The Quest for Non-Invasive Diagnosis: A Review of Liquid Biopsy in Glioblastoma

1. Introduction

Glioblastoma Multiforme (GBM) is the most common and aggressive primary brain tumour, with a median survival of around 15 months despite intensive treatments including surgery, radiation, and chemotherapy [1]. Its highly invasive nature, resistance to therapy, and recurrence make it one of the most challenging cancers to treat. GBM is also marked by substantial molecular heterogeneity, including gene mutations such as IDH-R132H, H3-K27M, and EGFR (EGFRvIII) amplification complicating treatment approaches and prognosis. Moreover, the blood-brain barrier (BBB) further limits the effectiveness of conventional therapies and monitoring methods [2,3,4].

Diagnosing GBM typically includes neuroimaging, followed by histopathological and molecular analysis for confirmation[5,6,7]. However, both imaging and tissue-based methods have limitations, such as distinguishing between relapse and pseudo progression, which are lesions related to treatment simulating recurrence of GBM; a continuous challenge [8]. These challenges underscore the need for alternative, non-invasive diagnostic tools, such as liquid biopsy, which has surfaced as a promising approach to detect early recurrence for GBM patients, principally through blood tests, urine, or cerebrospinal fluid (CSF), providing real-time insights into tumour dynamics and treatment response[9,10]. Liquid biopsies comprise the detection and quantification of tumour related content in biofluids that is released by tumours, such as circulating tumour cells (CTCs), circulating tumour DNA (ctDNA), circulating tumour microRNA (miRNA), extracellular vesicles (EVs) and nucleosomes or in response to tumours, all of which can provide critical insights into GBM progression and recurrence [11].

Being minimally or non-invasive, liquid biopsy offers a valuable alternative to invasive tissue biopsies for detecting early recurrence, monitoring minimal residual disease (MRD), and guiding treatment strategies in GBM. While the BBB hinders the amount of tumour-derived entities that may be found in body fluids, ongoing research aims to increase sensitivity and specificity of liquid biopsy tests, making them a promising tool for advancing clinical management of GBM. This review will explore recent advancements in liquid biopsy technologies and their potential for enhancing the diagnosis and treatment of GBM.

2. Circulating Biomarkers of Glioblastoma from Liquid Biopsies

2.1. Circulating Tumour Cells

CTCs are individual cells or cell clusters that are released from primary or metastatic solid tumours into bodily fluids. When disseminated into the bloodstream and other body fluids, they may contribute to metastasis and can be used as biomarkers for cancer detection, prognosis, and therapeutic monitoring[9,12]. The analysis of CTCs emerged as a potential tool for monitoring MRD in cancer patients[13,14,15]. Higher levels of CTCs are directly related to severity and spread of multiple cancers[16]. Key methods for CTC enumeration, which quantify CTC numbers include sensitive assays like CellSearch® (Menarini Silicon Biosystems, Italy), which detects as few as one CTC per 10 mL of blood. However, CellSearch® uses antibody-mediated capture, mainly targeting epithelial cell adhesion molecule (EpCAM) on CTC membrane surfaces using a positive selection method[17,18,19,20,21].

EpCAM is not expressed in GBM tumours due to their predominantly mesenchymal phenotype[12,22]. Hence, alternative techniques for isolation of CTCs in GBM patients have been explored (Table 1). One such method is negative selection, which enriches for CTCs by eliminating leukocytes from whole blood by means of antibodies against leukocyte antigens[9,19,29,32]. Several different techniques of CTC isolation have been used which are dependent on the CTCs physical properties[23,24]. These include density-gradient centrifugation[27,28], spiral microfluidic technology[30,31] and chip technology[21]. Although promising, it is important to highlight how rare CTC detection in the peripheral circulation of GBM patients actually is. Reported detection rates vary widely, with some studies identifying CTCs in as few as 20-40% of GBM patients, depending on the sensitivity of the detection method and sample volume and timing (20.6% - 77% of patients with detectable CTCs, Table 1). Along with the complexity of isolating CTCs from peripheral blood, this rarity underscores the need for highly sensitive and specific detection technologies to validate their clinical utility as reliable biomarkers. It also further complicates standardisation efforts and underlines the need for optimisation and harmonisation of isolation protocols.

