Systematic review of extracellular vesicle-based liquid biopsy for leptomeningeal metastasis: current evidence and future directions

Abstract

Leptomeningeal Metastasis (LM) is a catastrophic complication of systemic malignancies with poor prognosis and significant diagnostic challenges. Extracellular vesicle (EV)-based liquid biopsy is an emerging technology that may improve the detection, prognosis, and therapeutic monitoring for LM. This systematic review aims to critically appraise and synthesize existing evidence on the use of EV-derived biomarkers from cerebrospinal fluid (CSF) or serum for the diagnosis, prognosis, and therapeutic monitoring of LM. A systematic search of PubMed, ScienceDirect, Semantic Scholar, and Google Scholar was conducted up to January 2025. Eligible studies evaluated EV-associated molecular biomarkers (microRNAs or proteins) in LM patients using CSF or serum samples. Data extraction included study design, sample source, EV isolation method, biomarker type, validation method, and diagnostic/prognostic relevance. Risk of bias was assessed using QUADAS-2. Ten studies met inclusion criteria, encompassing EV miRNA and proteomic profiling in LM, primarily from non-small cell lung cancer (NSCLC) and medulloblastoma. miR-21 consistently correlated with poor prognosis and therapy resistance. Novel candidates such as miR-183-5p demonstrated high diagnostic accuracy (AUC 1.0 in a discovery cohort), targeting PTEN pathways. EV-derived proteins including fibronectin-1 (FN1), transketolase (TKT), and complement C2 also showed strong discriminative potential. Heterogeneity in EV isolation techniques and small sample sizes limited meta-analysis feasibility. EV-based liquid biopsy holds cautious promise as a minimally invasive tool for LM diagnosis, prognosis, and therapeutic monitoring. Larger, multicenter studies with standardized protocols are required to validate these findings and facilitate clinical translation.

Introduction

Leptomeningeal metastasis (LM), also commonly known as “leptomeningeal carcinomatosis” or “neoplastic meningitis,” is a catastrophic complication of systemic malignancies, wherein neoplastic cells penetrate the cerebrospinal fluid (CSF) and seed along the meningeal surfaces of the brain and spinal cord, resulting in multifocal neurologic impairment, accelerated disease, and poor prognosis [1, 2]. Although a rare metastatic complication, LM has increased in incidence in recent years due to improvements in the overall survival of systematic cancers [1]. LM occurs in around 5–10% patients with solid tumors, among the most common being non-small cell lung carcinoma (NSCLC) and breast cancer, followed by melanoma and other primary central nervous system (CNS) tumors; hematologic cancers are less frequent causes of LM [1, 3]. However, despite advances in management for systemic cancers, prognosis for patients who develop LM is unfortunately grim. The median overall survival after diagnosis ranges between 2–4 months, even with intrathecal chemotherapy, targeted therapy, and radiation [4] Prognosis can indeed vary depending on the primary tumor type, its molecular and immunological profile, and extent of CNS involvement [4].

Diagnosing LM itself is challenging and begins with a high index of clinical suspicion. Patients with a known history of malignancy may present with a myriad of nonspecific neurological symptoms that can overlap with other neurological complications of cancer, such as brain metastases, paraneoplastic syndromes, treatment-related neurotoxicity, and CNS infections [1].

The subsequent detection of LM involves several different techniques, such as a gadolinium-based contrast-enhanced magnetic resonance imaging (MRI) of the neuraxis to detect leptomeningeal enhancement and other patterns [1]. That being said, sensitivity is variable and especially limited in early-stage LM. If safe to do so, CSF cytology is the gold standard for diagnosis, with findings suggestive of LM including the presence of malignant epithelial cells, elevated protein levels and white blood cell counts, and decreased CSF glucose concentrations [1]. However, its sensitivity is limited in the first lumbar puncture, necessitating repeated sampling toward definitive diagnosis [1].

Given these diagnostic deficits, there is a pressing need for more precise and scalable biomarkers that enable early diagnosis, prognostication, and therapy monitoring in LM. Liquid biopsy techniques analyze tumor-derived material in biological fluids such as blood or CSF [5]. Compared to traditional tissue biopsies or serial lumbar punctures, which aim to identify whole, intact tumor cells, liquid biopsy focuses on the molecular profiling of tumor-derived material, capturing subcellular information toward detecting cancer even if no intact cells are present [6].

