioma Cell Culture resource (HGCC; www.hgcc.se) of patient-derived GBM cells and were cultured on laminin in serum-free medium supplemented with EGF, FGF, and stem cell supplements, as previously described (35). Primary human brain microendothelial cells (HBMEC) were purchased from 3H Biomedical, and cultured in Endothelial culture medium (EC medium; 3H Biomedical) supplemented with 5% FBS, 1% EC growth supplement, and 1% PEST. HBMECs at passages 2 to 4 were used for experiments. All cells were normally kept at 37°C in a humidified 5% CO2 incubator.
EVs were isolated from U87-MG cells by differential ultracentrifugation, as previously described (10). Briefly, subconfluent cells were grown in serum-free DMEM supplemented with 1% BSA at normoxic or hypoxic conditions for 48 hours, and conditioned media were collected and centrifuged at 300 × g twice to eliminate cell debris. Supernatant fractions were then centrifuged at 100,000 × g for 2 hours to pellet EVs, followed by washing twice with PBS at 100,000 × g for 2 hours.
For the isolation of EVs from plasma samples, sepharose-based CL-2B SEC columns (IZON Science) were used. Plasma samples were thawed on ice for the first time after freezing. Five hundred microliter plasma aliquots were applied to the column, and 15 fractions of 500 μL were collected immediately with 2 mmol/L CaCl2 in PBS as the elution buffer. Fractions 5 to 9, corresponding to the EV elution profile, were pooled and lysed by 10 cycles of freeze–thaw, with each cycle of freezing (15 minutes on dry ice) followed by thawing (3 minutes in an ultrasonic bath). Lysed EV proteins were desalted using PD-10 columns (GE Healthcare) according to the standard protocol. Purified EV proteins were concentrated by freeze-drying using Mini Lyotrap (LTE Scientific) and stored at −80°C until further use.
EV processing for LC-MS/MS
Lysed EVs in 6 M urea/50 mmol/L ammonium bicarbonate were reduced with 10 mmol/L dithiothreitol (DTT) for 1 hour at 56°C with gentle shaking and alkylated using 50 mmol/L iodoacetamide (IAA) for 30 minutes in the dark at RT. Thereafter, protein samples were digested with sequencing grade trypsin (Promega) overnight at 37°C with gentle shaking. The digestion was stopped by adding 2% trifluoroacetic acid (TFA; 1:10 v/v), and the samples were dried in a SpeedVac. Subsequently, they were either stored at −80°C or resuspended in 0.1% TFA for further analysis.
LC-MS/MS, normal gradient (NG), long gradient (LG), and tandem mass tag (TMT) labeling and high-resolution isoelectric focusing (HiRIEF) were performed as described in Supplementary Materials and Methods
Nanoparticle tracking analysis (NTA).
NTA was applied to determine the size and concentration of particles and to confirm that their size was equivalent to that of EVs (52). Particles were tracked on an LM10-HS system with a 405-nm laser (Malvern Instruments) and visualized with a Luca-DL EMCCD camera (Andor Technology). Standard silica beads (0.1 μm) were used to calibrate the analysis settings with a camera level of 10 and detection threshold 2 with blur 9 × 9. A total of five videos each of 30 seconds were recorded for the individual samples. Prior to analysis, the samples were diluted in PBS to ensure a particles/frame count within the manufacturer's recommendations. Particles were tracked, quantitated, and size enumerated using the NanoSight NTA software version 3.0 (Malvern Instruments).
Transmission electron microscopy (TEM).
Approximately 5 μL of isolated EVs was adsorbed onto 400-mesh carbon-coated gold grids. The dried samples were blocked with 1% BSA and incubated with anti-SDC1 rabbit monoclonal antibody (1:1,000; ab128936, Abcam) for 1 hour at RT. After washing, samples were incubated with 10-nm gold-conjugated anti-rabbit secondary antibody (1:20; 15726, TED Pella Inc.). Control samples were incubated with only secondary antibody. Subsequently, the samples were fixed with 1% glutaraldehyde, stained with 2% uranylacetate, and examined in FEI Tecnai BioTWIN TEM operated at an accelerating voltage of 100 KV. Images were recorded with an Olympus SIS Veleta CCD camera.