CTC analysis methods for GBM, other than enumeration, include, molecular profiling (e.g., next generation sequencing (NGS), Reverse Transcriptase-PCR (RT-PCR)), immunophenotyping using GBM-specific markers (such as GFAP, Nestin, Sox2, EGFR, GLAST)[21,27,29,30,31], single-cell sequencing[27], and epigenetic profiling[25], with CSF-based analysis often offering higher sensitivity than blood[26,27,28]. These methods enable the detection of clinically-relevant alterations, such as IDH1 mutations and MGMT promoter hypermethylation, which are important for prognosis and for guiding targeted therapy decisions in GBM[19].

Monitoring CTC kinetics over time can provide insights into the effectiveness of treatment and the risk of disease recurrence. Changes in CTC count, phenotype or genetic profile can act as early indicators of therapeutic response of tumour progression[21,27,28,29,31,32] (Table 1).

These studies demonstrate the potential of CTCs as a tool for cancer diagnosis and prognosis, however, there are still challenges that need to be addressed: (i) Low abundance: CTCs are present in very low numbers in the bloodstream being 1 CTC in 10⁹ normal blood cells per 10 mL of blood, making their isolation and analysis challenging[9,18]. Low abundance can lead to false-negative results and reduce the sensitivity of CTC-based assays. (ii) Heterogeneity: CTCs show intra- and inter-patient heterogeneity and may differ in morphology, phenotype, and genetic profile. This heterogeneity complicates the identification and characterisation of CTCs, and it may require the use of multiple markers or assays to detect different subsets of CTCs[35,36]. (iii) Technical variability: The isolation and analysis of CTCs require specialised equipment and protocols, and it can affect the accuracy and reproducibility of CTC-based assays. The choice of isolation and analysis method, as well as the handling and storage of blood samples, can all impact the results of CTC assays[37,38]. (iv) Lack of standardisation: There is currently no standardised protocol for the isolation and analysis of CTCs, and different studies may use different methods or markers to detect CTCs. This lack of standardisation can make it difficult to compare results across studies and to establish the clinical utility of CTC-based assays[39]. (v) Cost and accessibility: CTC-based assays can be costly and require specialised equipment and expertise, which may limit their availability and accessibility in a clinical setting[40].

Overcoming these barriers will require continued innovation in sensitive and specific CTC isolation and analysis techniques, along with greater collaboration between researchers and clinicians in establishing the clinical utility of this approach. By integrating CTC-based assays with other clinical and pathological parameters, and rigorously validating these methods in clinical trials, the clinical utility and potential for earlier accurate cancer detection and monitoring tumour progression may be determined.

2.2. Circulating Tumour DNA

Cell-free DNA (cfDNA) are short DNA fragments of 140-170 base pairs (bp) present in body fluids. In cancer patients, a portion of cfDNA originates from tumour cells following apoptosis or necrosis, referred to as circulating tumour DNA (ctDNA)[9,10,18,41,42]. Representing a promising tool for non-invasive cancer diagnosis, particularly in GBM, ctDNA consists of shorter DNA fragments (< 145 bp) that carry the genetic and epigenetic signatures of the originating tumour[12,18]. A number of studies have validated that ctDNA reflects the genetic profile of the original tumour, showing uniformity with its mutational profile and that of the corresponding tumour tissue across various cancers, including GBM[9,43,44,45], thereby offering insights into tumour genetics and treatment response (see Table 2). Key detection methods include assessing ctDNA levels and known point mutations using polymerase chain reaction (PCR)-based method such as digital droplet PCR (ddPCR), and next-generation sequencing (NGS) or whole genome sequencing (WGS), method that detects novel or unknown mutations[46,47,48]. ddPCR is particularly sensitive for identifying low-frequency mutations such as IDH1, TERT promoter mutations, and EGFRvIII in plasma samples[48]. NGS enables comprehensive genomic profiling, detecting a broad spectrum of mutations and copy number alterations[49]. Methylation-specific PCR targets epigenetic modifications, like MGMT promoter methylation, which are relevant for GBM prognosis and therapy selection[47].