Circulating tumor DNA (ctDNA) and circulating tumor cells (CTCs) are two well-established approaches in the liquid biopsy paradigm that have been studied in the context of LM [7]. ctDNA consists of fragmented DNA from tumors that enter the blood or CSF, allowing for the detection of tumor-specific genetic alterations. For LM, CSF-derived ctDNA has better sensitivity compared to plasma-derived ctDNA; dynamic changes in CSF ctDNA were found to directly correlate with clinical progression and therapeutic response in NSCLC patients with LM [8]. Likewise, CTCs extend from traditional CSF cytology analysis in that they use enrichment and molecular characterization techniques to analyze intact tumor cells in the CSF. Ning et al. demonstrated how tumor marker immunostaining-fluorescence in situ hybridization of CTCs was able to detect signs of LM in patients who otherwise had unremarkable findings from CSF cytology, suggesting better diagnostic sensitivity [9].

Regardless, both ctDNA and CTC-based approaches have their own limitations, especially in the context of LM. ctDNA analysis is extremely specific but could be hindered by the availability of tumor DNA in the CSF, especially in the setting of low disease burden subset or in patients with partial treatment response [10]. Furthermore, ctDNA can only capture the genomic landscape of the tumor and provides less information regarding its gene expression, proteomic alterations, or other features related to the tumor microenvironment that are more relevant to disease development and management [11]. Low DNA fragment yield, risk of contamination, and the need for high-depth sequencing platforms are other technical limitations that restrict widespread feasibility [12]. CTC analysis is also constrained by the scarcity and vulnerability of tumor cells in the CSF, and even with the most advanced enrichment technologies, it can be challenging without labor-intensive workflows.

In contrast to prior liquid biopsy approaches, extracellular vesicles (EVs) represent another area of interest. EVs are nanoscale particles covered in lipid bilayer that are secreted by cells, such as cancer cells, and can be present in CSF [13]. Unlike ctDNA and CTC analyses that either appear as exposed fragments or are subject to degradation, EV cargo—including DNA, messenger RNA, microRNAs (miRNAs), proteins, or lipid—are preserved in the lumen of the vesicle due to the membrane, giving them further stability upon CSF circulation [14]. Moreover, EVs are created in vast amounts, which would mean they outnumber CTCs in their infrequent occurrence.

In terms of analysis, EVs can provide a complex molecular portrait of the primary tumor, comprising not only genomic alterations but also providing subsequent transcriptional and proteomic information, giving a more comprehensive picture of tumor biology than ctDNA. EVs were also reported to effectively cross the blood–brain barrier, and for this reason, they present potential in the detection of CNS tumors or metastases, including LM, from simple serum analysis [15] Aside from their function as biomarkers, EV cargo is postulated to mechanistically contribute to LM pathogenesis by facilitating blood–brain barrier damage, immune escape, and drug resistance, making them appealing for understanding the unique trajectories and profile of patients who develop LM [16].

As a field of growing interest, there is no study to date systematically analyzing the advancements in EV-based liquid biopsy for LM. We conduct a systematic review that critically appraises and synthesizes the current evidence examining the role of EVs isolated from the CSF or serum in diagnosis, prognosis, and treatment monitoring for LM, primarily from NSCLC but other solid tumors.

Methods

This systematic review follows the Preferred Reporting Items for Systematic Reviews and Meta-Analyses [17]. The primary goal is to evaluate the role of EVs isolated from the CSF or serum in the diagnosis, prognosis, and treatment monitoring of leptomeningeal metastasis, primarily in non-small cell lung cancer (NSCLC) and other CNS-involved malignancies.

The eligibility criteria, search strategy, outcomes of interest, and synthesis plan were defined a priori before screening, although the review was not prospectively registered. Studies were included if they involved human subjects with leptomeningeal metastasis from solid tumors (e.g., particularly NSCLC or medulloblastoma), reported on extracellular vesicles or exosomes derived from CSF or serum, and analyzed EV cargo (such as microRNAs or proteins) with relevance to diagnosis, prognosis, therapy monitoring, or mechanisms related to LM. Both clinical and translational research studies were included as long as they reported relevant EV-related molecular outcomes. Studies that examined total CSF or serum biomarkers without specific EV isolation were excluded unless results clearly overlapped with vesicle-enriched fractions. Studies focusing solely on primary brain tumors or non-metastatic lesions were also excluded.