ProSeek multiplex proximity extension assay (PEA).
EV protein was analyzed using the ProSeek Multiplex Oncology II96x96 and CVD III96x96 panels (Olink Bioscience), as previously described (26, 27). The protein quantification is based on PEA technology, which provides high sensitivity and specificity based on the binding of oligonucleotide-labeled antibody probe pairs to their specific target protein, generating a PCR-amplified DNA template, which is proportional to the initial antigen concentration as quantified by real-time qPCR. Four internal and three negative controls were used to calculate the lower limit of detection (LOD) for each protein.
SDC1 levels in plEV samples were analyzed using the Human SDC1 ELISA Kit (Thermo Scientific) according to the manufacturer's instructions. Briefly, 100 μL of each standard and sample were added to appropriate wells and incubated for 2.5 hour at RT with gentle shaking. After washing, 100 μL of 1 × biotinylated antibody was added and incubated for 1 hour at RT. Following another wash, 100 μL of streptavidin–HRP solution was added to each well and incubated for 45 minutes at RT. One hundred microliters of TMB substrate was then added to samples for 30 minutes at RT in the dark with gentle shaking, followed by the addition of 50 μL of stop solution. The absorbance was measured at 450 nm and 550 nm in a spectrophotometer (FLUOstar OPTIMA).
HBMECs were starved overnight in serum-free endothelial cell culture medium supplemented with 1% L-Glut and 1% PEST. Cells were added in serum-free medium to the top chamber of 8-μm pore cell culture inserts (BD Biosciences) placed in a 24-well plate. Cells were incubated for 6 hours at 37°C to allow cell migration toward serum-free medium supplemented with plEVs isolated from patients or healthy controls (1.5 μg/mL) or serum-free medium with no additions. Migrated cells attached to the bottom membrane were fixed with 4% (w/v) paraformaldehyde, stained with crystal violet, and counted from pictures taken under the microscope (Axiovert 40C, 4 × objective; Carl Zeiss).
EV protein extracts were mixed with NuPAGE 4 × LDS Sample Buffer (Life Technologies) and heated for 10 minutes at 80°C. Proteins were resolved in a NuPAGE 4% to 12% Bis Tris gel (Life Technologies) at reducing conditions, and then transferred onto a polyvinylidene fluoride (PVDF) membrane (Immobilon-FL), followed by blocking in TBS containing 0.05% Tween-20, 5% nonfat dry milk for 1 hour at RT. To probe for SDC1, the membrane was incubated with rabbit monoclonal anti-SDC1 (1:10,000; ab128936, Abcam) in TBS 0.05% Tween 20 containing 5% nonfat dry milk overnight at 4°C. After washing, the membrane was incubated with HRP-conjugated anti-rabbit secondary antibody (1:3,000; 7074, Cell Signaling Technology). Protein bands were visualized by enhanced chemiluminescence Western blotting substrate (Pierce).
Cells were seeded on 4-chamber well slides (10,000 cells/well) overnight, fixed with 4% (w/v) paraformaldehyde for 10 minutes, and permeabilized with 0.5% Saponin/PBS for 15 minutes at RT. After washing, cells were incubated with rabbit monoclonal anti-SDC1 (ab128936, Abcam; dilution 2 μg/mL) for 1 hour at RT. After further washing, cells were incubated with AF-546-conjugated goat anti-rabbit secondary antibody (A11010, Life Technologies; dilution 2 μg/mL) for 1 hour in PBS/1% BSA in the dark for 10 minutes at RT. Control cells were stained with secondary antibody only. Cell nuclei were counterstained with Hoechst (1:20,000). Cells were analyzed using Zeiss LSM 710 confocal scanning equipment with a Plan-Apochromat20×/0.8 objective or a C-Apochromat 63x/1.2 W Korr objective and Zen software.