Detecting ctDNA in GBM is challenging due to tumour’s location and the BBB, which decreases yield of ctDNA in peripheral circulation. Despite these obstacles, significant work has been done towards ctDNA detection by employing highly sensitive molecular technologies as described in Table 2. It is noteworthy to mention the baseline level of cfDNA is greater in serum compared to plasma, which may be in result of contamination with DNA that is released from lysed immune cells, thus plasma samples are preferred for studying ctDNA[46,47,48]. While plasma ctDNA alone may not yet be sufficiently sensitive for reliable GBM diagnosis or monitoring, detectable levels offer promise for non-invasive molecular profiling and could help identify actionable mutations and resistance mechanisms, potentially improving treatment strategies. Studies have shown that ctDNA can be detected in a subset of GBM patients, though plasma detection rates are generally lower than in other cancers, likely due to the BBB limiting ctDNA release into the bloodstream[50]. Piccioni et al. found ctDNA mutations in the blood of 50% of brain tumour patients, with 55% of these being GBM cases[51]. However, dependent on patient cohort and cfDNA extraction method 75.0% sensitivity (95% CI: 64.1%–84.0%) and 88.7% specificity (95% CI: 77.0%–95.7%) has been achieved for detection of the IDH1 mutation R132H, and negativity for this mutation is a key classifier of GBM[52,53].

EGFRvIII ctDNA was detectable in plasma and can serve as a dynamic biomarker for preoperative and postoperative assessment, as well as for monitoring disease recurrence in GBM patients[54]. Zill et al. used next-generation sequencing of plasma ctDNA in 107 GBM patients to identify actionable mutations, supporting the use of ctDNA sequencing for personalised cancer treatment[55]. Liu et al., and Miller et al., further demonstrated that WGS of CSF ctDNA achieves high detection rates and can identify clinically relevant mutations, reinforcing its utility as a minimally invasive biomarker for GBM monitoring[56,57].

Furthermore, studies have demonstrated that CSF is a more reliable source for ctDNA in GBM patients compared to plasma, overcoming the limitations imposed by the BBB in regard to ctDNA release into the bloodstream[58,59,60]. For instance, ctDNA was detected in 61% of CSF samples compared to 37% in serum of 19 GBM patients using methylation-specific PCR[61], while nested PCR allowed for detection in 92% of CSF samples compared to only 8% in plasma samples of 38 GBM patients [62]. Another study using WGS on CSF samples of 11 GBM patients demonstrated a 100% ctDNA detection rate[63]. These findings highlight the superior sensitivity of CSF-based liquid biopsies, likely due to the proximity of CSF to the brain tumour and the reduced interference from background cfDNA found in blood[9,12,18,64]. However, CSF collection via lumbar puncture is invasive and may not be suitable for all patients[18]. As alternatives, less invasive fluids like nasal secretions or saliva may provide an alternative to CSF, offering a less invasive means to access brain-derived biomarkers but needed further investigation.

Urine has also emerged as a non-invasive source of ctDNA for GBM diagnosis, offering the advantage of easy and repeatable collection throughout treatment[10]. Following renal filtration, ctDNA can be detected as transrenal DNA (trDNA), enabling molecular diagnostics through urinary DNA analysis[10,65]. Several urinary biomarkers—including matrix metalloproteinases (MMP-2, MMP-9), neutrophil gelatinase-associated lipocalin (NGAL), and vascular endothelial growth factor (VEGF) have demonstrated high sensitivity (95.2%), specificity (95.7%), and accuracy (92.5%) in detecting primary brain tumours, including GBM (see Table 2). Notably, elevated levels of these urinary markers in GBM patients have been shown to correlate with their expression in tumour tissue and, in CSF (for MMP-9 expression)[65,66,67,68]. In addition, urine-based detection of ctDNA epigenetic alterations, such as DNA methylation, offers further potential. Methylation-specific PCR enables the quantification of tumour-specific methylation, such as O6-methylguanine-DNA methyltransferase (MGMT) promoter methylation[69,70], a well-established predictor of GBM patient response to alkylating agents like temozolomide (TMZ)[71,72,73].

In conclusion, ctDNA analysis has shown significant promise as a tool for detecting and monitoring genetic alterations in brain cancers, although some challenges still need to be addressed. These include variability in ctDNA levels depending on the stage and anatomical location of cancer, as well as its rapid degradation in circulation and sensitivity to pre-analytical factors, which can complicate collection and analysis. Overall, detection of ctDNA in CSF and plasma provides valuable insights into tumour evolution, resistance mechanisms, and potential therapeutic targets, offering a non-invasive approach to guide personalised treatment for CNS cancers. Further advancements in ctDNA isolation, sequencing methods, and clinical validation will be crucial to unlock its full potential for early diagnosis, treatment monitoring, and personalised cancer management.