A complete search was conducted across four electronic databases: PubMed, Science Direct, Semantic Scholar, and Google Scholar. The search was last updated in January 2025, with the following keywords and Boolean logic used: “(“extracellular vesicle” OR “exosome” OR “liquid biopsy”) AND (“leptomeningeal metastasis” OR “leptomeningeal carcinomatosis” OR “leptomeningeal”)” Additional manual searching of references from eligible articles was also performed. The two authors independently screened the titles and abstracts for initial eligibility, and discrepancies were resolved through discourse.

Data were extracted on study characteristics (authors, year, country), participant population (cancer type, LM confirmation), sample source, EV isolation and profiling techniques (e.g., ultracentrifugation, NTA, mass spectrometry, quantitative polymerase chain reaction, next-generation sequencing), molecular biomarkers assessed (miRNAs, proteins, etc.), main findings, and outcome relevance (diagnostic accuracy, survival association, drug resistance mechanisms).

Given the heterogeneity of biomarkers (different miRNAs and proteins) and outcome measures (e.g., diagnostic accuracy vs. mechanistic assays), a narrative synthesis was performed across five pre-specified themes: diagnostic EV miRNAs, prognostic markers, therapy resistance or monitoring, EV protein biomarkers, and mechanistic roles. The findings of specific biomarkers were tabulated.

A tailored risk of bias assessment for the selected studies was performed following the QUADAS-2 tool for biomarker studies [18]. In addition to QUADAS-2 assessment, each included study was appraised against the Minimal Information for Studies of Extracellular Vesicles 2018 (MISEV2018) framework [19]. For every study we recorded the nomenclature, EV isolation method, particle size and concentration characterization (nonparticle tracking analysis), morphological assessment (transmission electron microscopy), the positive EV protein markers reported, the method used for their detection, and whether negative or contaminant markers were also assessed. Two authors independently assessed risk of bias followed by discussion and consensus.

Results

The PRISMA flowchart describing the outcome of the literature screening is illustrated in Fig. 1. A total of 93 articles were identified across the databases, and 11 articles remained after preliminary screening of the titles and abstract for relevance and removal of duplicates. After detailed screening, ten studies were ultimately included in this systematic review [8, 20–28].

Fig. 1.

PRISMA flow. EV,  Extracellular vesicle

One article was excluded because one focused on EVs in the context of glioma patients instead of leptomeningeal carcinomatosis.

The ten included studies were published between 2018 and 2024, and all investigated EV contents–either microRNAs (miRNAs) or proteins–in patients with LM, most commonly from NSCLC. One study focused on medulloblastoma, and another study also included data from a patient with metastatic breast cancer [20, 22].

Sample sources were either from the CSF or serum, though most (n = 8) used CSF as the primary source for EV isolation. Among studies using CSF as the sample source, sampling sites varied, with most obtained through lumbar puncture, followed by intraventricular or cisternal puncture. Two studies additionally or exclusively profiled serum-derived EVs [8, 27]. Across the studies, participant groups typically included LM-positive patients, with comparators such as LM-negative patients (e.g., NSCLC without LM), non-cancer neurological controls (e.g., patients with hydrocephalus, idiopathic intracranial hypertension, unruptured aneurysms, head trauma, etc.), or healthy individuals. Sample sizes varied, with the smallest including 8–10 LM patients and the largest profiling with over 400 CSF samples. Diagnostic value was generally reported using area under the receiver operating characteristic curve (AUC), sensitivity, specificity, or the log₂-transformed fold change (Log2FC) value, which reflects the magnitude of differential expression. Studies described whether the miRNA was significantly upregulated or downregulated in cases where these quantitative metrics were not provided. All studies used validated EV isolation techniques such as ultracentrifugation, size-exclusion chromatography, and characterization to corroborate these values.

Figure 2 illustrates the risk of bias and applicability concerns across the ten included studies following the QUADAS-2 framework. Overall, the methodological quality of all papers was high, with most having a low risk of bias and strong applicability across all domains. However, some studies had some considerations. We flagged Xu et al., Li et al., and Hou et al. out of concern for patient selection and potentially flow [23, 25, 27]. Xu et al. had a small case–control design with seven LM-positive patients. Similarly, Li et al.’s design was single-arm with a small cohort (n = 9 LM patients, 5 controls). Last, although pooled analyses are common in early-stage biomarker discovery, Hou et al. pooled CSF samples from 10 LM patients into three group-level samples before proteomic sampling, limiting diagnostic assessment at the individual level. Although they did follow up with validation on individual samples using PRM, the primary discovery phase was still group-level. Most studies followed clear LM diagnostic criteria, including clinical features, CSF cytology, and MRI, although this information was slightly unclear in Xu et al.