IHC staining for SDC1 was performed on an automated IHC staining platform (Ventana Medical Systems, Roche) according to the manufacturer's instructions. Briefly, tissue was fixed with 4% formaldehyde, dehydrated, and paraffin-embedded. Following deparaffinization, antigen retrieval was performed by treatment in Cell Conditioning solution (CC1, 950-124, Ventana Medical Systems, Roche) for 1 hour at 95°C. Sections were then stained with mouse anti-human SDC1 antibody (B-A38; 760-4248, Ventana Medical Systems, Roche) for 50 minutes at 36°C, followed by visualization using 3,3′-diaminobenzidine (DAB) Ultraview (Ventana Medical Systems, Roche) according to the manufacturer's instructions.
Functional enriched pathway analyses of proteins were performed using the ConsensusPathDB–human interaction network database (http://cpdb.molgen.mpg.de/). GBM WHO grade IV and LGG (astrocytoma grade II) RNA-seq data for survival were downloaded from The Cancer Genome Atlas (TCGA) data portal (https://tcga-data.nci.nih.gov/docs/publications/tcga/) as per recommended network instructions. The TCGA RNA-seq data set for SDC1 expression was analyzed using data from the cBioPortal (http://www.cbioportal.org/). The TCGA GBM RNA-seq data sets for SDC1 expression in IDH wild-type and mutant, tumor subtypes, and gene correlations (r-value) were analyzed using data obtained from the GlioVis portal (http://gliovis.bioinfo.cnio.es/). In the PEA analysis, protein levels were expressed as Normalized Protein eXpression (NPX) values, an arbitrary unit on log2-scale. The LOD for a protein is computed based on blank samples as three times the standard deviation above the mean. The association between a protein NPX value and patient group (GBM or LGG) was investigated using linear regression, adjusting for age (there was no significant association between gender and group), and assessed using a likelihood ratio test. The linear regression approach was applied to all proteins with less than 20% values below LOD, and all samples with a value below LOD were excluded in the analysis. Proteins with 80% or more values below LOD were excluded from the analyses. Proteins with 20% or more values below LOD (but less than 80%) were discretized (coded as 0 = below LOD, 1 = above LOD). The association between a discretized protein variable was estimated using logistic regression, adjusting for age and using a likelihood ratio test, as for the linear regressions described above. The association of plEVSDC1 and plEVITGB2 expression with glioma grade (GBM WHO grade IV and LGG/astrocytoma WHO grade II) was calculated using unpaired two-tailed Student t test. Associations between plEVSDC1 and plEVITGB2 expression and GBM patient survival were analyzed using log-rank (Mantel–Cox) χ2 test. Receiver operating characteristic (ROC) curves were used to determine the specificity and sensitivity and were expressed as area under the ROC curve (AUC). The cutoff points were selected using the Youden index, which maximizes the sum of sensitivity and specificity. Group differences involving GBM subtypes were tested by the simple one-way ANOVA test. In the PEA analyses, P values were adjusted for multiple tests using the Benjamini–Hochberg method for controlling the false discovery rate (FDR < 0.05). P < 0.05 was considered statistically significant. Data are presented as the means ± SD. All figures were prepared and analyzed using either GraphPad Prism version 7.0 (GraphPad Software) or in R version 3.4.2.
