Glioblastoma biomarkers in urinary extracellular vesicles reveal the potential for a ‘liquid gold’ biopsy

Glioblastoma (GBM), IDH-wildtype, is the most common and lethal primary brain tumor in adults. Despite standard therapy, nearly all GBMs recur, and monitoring disease activity is hampered by radiologic ambiguities such as pseudoprogression. There is a pressing need for sensitive, specific, and minimally invasive biomarkers to track tumour burden, treatment response, and recurrence to guide timely clinical decisions. Extracellular vesicles (EVs) cross the blood–brain barrier and carry biomolecules reflective of parent tumor cell states; the urinary system is a major route for clearing circulating EVs. The study investigates whether urinary EVs (uEVs) can serve as a non-invasive ‘liquid gold’ biopsy for GBM by identifying proteomic signatures associated with diagnosis, tumour burden changes after surgery, and recurrence.

Prior work has demonstrated EV-associated GBM biomarkers in neurosurgical fluids, cerebrospinal fluid (CSF), and blood, but these matrices pose limitations due to invasiveness (CSF, surgical fluids) and complexity (blood). Urinary EV biomarker studies have largely focused on renal diseases, but uEV cargo has distinguished systemic and neurological conditions, including breast cancer and neurodegenerative diseases (Parkinson’s, Alzheimer’s, Huntington’s). In GBM, urinary biomarkers such as matrix metalloproteinases were previously elevated and decreased with treatment, and proteomics of urine at diagnosis versus post-resection identified hundreds of proteins linked to autophagy and angiogenesis. Comprehensive proteomic characterization of EVs from urine is challenged by abundant uromodulin (Tamm–Horsfall protein) co-isolation, potentially masking low-abundance biomarkers in shotgun LC-MS/MS. Data-independent acquisition (DIA) MS strategies with spectral libraries enable sensitive, reproducible, and accurate quantification of low-abundance peptides, motivating their use for uEV proteome profiling in GBM.

Study design and cohorts: Fifty urine specimens were collected from 24 catheterised GBM patients at defined clinical timepoints: immediately prior to primary surgery (Pre-Op, n = 17), within 30 days post gross total resection (Post-Op, n = 9 matched cases), and prior to recurrence surgery (REC, n = 7). Urine from age- and gender-matched healthy controls (HC, n = 14) was collected mid-stream after the morning peak. Nearly all GBM patients (23/24) had normal renal function. Ethics approvals and biobanking procedures were followed. EV isolation and characterization: Urine (approximately 15–100 ml) underwent differential centrifugation and ultracentrifugation to remove cells/debris, pellet large EVs, and isolate small EVs (<300 nm). Reduction with TCEP/Tris was applied to mitigate uromodulin networks and improve EV recovery. EVs were characterized per ISEV guidelines using nanoparticle tracking analysis (particle size/concentration), cryo-transmission electron microscopy (vesicular morphology), and LC-MS/MS to verify canonical EV markers (including coverage of top EV proteins from Vesiclepedia). Proteome preparation and LC-MS/MS: EV pellets were lysed in RapiGest/TEAB with heat and sonication, quantified (Qubit), reduced/alkylated as per protocol, digested with sequencing-grade trypsin, and desalted (HLB cartridges). Peptides were analysed on a Q-Exactive HF/X Orbitrap in DIA mode with nano-LC separation (reverse-phase gradients). DIA employed variable-width windows across a broad m/z range. Instrument performance was validated using BSA standards and monitored throughout runs. Spectral library and data processing: A comprehensive GBM spectral library derived from patient GBM cells, GBM-derived EVs, and related tissues was used for targeted data extraction in Spectronaut/DIA-NN. Identification thresholds included peptide length 7–30 aa, precursor m/z 310–1800, 1% FDR at peptide/protein levels. Normalized protein abundance matrices were generated. Technical reproducibility was assessed via replicate injections (R² = 0.923–0.981). Filtering, statistics, and biomarker modeling: Proteins present in at least 80% of samples across cohorts were retained (903 proteins) for differential analysis. Abundances were log-transformed, missing values imputed from a normal distribution, and quantile normalized. Group comparisons used t-tests with BH-adjustment. Criteria for candidate biomarkers: fold change (FC) ≥ 2 and adjusted p ≤ 0.05. ROC analyses determined AUC; proteins with AUC > 0.9 were prioritized. Stepwise logistic regression identified minimal panels with optimal diagnostic (GBM vs HC) and progression (Post-Op vs REC) performance. PCA visualized group separation. Functional annotation: Differential proteins were analysed by Ingenuity Pathway Analysis (IPA) and annotated to KEGG/Reactome via DAVID. Enrichment focused on EV-associated compartments and cancer-relevant pathways, including TRiC/CCT chaperonin networks.