Fig. 2.

QUADAS-2 risk of bias and applicability concerns summary table

EV miRNAs for diagnosis

miRNAs are small, non-coding RNA molecules, generally 19–25 nucleotides in length, that regulate gene expression by post-transcriptionally silencing messenger RNAs. Four studies identified specific miRNAs within EVs from CSF or serum that could distinguish LM-positive patients from controls or LM-negative patients (Table 1).

Table 1.

Extracellular vesicle (EV)-associated microRNAs with diagnostic value for leptomeningeal metastasis (LM)

miRNA	Primary cancer	Study	Specimen	Diagnostic value*	Validation
hsa-miR-183-5p	NSCLC	Li et al. [23]	CSF	Sn: 1.0 Sp: 1.0 AUC: 1.0	NGS + qRT-PCR
hsa-miR-184				Sn: 0.308 Sp: 1.0 AUC: 0.5
hsa-miR-96-5p				Sn: 0.846 Sp: 1.0 AUC: 0.908
hsa-miR-182-5p				Sn: 1.0 Sp: 1.0 AUC: 1.0
hsa-miR-21-5p	NSCLC, breast cancer	Lee et al. [22]	CSF	Log2FC: 11.5	smRNA-seq + qRT-PCR
miR-19b-3p				Log2FC: 9.5
miR-25-3p				Log2FC: 9.3
miR-200c-3p				Log2FC: 9.1
miR-19a-3p				Log2FC: 8.6
miR-34b-3p				Log2FC: 5.7
hsa-miR-423-5p				Log2FC: 3.9, downregulated
hsa-miR-1273 g-3p				Log2FC: 4.1, downregulated
hsa-miR-4271				Log2FC: 8.6, downregulated
miR-483-5p	NSCLC	Xu et al. [27]	Serum	Upregulated	Exosomal miRNA-seq
miR-342-5p				Upregulated
miR-374a-5p	NSCLC	Jin et al. [8]	Serum and CSF	Upregulated	NGS + Functional assays
CSF exosomal signature (let-7e-5p, miR-142-5p, miR-148a-3p, miR-200b-3p, miR-451a, miR-483-5p, miR-486-5p, miR-96-5p)	NSCLC	Jiang et al. [28]	CSF	Sn: 0.97 Sp: 1.0 (combined panel value)	Exosomal RNA-Seq

The table summarizes miRNAs identified across multiple studies that differentiate LM-positive from LM-negative patients or healthy controls. It includes the biological sample source, cancer type, validation methods used, and diagnostic performance metrics where available

NSCLC, Non-small cell lung cancer; CSF,  Cerebrospinal fluid; Sn, Sensitivity; Sp, Specificity; AUC, Area under the curve; FC, Fold change; qRT-PCR, Quantitative reverse transcription polymerase chain reaction; NGS, Next generation sequencing; smRNA-seq,  Small RNA sequencing

*All findings demonstrate statistically significant differentiable expression with respect to controls (p < 0.05)

Unlike other studies, Xu et al. and Jin et al. placed greater emphasis on serum-derived exosomes [8, 27]. Xu et al. found that serum samples offered higher miRNA expression levels in general compared to CSF, suggesting that blood-based assays may not only be sufficient, but also enhance detection sensitivity and practicality in clinical settings [27].

Members of the miR-200 family (e.g., miR-200b-3p and miR-200c-3p) were frequently identified across different cohorts [22, 28] miR-96-5p and miR-483-5p were also implicated in multiple studies [23, 27, 28].

EV miRNAs for prognosis and therapeutic response

Across the studies, miR-21 emerged as the most well-validated prognostic and treatment response biomarker. Im et al. found that elevated CSF EV-associated miR-21 levels were significantly associated with worse overall survival (OS) in NSCLC patients with LM undergoing intraventricular methotrexate therapy; CSF EV miR-21 expression (n = 9) had significantly shorter OS than those with low expression (n = 27); across the full cohort of 36 patients the OS ranged from 191 to 395 days (median 237 days) (hazard ratio: 2.57, p < 0.01) [21]. The magnitude of miR-21 expression was also weakly but significantly correlated with OS. Lee et al. further reinforced these findings in a larger cohort, describing how LM patients with post-treatment declines in miR-21 expression showed improved survival, while those with rising levels had worse outcomes [26].