Optimization of EV isolation from patient plasma
To optimize procedures for the isolation of EVs from patient plasma, we initially compared well-established centrifugation procedures for in vitro isolation of glioma cell-derived EVs (gcEV; refs. 10, 21) with SEC-based isolation of plEVs from patients (Fig. 1A). The distribution of total protein and particles in separate SEC fractions (15 fractions were collected) was assessed by gel electrophoresis and NTA, revealing that the bulk of plasma proteins gradually increases from fraction 10 (Fig. 1B), and that the particle concentration peaked in SEC fraction 9 (Fig. 1C). This is consistent with the typical elution pattern of EVs from sephadex-based SEC columns with a pore size of approximately 75 nm (19, 22, 23). NTA of gcEVs (Fig. 1D) and plEVs from SEC fractions 5 to 9 (Fig. 1E) showed comparable size distributions with a median particle size of approximately 130 and 110 nm, respectively. Electron microscopy (EM) studies corroborated these data, showing similar size distribution and morphology of gcEVs and plEVs (Fig. 1F and G). The composition of plEVs was initially determined by NG shotgun LC-MS/MS proteomics and compared with the gcEV proteome (Fig. 1H; Supplementary Data File S1). We found 328 overlapping protein identities, and there was expectedly high coverage for abundant plasma proteins such as complement factors, coagulation factors, and apolipoproteins in plEV isolates (Supplementary Data File S1). With the aim of improving the detection of low abundant plEV proteins, we used an LG LC-MS/MS approach that increased the overlap with gcEVs (n = 553), but did not substantially enhance the coverage of typical EV proteins (Fig. 1H; Supplementary Data File S1). HiRIEF fractionation and TMT LC-MS/MS of plEVs further increased the number of overlapping protein identities with gcEVs (n = 772; Fig. 1H) and, more importantly, revealed the presence of several established EV markers, including tetraspannins (e.g., CD81, CD9, CD63), annexins, RABs, heat shock, and ESCRT proteins (Supplementary Table S1; Supplementary Data File S1; refs. 24, 25). Functional pathway analyses showed the enrichment for several pathways implicated in glioma biology, e.g., EGFR and integrin mediated signaling and angiogenesis (Supplementary Fig. S1A). We conclude that SEC-based separation and high-resolution MS/MS allowed the identification of several EV proteins previously unidentified by LC-MS/MS in complex samples such as plEVs (19), supporting the feasibility for downstream analyses.
Identification of syndecan-1 by targeted analysis of glioma patient plEVs
To explore plEVs as a minimally invasive tool for the discrimination between high-grade and low-grade gliomas (LGG), we applied the SEC protocol on plasma collected within a population-based, clinical study cohort encompassing consecutive patients referred to the neurosurgery department with a suspected brain tumor lesion (n = 136). In this cohort, 69 patients were diagnosed with high-grade (WHO IV) GBM and 17 with low-grade (WHO II) astrocytoma (hereafter referred to as LGG), as determined by surgery and histopathologic examination according to the clinical routine (Supplementary Fig. S1B). Baseline (preoperative) plasma samples were available from all LGG patients and 65 out of 69 GBM patients (in total n = 82; patient characteristics are presented in Supplementary Data File S2) that were included for SEC separation and further processing. The representativeness of the present cohort was determined by a relative survival of GBM vs. LGG patients (HR: 0.21; 95% CI: 0.12–0.36; P < 0.001; Supplementary Fig. S2A) that was comparable with data from TCGA cohort encompassing 850 patients (HR: 0.17; 95% CI: 0.14–0.19; P < 0.001; Supplementary Fig. S2B). PlEVs were isolated from all patients by SEC and analyzed using an ultrasensitive immunoassay based on PEA technology (Fig. 2A; refs. 26, 27). The multiplex format allowed the simultaneous analysis of 183 proteins that were selected based on their known involvement in glioma biology and potential sorting to EVs (Supplementary Table S2). Half of the analyzed proteins (n = 92) were above LOD in at least 20% of the samples. Using linear regression and adjusting for age (there was no gender difference between GBM and LGG groups), in total 12 plEV-associated proteins significantly differed between GBM and LGG (P < 0.05), half of which were detectable in at least 98% of plEV samples (Fig. 2B). When adjusting for multiple testing, specifically syndecan-1 (SDC1) differed significantly between GBM and LGG (Fig. 2C), whereas integrin beta chain-2 (ITGB2) showed a strong trend, however not significant (Fig. 2D). Notably, SDC1 and ITGB2 are single-pass type I membrane proteins, which identifies them as potential EV surface biomarkers. To explore the diagnostic accuracy of plEV, ROC curves were established. The ROC analysis revealed that plasma EV-SDC1 (plEVSDC1) could discriminate patients with GBM from LGG, with an AUC value of 0.81 (95% CI: 0.69–0.93; Fig. 2E), and a sensitivity of 71% (95% CI: 44%–89%) and specificity of 91% (95% CI: 81%–97%). For plasma EV-ITGB2 (plEVITGB2) AUC was 0.77 (95% CI: 0.65–0.88; Fig. 2F) with a sensitivity of 65% (95% CI: 38%–86%) and specificity of 71% (95% CI: 58%–81%).