2.3. Nucleosomes

Chromatin structure was first described in 1974 as repeating nucleosome units. Each nucleosome consists of an octamer of two copies, each of the core histones H2A, H2B, H3 and H4, wrapped by 146-147 base pairs of DNA[80,81]. Nucleosomes are generated by DNA associated with proteins, including histone H1. Histones have a globular domain for histone-histone interactions and flexible, positively charged tails (20-35 amino acids) that undergo post-translational modifications (PTMs)[82]. As the basic units of chromatin, nucleosomes regulate key nuclear processes like transcription, replication and repair, primarily through PTMs[83]. These modifications influence gene expression on an additional level to DNA methylation, and activating oncogenes or silencing tumour suppressor genes may contribute to cancer development and progression[83,84]. Histone PTMs include methylation, acetylation, phosphorylation, ubiquitination, sumoylation, glycosylation, homocysteinylation, and crotonylation[85]. Methylation may activate or repress gene expression depending on the site and degree[83], while acetylation generally promotes transcription by neutralising lysine’s positive charge, reducing chromatin compaction and enhancing accessibility for transcription factors[85]. The interplay of these PTMs forms a “histone code” read by cellular proteins to control transcription, replication, and repair, with core histone variants adding further complexity[85,86].

Nucleosomes and their PTMs can be detected in plasma or serum using immunoassays like chemiluminescence immunoassay (ChLIA) or enzyme-linked immunosorbent assay (ELISA) (Table 3). These assays detect specific biomarkers from complex matrices, and can be applied on automated platforms which would allow faster and more reproducible results[87]. Cancer cell death, including that of GBM cells results in the release of nucleosomes into the bloodstream, which are transported as mono- or oligo- nucleosomes bound to ctDNA.

Cancer patients have a greater count of circulating nucleosomes due to the increase in cellular turnover in comparison to healthy controls and due to the cytotoxic effect of treatments which lead to cell death[83]. The high levels of nucleosomes are not explicit to cancer or GBM per se, since it is also linked to trauma, stroke and sepsis which limits its clinical use as a unique biomarker for GBM detection. The latest studies have however, reported that the PTMs found on circulating nucleosomes may have cancer specificity and are thus be investigated as biomarkers[88].

Enzymes that regulate PTMs are often dysregulated in GBM. Histone deacetylases and demethylases like lysine-specific histone demethylase 1, can alter the epigenetic status of brain cells, influencing cancer development and progression [88,89]. In paediatric high-grade gliomas, recurrent histone mutations are common. The H3K27M mutation, a lysine-to-methionine substitution at position 27 of H3.1 or H3.3 histone genes, is associated with childhood diffuse intrinsic pontine glioma, disrupting polycomb repressive complex 2 function, leading to global H3K27 methylation loss and altered gene expression[90]. H3G34R/V mutations, involving glycine 34 substitutions in H3.3 are seen in gliomas of the cerebral hemispheres in both children and adults causing redistribution of the activating H3K36 methylation mark and transcriptional dysregulation[88,90].

Another epigenetic marker dysregulated in GBM is histone 3 lysine 4 (H3K4), which is decreased in severe GBM cases, leading to gene repression[89]. Acetylation of histone 3 at lysine 18 (H3K18Ac), is another common PTMs with dysregulated expression seen in a number of cancers including GBM, where lower levels identify better prognosis for the patient[91]. Accordingly, a profile of PTMs of circulating nucleosome linked histone modifications has potential to be studied in GBM patients which would be in addition to MRI, a guide to diagnosing and monitoring tumour progression in GBM patients[9].

Nucleosomes are generally less useful for cancer diagnosis due to elevated levels in benign diseases. However, in gastrointestinal cancers, nucleosome levels correlate with tumour stage and metastasis. Nucleosomes are more valuable for therapy monitoring, as their levels decrease in remission and increase with disease progression during chemotherapy and radiotherapy, a pattern observed in lung, pancreatic, colorectal cancers, and haematological malignancies [92,93].In GBM, circulating nucleosomes in serum and CSF have shown potential for diagnosis and prognosis. Patients developing cerebral oedema post-surgery exhibited a ~200-fold rise in CSF nucleosomes. Monitoring these levels may help detect complications and track tumour progression and treatment response[94]. However, further research is needed to refine their specificity and clinical application, especially in distinguishing cancer from other conditions.