In addition to miR-21, Li et al. also pointed to other miRNAs that can help predict treatment response and prognosis [23]. miR-1-3p expression significantly increased in patients who responded to intrathecal chemotherapy, with a log2FC greater than 4 (p < 0.01, false discovery rate < 0.5), suggesting its utility as a marker of positive therapeutic response. Similarly, miR-509-3p levels were elevated in patients demonstrating partial responses to intrathecal permetrexed and were lower in those with progressive disease. In contrast, miR-449a was significantly downregulated in the progressive disease group, potentially reflecting its association with poor prognosis or a more aggressive disease phenotype.

EV protein biomarkers

Proteomic profiling of EVs, all derived from CSF, has also revealed protein biomarker candidates across three studies. These proteins, tabulated in Table 2, all primarily serve a diagnostic rather than prognostic or treatment monitoring role across the studies.

Table 2.

CSF extracellular vesicle (EV)-associated proteins with diagnostic value for leptomeningeal metastasis (LM)

Protein	Function/Use	Study	Diagnostic value*	Validation
Fibronectin 1 (FN1)	NSCLC	Hou et al. [25]	Upregulated	Parallel reaction monitoring
Transketolase (TKT)	Medulloblastoma	Kim et al. [20]	Log2FC: 2.20	ELISA
Complement C2	Lung adenocarcinoma (NSCLC)	Liu et al. [24]	Sn: 0.70 Sp: 0.97 AUC: 0.88	ELISA
Peptidoglycan recognition protein 2 (PGLYRP2)			Sn: 0.97 Sp: 0.70 AUC: 0.90	ELISA
Serum amyloid A4 (SAA4)			Sn: 0.60 Sp: 0.87 AUC: 0.87	ELISA
Alpha-1-microglobulin (AMBP)			Sn: 0.37 Sp: 0.80 AUC: 0.68	ELISA

The table summarizes proteins identified across selected studies that differentiate LM-positive from LM-negative patients or healthy controls. It includes the biological sample source, cancer type, validation methods used, and diagnostic performance where metrics are available

NSCLC, Non-small cell lung cancer; Sn, Sensitivity; Sp, Specificity; AUC, Area under the curve; FC, Fold change; ELISA, Enzyme-linked immunosorbent assay

*All findings demonstrate statistically significant differentiable expression with respect to controls (p < 0.05)

Fibronectin-1 (FN-1) was highlighted by Hou et al. as a central hub protein after showing the highest betweenness centrality in the protein–protein interaction network for the comparison of LM versus NSCLC without LM. This was further confirmed with targeted proteomics analysis using parallel reaction monitoring. Kim et al. also used a similar proteomic pipeline for medulloblastoma, where they revealed how transketolase (TKT) was one of the top feature-selected differentially expressed proteins (log2 fold change 2.20, corresponding to a 4.59-fold increase versus controls) and was the only one of four ELISA-tested candidates whose elevation in medulloblastoma was confirmed by ELISA and single-vesicle (ExoView) analysis. The number of TKT-positive EVs was positively correlated with the burden of leptomeningeal seeding.

Liu et al. focused specifically on NSCLC patients with lung adenocarcinoma. Proteins such as Complement C2 and Peptidoglycan Recognition Protein 2 (PGLYRP2) demonstrated strong discriminative power, especially when combined; the AUC of C2- PGLYRP2was 0.97 when evaluated using ELISA. There were other markers in lung adenocarcinoma as well, with moderate diagnostic efficacy.

EV characterization by study

The completeness of EV characterization also varied widely when assessed against MISEV2018 (Table 3): although most source included nanoparticle tracking analysis, electron microscopy, and tetraspanin markers, few documented negative or contaminant-marker controls [19]. For this reason, we were unable to statistically pool our results and are presenting it as a descriptive summary and stratifying it by analyte, sample source, and EV method. Furthermore, two sources fell short of the criteria for an EV-specific study, namely Jiang et al., which was only available as a conference abstract without significant methodological detail, and Liu et al., which profiled the total CSF proteome instead of a specifically EV-enriched fraction (EV and exosome appearing only as gene-ontology annotations) [24, 28]. Therefore, the protein findings from Liu et al. should be carefully interpreted as CSF-proteome signals that are not yet entirely confirmed to be derived from a single EV analyte. The body of evidence is also strongly weighted towards NSCLC as the primary cancer, with a single medulloblastoma study and only isolated breast-cancer cases, so conclusions about EV-based liquid biopsy for LM apply most directly to NSCLC and cannot be easily generalized to other primary cancers.