Syndecan-1 tumor expression discriminates between GBM and LGG
Interestingly, SDC1 and ITGB2 tumor mRNA expression data retrieved from the TCGA cohort (n = 850) revealed that both transcripts were significantly higher in GBM as compared with LGG tumors (Fig. 3A; Supplementary Fig. S2C). ROC analysis of SDC1 mRNA showed an AUC of 0.91 (95% CI: 0.89–0.93; Fig. 3B) with a sensitivity of 79% (95% CI: 75%–82%) and specificity of 91% (95% CI: 87%–94%). For ITGB2, AUC was 0.76 (95% CI: 0.72–0.79; Supplementary Fig. S2D), with a sensitivity of 70% (95% CI: 66%–74%) and specificity of 67% (95% CI: 59%–74%). Moreover, high SDC1 expression was associated with wild-type isocitrate dehydrogenase (IDH) typically found in primary GBM (Fig. 3C), and the mesenchymal GBM subtype (ref. 28; Fig. 3D), and correlated with a more aggressive tumor phenotype, as supported by an association between high SDC1 expression and worse patient outcome (Fig. 3E; high vs. low SDC1; HR: 0.66; P < 0.05). However, there was no significant association between tumor ITGB2 and GBM patient outcome (Supplementary Fig. S2E; P = 0.24). These gene-expression data from a large, independent glioma cohort data thus corroborated our findings from proteomic analyses of plEVs (Fig. 2). We conclude that plEVSDC1 as well as tumor SDC1 mRNA can discriminate between GBM and LGG, and correlate with GBM aggressiveness.
Syndecan-1 expression correlates with a hypoxic and proangiogenic phenotype
In accordance with its association to the mesenchymal GBM subtype, SDC1 expression correlated with several components involved in stroma remodeling and angiogenesis (Fig. 3F; Supplementary Fig. S3A). SDC1 further correlated with molecules known to be hypoxia-induced (29), which was corroborated in GBM cells in vitro by gene array analyses (Supplementary Fig. S3B). EVs derived from GBM cells have shown potent proangiogenic effects (11), especially when derived from hypoxic conditions (10). Accordingly, in pilot studies we could show that plEVs from GBM patients (SDC1-positive, as determined by ELISA; n = 3) were significantly more potent at stimulating primary HBMEC migration as compared with plEVs from healthy subjects (SDC1-negative; n = 3; Fig. 3G). We conclude that SDC1 tumor expression associates with glioma grade and characteristics typical of GBM aggressiveness.