Collectively, circulating nucleosomes and their PTMs present a promising avenue for GBM detection and monitoring. While total nucleosome levels in plasma and CSF lack disease specificity, unique histone modifications, such as H3K27M and H3K18Ac, provide greater diagnostic potential. Emerging technologies, including immunoassays and single-molecule detection methods, have enabled precise profiling of nucleosome-bound epigenetic markers, distinguishing GBM patients from healthy individuals. These findings suggest that nucleosome-based liquid biopsies could serve as a valuable adjunct to MRI and conventional diagnostics, particularly in monitoring disease status and therapy response. Additionally, nucleosome levels may help identify postoperative complications, such as cerebral oedema, offering further insight into tumour progression and treatment outcomes. However, further validation in larger clinical studies is necessary to refine their specificity and clinical application, especially in differentiating cancer from other conditions.

2.4. Circulating Tumour microRNA

Circulating tumour microRNA (miRNA), is the most common small non-coding RNA, typically 21-23 nucleotides long, and regulates up to 30% of protein-coding genes[96,97]. It plays a key role in cancer by downregulating tumour suppressor genes. miRNAs are highly stable and detectable in bodily fluids such as blood, urine, and CSF, and are also present in extracellular vesicles (EVs)[12,98,99,100].

Altered expression of specific miRNAs have been found to distinguish between GBM patients and lower-grade glioma patients (Table 4). For instance, downregulation of miR-125b, miR-16, miR-497, miR-128, miR-342-3p, miR-205 and upregulation of miR-210, miR-182, miR-20a-5p, miR-454-3p, miR-106a-5p, miR-181b-5p have been linked to GBM pathophysiology[101,102,103,104,105,106,107,108]. Thus, miRNA dysregulation in GBM involves both upregulation of oncogenic miRNAs (“oncomirs”) and downregulation of tumour-suppressive miRNAs, each playing distinct roles in tumour biology.

Most miRNAs which show an increase in their expression level in GBM are oncogenic and are known as “oncomirs”[109]. Oncomirs usually promote tumour growth by inhibiting tumour suppressor genes and genes linked to cell differentiation or apoptosis. The first oncomir which showed significantly increased levels in GBM tissue samples compared to normal tissue was miR21. It can target several tumour suppressors such as PTEN[110]. The levels of miR-21 in GBM tissue and plasma samples was higher in grade II and III GBM samples compared to controls, yet there was no significant change between pre-surgery and post-surgery sample measures, while miR21 level was decreased after chemo irradiation therapy [101].

Several miRNAs function as tumour suppressors and are commonly downregulated in GBM patients[111]. Tumour-suppressive miRNA such as Let-7, which targets oncogenes like K-RAS and MYC inhibit GBM cell proliferation by interfering with histone methyltransferase EZH2[112,113,114]. miR-128 and miR-342-3p, significantly reduced in the plasma of GBM patients, have been shown to return to normal levels after surgery and chemo-radiation, indicating the patient’s response to treatment[102]. These miRNAs are glioma-specific, and their loss or downregulation may contribute to tumour progression including malignant transformation of meningioma and pituitary adenomas into GBM [115,116]. A meta-analysis demonstrated that serum miR-125b, markedly reduced in GBM patients, shows a 3.5-fold higher detection frequency compared to healthy controls, highlighting its potential as a screening biomarker[117]. Similarly, miR-205, another tumour suppressor, was significantly decreased in GBM patients, increased post-surgery, and declined again upon recurrence, suggesting its value as a dynamic biomarker[107]. Accordingly, the dysregulated expression of miRNA, shapes the prediction of cancer, early diagnosis, prognosis and histological classification[9,98]

Other studies have advocated that the BBB is not permeable for some miRNAs which increases the value of miRNA profiling of the CSF samples. As miR-10b is greatly overexpressed in GBM but absent in normal brain, it can thus be detected in CSF of majority of GBM patients (89%) but not in the CSF of healthy controls. Considering extracranial tissues express miR-10b, its absence in CSF of healthy controls designated that it may not pass the BBB under non-neoplastic conditions[98,118].