Table 3.

Appraisal of the included studies against MISEV2018 criteria for extracellular vesicle (EV) characterization

Study	EV Isolation	NTA	TEM	Positive markers (method)	Negative/contaminant control	MISEV2018 Shortfall
Jiang [28]	Not reported	No	No	None reported	None	Conference abstract; no EV characterization reported
Li [23]	SEC	Yes	Yes	CD9, CD63, TSG101 (WB)	Calnexin (absent)	Meets core criteria
Lee (2024)	Precipitation kit	Yes	Yes	Flotillin-1, CD63, CD81, CD9 (WB)	GM130, cytochrome C	Meets core criteria
Lee [26]	Centrifugation; SEC for verification	Yes (with DLS)	Yes	CD81, CD63 (WB)	GM130, cytochrome C	Meet core criteria
Xu [27]	Precipitation kit	Yes	Yes	CD63, TSG101 (WB)	None reported	No negative/contaminant control
Jin [8]	SEC + ultrafiltration	Yes	Yes	TSG101, CD63, Alix (WB)	Calnexin (absent)	Meet core criteria
Im [21]	Ultracentrifugation	Yes	Not reported	ExoView/MACSPlex (supplementary)	None reported	No TEM; markers in supplement; no negative control
Hou [25]	Ultracentrifugation	Yes	Yes	CD63 (WB); CD9/CD81 proteomic only	None reported	No negative control; pooled discovery samples
Kim [20]	Precipitation kit	Yes	Yes	CD9, CD63, CD81 (ExoView)	None reported	No negative/contaminant control
Liu (2022)	None (total CSF proteome)	No	No	None	None	Not an EV-specific study; total CSF proteome

NTA, Nanoparticle tracking analysis; TEM, Transmission electron microscopy; SEC, Size-exclusion chromatography; WB, Western blot

Discussion

EV-based liquid biopsy for LM is a field of active research, with the existing literature reporting diverse results on the best biomarkers, isolation methods, and ultimately, their clinical relevance. The majority of evidence has focused on either candidate miRNA or protein discovery differentially expressed in LM patients versus controls. Of particular note is that miR-21 was a common correlate with reduced overall survival and also therapy resistance in patients with LM, validated in different cohorts. Similarly, EV-derived proteins such as FN-1, TKT, and complement protein C2 possessed potential diagnostic value, where some of these protein-based EV markers reached AUC values of greater than 0.90 in detecting LM. Regardless, it is key to distinguish discovery-phase findings across the studies from clinically validated biomarkers, as all of the included results derive from a single-center discovery cohort or early-validation cohort, several with fewer than ten LM-positive patients. Of note, several studies reported high diagnostic metrics for some markers, including AUCs of at least 0.90 and perfect sensitivity or specificity. These should only be interpreted from the standpoint of hypothesis-generating signals influenced by overfitting and optimism bias in small case–control designs, rather than as evidence of established diagnostic performance. To date, from our studies, only miR-21 was a marker reproduced across more than one independent cohort, while the remaining miRNA and protein candidates remain at the discovery stage.

miR-183-5p is another candidate biomarker with reported diagnostic value, demonstrating perfect sensitivity and specificity (AUC 1.0) in distinguishing LM-positive patients from controls. Notably, this estimate derives from a single small discovery cohort (nine LM patients and five controls) in which companion microRNA from the same study, miR-184, was non-discriminatory: the perfect accuracy in this study might plausibly be explained by overfitting and optimism bias, and the marker needs further independent validation. Indeed, functional analyses have revealed that miR-183-5p is upregulated in NSCLC, promoting cell proliferation, migration, and tumor growth. miR-183-5p suppresses the tumor suppressor gene PTEN, leading to abnormal activation of the AKT pathway and subsequent downregulation of p53, therefore accelerating tumor progression [29]. Overexpression of this microRNA has also been implicated in pre-neoplastic phenotypes of breast cancer, although through several other potential pathways in addition to the PTEN/PI3K/AKT pathway [30].