Syndecan-1 is sorted to GBM patient plEVs and GBM cell–derived EVs
The identification of SDC1 in plEVs was of particular interest given its suggested functional role in GBM (30–32) and other malignancies (33), and its key role in EV biogenesis and release as part of the SDC1–syntenin–ALIX pathway (34). Immunofluorescence staining for SDC1 in U3034 primary GBM cells (35) visualized a cytoplasmic, vesicular as well plasma membrane distribution (Fig. 4A). The vesicular localization of SDC1 was corroborated by costaining experiments for SDC1 and vesicle markers, showing colocalization of SDC1 with the endosomal marker EEA1 and the late endosome/lysosomal and EV marker CD63 (Supplementary Fig. S4A). A similar distribution of SDC1 was found in GBM patient tumors (Fig. 4B), while SDC1 was not detected in normal brain tissue (Fig. 4C). Further, SDC1 was efficiently sorted to the EV fraction while undetectable in the soluble fraction in culture media from primary (Fig. 4D) as well as established (Fig. 4E) GBM cells. Notably, low density lipoprotein (LDL), a common contaminant of plEV preparations (19, 20), was negative for SDC1 (Supplementary Fig. S4B). We next performed EM studies of plEVs isolated from GBM patients and gcEVs from GBM cell medium. Both EV sources showed positive labeling for SDC1 (Fig. 4F; Supplementary Fig. S4C). We conclude that SDC1 is efficiently sorted to GBM-derived EVs and provide evidence that SDC1 is associated with plEVs isolated from GBM patients.
PlEVSDC1 correlates with glioma grade and tumor SDC1 protein expression
We next sought to validate plEVSDC1 data from the glioma PEA cohort by a quantitative immunoassay. Interestingly, ELISA-based quantification of plEVSDC1 could discriminate between GBM and LGG patients with a higher significance level (P = 0.0002; Fig. 5A) as compared with the PEA-based analysis (P = 0.0498; Fig. 2C). ELISA appeared less sensitive than PEA with 13% (8 of 61), 65% (11 of 17), and 100% (4 of 4) samples below detection level in GBM, LGG, and healthy controls, respectively. ROC analysis showed an AUC of 0.82 (95% CI: 0.71–0.93; Fig. 5B) with a sensitivity of 71% (95% CI: 0.44%–0.89%) and specificity of 80% (95% CI: 68%–89%). In concordance with the TCGA data (Fig. 3C), we found a significant correlation between plEVSDC1 and IDH status, i.e., high plEVSDC1 was associated with wild-type IDH (Supplementary Fig. S2F). Further, the association between plEVSDC1 and GBM patient survival (high vs. low plEVSDC1; HR: 0.61; P < 0.05; Fig. 5C) was similar to SDC1 survival data in TCGA patients (high vs. low SDC1; HR: 0.66; P < 0.05; Fig. 3E), thus providing indirect support that plEVSDC1 levels reflect tumor SDC1 expression. To investigate this more directly, matched tumors from patients with low, moderate, and high plEVSDC1 were analyzed for SDC1 expression by IHC. Importantly, SDC1 expression in paired plEV and tumor samples closely correlated; we observed virtually no staining in plEVSDC1 low patients, and limited and strong signal in tumors from patients with moderate and high plEVSDC1 levels, respectively (Fig. 5D). To further investigate how plEVSDC1 levels correlate with tumor status, we evaluated plEVSDC1 of GBM patients preoperatively and postoperative day 21 prior to start of the oncological treatment (n = 15). Overall, the results showed a significant decrease in plEVSDC1 following surgery (Fig. 6A). However, this analysis also revealed interindividual variations; in addition to the patients showing decreasing plEVSDC1 levels (n = 10), 2 patients had unchanged plEVSDC1 and 3 patients showed increased plEVSDC1 (Fig. 6A). We hypothesized that this reflects differences in the extent of surgery that ranged from biopsy to complete resection. To test this possibility, we analyzed available MRI data, revealing that patients who had a total or subtotal resection also showed a postoperative decline in plEVSDC1 (Fig. 6B). On the contrary, MRI data from patients with unaltered or even increased postoperative plEVSDC1 showed remaining contrast enhancement following surgery (Fig. 6C). We conclude that glioma tumors release plEVSDC1 into the circulation depending on their histologic grade, aggressiveness, and extent of surgical removal.