Several studies have identified specific miRNAs that are dysregulated in GBM patients, with both upregulated and downregulated miRNAs linked to tumour progression and prognosis. miRNAs such as miR-21, miR-128, and miR-342-3p have been highlighted for their potential as biomarkers for early GBM detection, offering high sensitivity and specificity, particularly in blood and urinary samples. These clinical data signify that the dysregulation of miRNAs, whether through upregulation or downregulation, can be used to diagnose or monitor GBM patients. The circulation biomarker: miRNA should serve as a non-invasive method for screening patients suspecting GBM. Though the results are thought-provoking, some limitations do exist, like the small size of regimens and the lack of a standard method for blood collection, RNA extraction and sequencing. Of note is that specificity of miRNA detection may be lower than that of ctDNA[9], but it may be considered complementary to ctDNA detection. Despite challenges, miRNA-based liquid biopsy offers a promising, less-invasive alternative to traditional tissue biopsies for GBM diagnosis, prognosis, and treatment monitoring.

2.5. Extracellular Vesicles

Extracellular vesicles (EVs), including exosomes and microvesicles (MVs) are small membrane enclosed particles released by normal and tumour cells. They serve as carriers of macromolecules in liquid biopsies of brain tumour patients[18]. EVs are heterogeneous in size, quantity and origin of molecular content and biological activity. Two main groups of EVs exist, which differ in origin and size. MVs (50 to 500 nm) bud from the cell membrane, while exosomes (50 to 150 nm) originate from the endosomal system[125,126]. EVs protect their cargo; mRNAs, miRNAs, lipids or proteins, which are specific to the cell origin[127] from enzymatic degradation via their lipid bilayer, enabling them to cross the BBB[125].

The significance of EVs is emphasised by the fact that their transcriptomic and proteomic profile is specific to the cell of origin and can differ in response to diverse stimuli[10]. Their imperative role in intercellular communications allows them to alter the phenotype of recipient cells by transferring genetic information and proteins[12,128]. These molecules often reflect malignant processes, underlining their importance as liquid biopsy tools in cancer, including GBM[127]. EV-mediated crosstalk has been demonstrated between GBM and its tumour microenvironment, fostering tumour progression[129]. Exosomes from hypoxic GBM cells overexpress VEGF-A, disrupting the BBB by downregulating occludin and claudin-5[130].

EV isolation typically involves immunoaffinity capture or differential centrifugation gradients[9]. Exosomes are characterised by transmission electron microscopy (TEM), nanoparticle-tracking analysis (NTA) and surface markers such as CD63, CD81, CD9, CD37, CD53, CD82, ICAM-1 and integrins, detectable by flow cytometry or Western blot[127].

Exosome-based liquid biopsies show promise in cancer diagnostics, with tests like ExoDx™ Lung (ALK) and ExoDx Prostate IntelliScore (EPI) already approved. For GBM, research is ongoing[131]. Emerging biomarkers include exosomal circular RNAs (circRNAs), EGFRvIII RNA for monitoring CAR-T cell therapy response, and MGMT methylation for predicting treatment outcomes. Despite promising findings (see Table 5), no exosome-based GBM test has yet reached clinical practice, pending further validation in larger cohorts [131,132,133,134,135].

Studies indicate that EVs from GBM stem cells exhibit increased adhesion-related proteins after TMZ treatment, potentially promoting tumour progression[136]. TMZ resistance may be associated with EVs carrying MGMT mRNA, as a potential marker of resistance, alongside adhesion proteins (e.g., TGM2, CD44 and CD133), and stemness markers like NESTIN[137,138,139]. These EV-associated molecules may serve as biomarkers to monitor drug resistance and treatment failure[10].

Overall, EVs, particularly exosomes, are promising biomarkers for GBM detection, monitoring, and prognosis. Exosomal RNA, proteins, and lipids include disease-specific molecular cargo, such as EGFRvIII RNA and syndecan-1 with high diagnostic potential. EV-derived molecular cargo may reflect or potentially even contribute to treatment resistance and tumour progression, though their mechanistic role remains under investigation. Although exosome-based liquid biopsy presents a minimally invasive alternative to traditional tissue biopsies, larger patient cohorts need to be analysed and standardised isolation and characterisation techniques are needed to validate their utility in GBM. Despite these limitations, ongoing research suggests that EV-based liquid biopsies could revolutionise GBM management, offering real-time insights into disease progression and therapeutic response.