- Proteome depth: 6857 proteins were confidently identified in uEVs (p ≤ 0.01), including 94 EV marker proteins; 1667–2001 proteins were detected in >80% of samples within cohorts, and 903 proteins were present in ≥80% of all samples and carried forward for differential analyses. - Diagnostic signature (GBM vs HC): Among 1545 proteins common to Pre-Op GBM and HC, 290 showed nominal changes (p ≤ 0.05) and 31 were significant at BH-adjusted p ≤ 0.05. Several ribosomal subunits (e.g., RPL8, RPL23, RPL7A; and S-subunits in figures) and keratin KRT19 were elevated in GBM uEVs. A stepwise logistic model using five proteins (including KRT19, RPL8, RPL23, RPL7A) achieved AUC = 0.958 for distinguishing GBM from HC. PCA based on high-performance proteins improved group separation. - Tumour burden (Pre-Op vs Post-Op): Of 96 commonly detected proteins, 72 changed significantly after gross total resection (p ≤ 0.05; several at adjusted p ≤ 0.05), defining a ‘tumour burden’ signature. Twenty of these also differed between Pre-Op GBM and HC with concordant direction of change, indicating reversibility post-resection. - Recurrence (Post-Op vs REC): Sixty-four proteins differed significantly at recurrence (p ≤ 0.05), including multiple TRiC/CCT chaperonin subunits (e.g., TCP1/CCT1, CCT2–CCT4, CCT6, CCT7, CCT8). Network analysis revealed interactions with cancer-relevant proteins (ANXA1/2, GNAS, HSP90AA1/AB1, VCP, YWHAE). Enriched pathways included PI3K–Akt signaling and ‘pathways in cancer’; Reactome highlighted ‘Folding of actin by CCT/TRiC’ with very high enrichment. - Recurrence biomarkers and performance: Three proteins (ITM2B, GGH, GRN) showed consistent longitudinal trends across Pre-Op, Post-Op, and REC, with strong individual ROC performance (AUC > 0.92). A logistic regression panel combining GRN and ITM2B yielded AUC = 1.0 ± 0.2 for classifying recurrence (Post-Op vs REC), with clear PCA separation. - Technical robustness: Replicate DIA-MS injections showed high repeatability/reproducibility (R² 0.923–0.981).

This study demonstrates that urinary extracellular vesicles are a viable, minimally invasive source of GBM biomarkers that reflect tumor biology and clinical status. The identification of robust diagnostic signals in uEVs, including elevated ribosomal proteins and KRT19, supports their potential for non-invasive GBM detection. Significant post-resection decreases across a set of proteins define a ‘tumour burden’ signature, indicating that uEV proteomes can capture dynamic changes following surgery. The recurrence-associated increases in TRiC/CCT chaperonin subunits and their cancer pathway enrichment align with known GBM oncogenic signaling (e.g., PI3K–Akt, Wnt/β-catenin), suggesting biological plausibility. Individually strong and panel-based biomarker performances (diagnostic AUC ~0.96; recurrence panel AUC ~1.0) indicate that uEV profiling may improve surveillance, enabling earlier detection of recurrence and potentially helping to differentiate true progression from treatment-related changes. The findings extend GBM EV biomarker observations from invasive or complex matrices (CSF, surgical fluid, blood) to urine, facilitating broader clinical adoption.

Using DIA-LC-MS/MS with a GBM-focused spectral library, the study provides the first in-depth characterization of GBM patient urinary EV proteomes and identifies candidate biomarker panels for diagnosis, tumour burden monitoring, and recurrence detection. Key contributions include delineation of uEV protein signatures specific to GBM, demonstration of dynamic proteomic shifts across clinical timepoints, and high-performing biomarker panels (e.g., KRT19/ribosomal proteins for diagnosis; GRN+ITM2B for recurrence). Future work should validate these candidates in larger, longitudinal cohorts with comprehensive clinical and radiologic correlation (including pseudoprogression), optimize EV isolation/purity, and expand to multi-omic EV cargo (e.g., miRNAs) to enhance sensitivity and specificity.

- Sample size and cohort design: Modest number of patients and specimens; recurrence cohort is small. Longitudinal sampling frequency may limit temporal resolution. - Post-operative timing and confounders: Post-Op sampling may not fully reflect steady-state tumour burden reduction; surgical stress, anaesthesia, blood–brain barrier disruption, and perioperative factors can influence uEV profiles. - EV isolation and purity: Differential ultracentrifugation can co-isolate contaminants; uromodulin (THP) remained abundant despite reduction strategies, potentially masking low-abundance biomarkers. Alternative methods (e.g., SEC) may improve purity but reduce yield. - Generalizability across biofluids: EV biomarker levels can differ across fluids (e.g., PSAP discrepancies in urine vs plasma), complicating cross-matrix thresholds. - Analytical considerations: Library-based DIA enhances specificity but may miss proteins absent from the library; library-free methods have different trade-offs. Some reported identifiers show OCR-related ambiguities that warrant orthogonal validation.