We report the establishment of a procedure for the isolation of plEVs combined with targeted analyses of almost 200 proteins using ultrasensitive immunodetection technology, and identify SDC1 as a plEV constituent for noninvasive differentiation between GBM and LGG. SDC1 tumor expression similarly discriminated between GBM and LGG in a large glioma patient population from the TCGA cohort. Importantly, tumor-derived EVs remain a rarity in clinical samples, i.e., it was essential to elucidate how plEVSDC1 correlates with tumor SDC1. We find strong support of plEVSDC1 originating from GBM tumors by analyses of SDC1 expression in matched patient tumor specimens as well as in longitudinal (pre- and postoperative) plEV isolates. Together, our results provide important support to the concept of EV proteomics for noninvasive brain tumor diagnosis. In addition, we provide new insights into the role of SDC1 in glioma biology.
The attractiveness of EVs as a circulating biomarker relies on their relative structural robustness as compared with free nucleic acids and circulating tumor cells that may have limited use in glioma diagnosis due to, e.g., poor passage over the blood–brain barrier and a glial cell origin. Apart from in GBM, considerable interest has recently been focused on EV-based approaches in pancreatic ductal adenocarcinoma (PDAC; ref. 36), including a recent study (37) that identified a signature of five plEV proteins to diagnose PDAC with an AUC of 0.84 and a sensitivity and specificity of 86% and 81%, respectively (n = 43). As a comparison, we find that plEVSDC1, i.e., a single marker, can discriminate between GBM and LGG with an AUC of 0.82 and a sensitivity and specificity of 71% and 80%, respectively (n = 82). This can be compared with MRI, which is the currently used standard modality in the diagnosis and prognosis of brain tumors. The discrimination between GBM and LGG is based primarily on gadolinium (Gd) enhancement (a marker of blood–brain-barrier disruption). However, it can still be a challenge in clinical practice as high-grade tumors may demonstrate no Gd enhancement, while low-grade tumors occasionally do. Accordingly, diagnostics by conventional MRI has shown a sensitivity and specificity of approximately 70%. Perfusion MRI and MR spectroscopy may improve the accuracy in discriminating between HGG and LGG tumors, but has not been implemented for preoperative, diagnostic purposes (38, 39).
Notably, most previous studies apply differential centrifugation to isolate EVs from plasma; even highly sensitive approaches using, e.g., microfluidic chip technology (14) and advanced nanoplasmonic sensing systems (37), were preceded by ultracentrifugation that may largely influence the integrity and contents of EVs for downstream analyses. Further, time-consuming centrifugation procedures are practically incompatible with future clinical protocols. SEC has recently emerged as the “gold standard” to yield purified EVs from complex biological samples (19, 22–24, 40–42). For experimental consistency and reproducibility, we used a standardized, commercially available column based on a sephadex resin of approximately 75-nm pore size. Using the same SEC procedure, others similarly found that the purest EV pool is contained up to fractions 9, and that the bulk of plasma proteins will elute in later fractions (19, 22). In fact, typical EV markers, such as CD9, CD63, and CD81, have previously been identified in plEVs with Western blot, EM, or flow cytometry, however, with LC-MS/MS only by a recent study using a similar SEC protocol combined with density cushion ultracentrifugation (19).
SDC1 has a well-established role as a growth factor and lipoprotein coreceptor during tumor development and is overexpressed in myeloma and in epithelial tumors of various origins, including in breast cancer (33, 43, 44). In line with our findings, SDC1 mRNA was nondetectable in normal brain while expressed in several GBM cell lines (30), and the survival rate was lower in patients with SDC1-positive as compared with SDC1-negative tumors (31). SDC1 expression has been found to be low in most normal tissues and, accordingly, systemic treatment with the indatuximab–ravtansine antibody–drug conjugate targeted at SDC1 was well tolerated in a I/IIa trial with myeloma patients (45). However, SDC1 is expressed in epithelial cells of, e.g., the skin and gastrointestinal tract (46) and can be shed as a soluble component into plasma (47). The increased secretion of plEVSDC1 by GBM tumors may relate to the direct role of SDC1 in EV biogenesis as part of a machinery involving syntenin and heparanase, i.e., proteins implicated in GBM pathogenesis (48, 49). Interestingly, we found other important promoters of EV biogenesis, RAB27A and ARF6 (50, 51), to be correlated with SDC1 in GBM (Fig. 3F), which may also contribute to the excessive release of plEVSDC1 from GBM tumors. Future studies should determine whether these independent observations are causally connected, i.e., if SDC1-mediated induction of the EV release pathway confers a more aggressive tumor phenotype in glioma. If so, SDC1 would appear as an interesting target for perturbation of the EV machinery and plEVSDC1 as an attractive biomarker of any such intervention.