3. Application of Machine Learning and Artificial Intelligence

The diagnosis of glioblastoma using signals derived from liquid biopsy samples remains particularly challenging, traditional statistical approaches frequently fall short in detecting the nonlinear patterns and subtle molecular variations that characterize GBM biology, increasing the risk of misdiagnosis or missed detection when relying solely on liquid biopsy data. To address these challenges, artificial intelligence (AI) and machine learning (ML) approaches are increasingly integrated into GBM liquid biopsy workflows. Novel ML algorithms can extract meaningful insights from noisy and heterogeneous datasets, enabling earlier and more accurate detection of GBM tumours, even when biomarker levels are extremely low in blood or CSF. Various liquid biopsy-based signatures have shown potential for tumour detection, including plasma denaturation profiles, CSF proteomic signatures, and serum miRNA signatures120, 160. In omics analysis for GBM, ML techniques uncover complex molecular signatures linked to tumour subtype classification, disease progression, and therapeutic response[163,164]. This allows the development of robust predictive models that outperform traditional biomarker-based approaches, advancing precision diagnostics and personalized treatment strategies for glioblastoma patients. Once trained, these models can be applied in clinical practice[163,164].

Fragmentation patterns and personalised cfDNA sequencing in urine and plasma have differentiated glioma patients from healthy individuals[165]. Tumour-educated platelets (TEPs), analysed with swarm algorithms, have identified spliced RNA biomarkers to distinguish false positives from true disease progression[166].

ML has also identified tumour-specific DNA methylation markers in blood and tumour-associated MRI features[167,168]. In several studies, ML algorithm identified MRI based tumour-associated features, while serum spectroscopy combined with MRI detected spectra variations between healthy and tumour-affected patients. Additionally, ML has been applied to investigate the relationship between brain tumour volume and liquid biopsy test performance. In a cohort of 177 patients (90 patients with high-grade glioma (GBM or anaplastic astrocytoma), or low-grade glioma (astrocytoma, oligoastrocytoma and oligodendroglioma) spectroscopic liquid biopsy approach detected small and low-grade gliomas, supporting early diagnosis[169,170].

AI enhances liquid biopsy applications and diagnostic performance by enabling rapid analysis of cancer-related circulating proteins and nucleic acids and enhancing imaging from subjective interpretation to quantitative analysis.[171,172,173] Multimodal AI models integrate proteomic and genomic data from liquid biopsies and quantitative imaging to create a more comprehensive diagnostic pathway[174,175]. This helps distinguish benign from malignant lesions, reduces unnecessary follow-ups and guides therapeutic decisions.

Despite its potential, the application of AI and ML in liquid biopsy and cancer diagnostics faces several challenges. One major hurdle is the need for large, high-quality datasets to train robust models, as variability in sample collection, processing, and sequencing methods introduces bias[176]. Additionally, the interpretability of AI-driven models remains a concern, as complex algorithms often function as "black boxes," making it difficult for clinicians to understand the reasoning behind predictions[164,174]. Regulatory approval and clinical validation also pose significant barriers, requiring extensive trials to ensure reliability and accuracy. Finally, integrating AI into existing healthcare systems demands substantial computational resources, infrastructure, and clinician training, which may limit widespread adoption[168,174,176,177,178].

4. Conclusions

In the coming years, clinical studies are expected to clarify which liquid biopsy techniques provide the most value for diagnosing, prognosing, and treating primary brain tumours, potentially surpassing traditional imaging and tissue-based methods. GBM liquid biopsy research is rapidly gaining traction. Analysis of body fluids, such as blood, or CSF, from GBM patients reveal various tumour-derived components, including CTCs, ctDNA, miRNAs, EVs, nucleosomes, and metabolites. These biomarkers may serve as alternatives to tissue samples and support genotype-guided therapies and personalised GBM management. They also show promise for monitoring tumour progression, treatment response, and guiding therapeutic choices. However, accessing tumour-derived material is challenging due to the BBB, which limits the release of biomarkers into the bloodstream—currently the most studied liquid biopsy. Consequently, tumour-derived nucleic acids in serum or plasma often occur at low levels, hindering routine clinical use. Improving sensitivity will require optimising sample volumes, technological platforms, and the use of artificial intelligence. Combining the measurement of multiple different liquid biopsy-based markers from the same biopsy will likely improve overall sensitivity of single modality testing but this idea needs to be further validated in the future.