The role of EVs in modeling LM diagnosis and prognosis can be explained by the pathophysiological basis of LM. Although the pathogenesis is not fully understood, EVs have been implicated in several key processes leading to disease development. Tumor-derived EVs can disrupt the integrity of the blood–brain barrier, with potential invasion of tumor cells into CSF and metastatic seeding along the leptomeninges. For example, EV cargo like miR-374a-5p directly targets ADD3 (gamma-adducin) and disrupts the junctional distribution of tight-junction proteins like ZO-1 and occludin, ultimately increasing BBB permeability without reducing the overall expression of these proteins [8]. Similarly, FN-1 is involved in extracellular matrix remodeling and epithelial-mesenchymal transition, both of which are involved in metastatic seeding and progression within the leptomeninges [25].

While most studies in this review used CSF as the primary sample for EV harvesting, a few explored serum-based EVs. Given that CSF is in direct contact with LM, it is expected that CSF-derived EVs would be enriched with tumor-specific cargo and more appropriately reflect the microenvironment. However, CSF accessed by lumbar puncture is invasive and may not be feasible, especially in cases where tumor infiltration occludes the subarachnoid space, resulting in a dry tap [31]. In such cases, aside from attempting higher-level access, serum samples can be a convenient alternative, in addition to being less invasive for analysis [32]. Xu et al. and Jin et al. explored serum-derived exosomal biomarkers, noting that serum EV miRNA expression levels were generally higher than their CSF sample counterparts, potentially improving sensitivity in detection [8, 27]. That being said, the use of serum may potentially introduce confounding signals from systemic disease or non-tumor-derived EVs, making it necessary for future studies to distinguish LM-specific EVs from those originating from the extracranial primary tumor or systemic circulation [33]. Current evidence supports CSF as the reference compartment for EV-based LM biomarkers, as CSF is in direct contact with leptomeningeal disease and its EV pool is less diluted by systemic, non-tumor sources. Serum can be regarded as a complementary or triage compartment rather than an equivalent substitute. Based on the included studies and their findings, serum can reasonably be considered as an acceptable surrogate only under specific conditions: the target marker has been shown to be concordant between paired serum and CSF samples from the same patient; CSF sampling is contraindicated or yields a dry tap in the setting of subarachnoid obstruction; the assay is used for longitudinal assessment of a marker already validated in CSF rather than for primary diagnosis; and the analysis accounts for EVs arising from the extracranial primary tumor or systemic inflammation. In contrary, serum would be a poor proxy for initial diagnosis if the primary tumor or systemic disease contributes to the same marker, when the marker has not been cross-validated against CSF, or when low-burden disease would produce only small numbers of LM-specific EVs.

EVs have also been considered for other CNS neoplastic lesions. Chen et al. explored the use of CSF-derived EVs in patients with glioma for the detection of tumor-specific mutations [34]. Using BEAMing and droplet digital PCR technologies, they analyzed EV samples to identify mutant IDH1 mRNA. This positions EV-based liquid biopsy as a potential alternative to traditional brain tumor tissue sampling.

There are some limitations within the studies aside from the risk of bias analysis. Most studies enrolled relatively small samples of LM patients, often fewer than 50 participants, and most were single-center studies. There was also heterogeneity in study designs, where some were retrospective, others prospective, and others case–control. These differences may have influenced the reported diagnostic measures. The heterogeneity in EV isolation and profiling strategies and lack of uniform protocols can also partly account for inconsistent biomarker discovery; some studies used ultracentrifugation, while others used size-exclusion chromatography or affinity capture, impacting the yield, purity and characterization of the EV populations.

We could not perform a formal meta-analysis given the heterogeneity in study designs and isolation techniques. The differences in outcome reporting, such as differences in diagnostic metrics, and variable approaches to confirm LM across studies, make it difficult to quantitively pool results when there are limited studies. In addition, this review was not prospectively registered, which raises the possibility of selective reporting. To mitigate this limitation, we have provided a full search strategy, eligibility criteria, screening process, full-text exclusion reasons, extracted study characteristics, and study-level risk-of-bias assessments, all of which can be replicated. There is also a possibility of publication bias, as studies reporting negative or inconclusive results with respect to EV-based biomarkers were underrepresented in our analysis.