Although the identification of plEVSDC1 provides important proof of principle, this study has limitations. The current study was designed as a feasibility study and needs to be corroborated in expanded clinical cohorts to also include other types of brain lesions that can be challenging to differentiate by MRI. Future studies should aim at identifying additional EV membrane markers that can be utilized for the enrichment of tumor-derived EVs. This may require a more comprehensive approach, as encouraged by our pilot studies using advanced LC-MS/MS during optimization of the SEC procedure. We found hundreds of proteins in plEVs that overlapped with GBM cell–derived EVs, several of which are established EV membrane markers.
In summary, we identify plEVSDC1 as a potential tool to facilitate noninvasive diagnosis of gliomas. Our results provide important support to the concept of EVs as a circulating “miniature of its cell of origin” and further motivate the future development of high-throughput, quantitative LC-MS/MS procedures to take full advantage of EVs as liquid biopsy biomarkers in cancer.
Disclosure of Potential Conflicts of Interest
J. Lehtiö reports receiving commercial research grants from AstraZeneca. No potential conflicts of interest were disclosed by the other authors.
Conception and design: V. Indira Chandran, M. Pernemalm, P.C. Sundgren, M. Belting
Development of methodology: V. Indira Chandran, C. Welinder, M. Pernemalm, J. Lehtiö
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C. Welinder, S. Offer, M. Pernemalm, S.M. Lund, S. Pedersen, J. Lehtiö, G. Marko-Varga, M.C. Johansson, E. Englund, P.C. Sundgren, M. Belting
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): V. Indira Chandran, C. Welinder, S. Offer, E. Freyhult, M. Pernemalm, S. Pedersen, J. Lehtiö, E. Englund, M. Belting
Writing, review, and/or revision of the manuscript: V. Indira Chandran, E. Freyhult, M. Pernemalm, S. Pedersen, J. Lehtiö, E. Englund, P.C. Sundgren, M. Belting
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): V. Indira Chandran, A.-S. Månsson, M. Pernemalm, G. Marko-Varga, M.C. Johansson, P.C. Sundgren
Study supervision: P.C. Sundgren, M. Belting
Other (managed patient samples and performed experiments): V. Indira Chandran
The authors thank all the patients and their families who participated in the study, and all the investigators and staff who contributed their time and effort to this study. They also thank Anna Weddig, research nurse, for help with the collection and handling of patient samples, and the Uppsala University Human Glioma Cell Culture resource (HGCC, www.hgcc.se) for sharing U3034 cells. This study was funded by grants from the Swedish Cancer Fund CAN 2017/664 and 2016/365 (M. Belting and P.C. Sundgren), the Swedish Research Council VR-MH 2014-3421 and K2011-52X-21737-01-3 (M. Belting and P.C. Sundgren), the Swedish Childhood Cancer Foundation PR2015-0078 (M. Belting), the Gunnar Nilsson Cancer Foundation, the Fru Berta Kamprad Foundations (M. Belting), the Skåne University Hospital donation funds (M. Belting), the Governmental funding of clinical research within the national health services, ALF (M. Belting and P.C. Sundgren), EU Horizon 2020 AiPBAND MSCA-ITN-ETN (M. Pernemalm and J. Lehtiö), and a donation by Viveca Jeppsson (M. Belting).
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.