Integration with existing LM diagnostics

Given the current state of evidence, EV-based assays are not likely to replace current LM diagnostic approaches in the near term and are better positioned as a complement to them. Contrast-enhanced MRI of the neuraxis and CSF cytology should remain as the current diagnostic pillars, with cytology providing high specificity but limited and operator-dependent sensitivity on a first lumbar puncture, and MRI providing the anatomical localization but variable sensitivity in early disease. ctDNA and circulating tumor cells currently appear to extend molecular profiling: CSF ctDNA shows high sensitivity for actionable mutations and is moving toward prospective validation, while CTC enumeration can enhance diagnostic yield over cytology alone. Within this domain, EV cargo adds a distinct, stable, and multi-analyte readout (miRNA and protein) that can be sampled from CSF and, potentially, serum. A more feasible role would be to use EV markers, such as miR-21, for prognostic stratification and serial monitoring alongside cytology and MRI, and as an adjunct that raises diagnostic sensitivity when cytology is negative but clinical suspicion continues to be high. Demonstrating further value with EV-based assays will require head-to-head studies that report EV performance against, and in combination with cytology, MRI, ctDNA, and CTCs in the same patients, rather than against healthy controls alone.

Particularly, translation of these discovery-phase findings into clinical practice will need a staged pathway. First, analytical standardization is necessary: EV isolation and characterization studies need to conform to MISEV2018 standards, with defined pre-analytical handling, reporting of nanoparticle tracking analysis, electron microscopy, and positive and negative markers, and demonstration of assay reproducibility across laboratories. Second, the most reproducible candidates, such as miR-21, miR-183-5p, FN1, and TKT, should advance to analytical validation with locked assays predefined positive thresholds, instead of thresholds placed post hoc. These thresholds should also be tested in adequately powered, prospective, multicenter cohorts that enroll consecutive patients with suspected LM, use blinded comparison against a reference standard (cytology and MRI) and pre-register the analysis plan. Finally, clinical utility, such as improved diagnosis, prognostication, or treatment decisions relative to existing tests, should be demonstrated before adoption, ideally including the incremental value of EV assays added to ctDNA or CTC testing. Currently, prospective, registered validation in LM exists only for CSF ctDNA, and from our search, we are unaware of a completed prospective diagnostic trial or formal cost-effectiveness assessment specific to EV-based assays in LM. Cost, turnaround time, and technical demands of ultracentrifugation or size-exclusion workflows also need to be weighed before routine use can be justified.

Regardless, the current findings of this systematic review suggest that EVs, especially miRNAs and proteins, might, with further validation, complement existing diagnostic approaches to LM (Fig. 3). Serial EV profiling, such as tracking miR-21 or miR-1-3p levels, could help monitor treatment response and enhance the real-time management of patients with LM, allowing for earlier, targeted therapeutic intervention if needed. However, there are several hurdles before EV-based assays can be translated into clinical practice. In addition to the heterogeneity of EV populations, lack of standardized isolation protocols, and risk of contamination by non-vesicular material, technical limitations in EV yield and sample handling can limit its reproducibility [35]. With continued large-scale validation, EV-based liquid biopsies can eventually be standardized to improve the care of LM, with earlier diagnosis, more personalized management, and better monitoring.

Fig. 3.

Schematic overview of extracellular vesicle-based liquid biopsy for leptomeningeal metastasis (LM) detection. Solid tumors such as non-small cell lung cancer (NSCLC) can metastasize to the leptomeninges, releasing EVs into the cerebrospinal fluid and blood circulation. Tumor-derived EVs contain cargo of diagnostic and prognostic interest, such as microRNAs and proteins. Biological samples can be obtained via lumbar puncture or blood draw, allowing for EV isolation and molecular profiling for diagnosis, prognosis, and therapy monitoring. Made using biorender.com

Abbreviations

AUC

Area under the curve

BBB

Blood–brain barrier

CNS

Central nervous system

CSF

Cerebrospinal fluid

CTC

Circulating tumor cell

ctDNA

Circulating tumor DNA

ELISA

Enzyme-linked immunosorbent assay

EV

Extracellular vesicle

FC

Fold change

FN1

Fibronectin 1

IDH1

Isocitrate dehydrogenase 1

LM

Leptomeningeal metastasis

miRNA (or miR)

MicroRNA

MRI

Magnetic resonance imaging

NGS

Next generation sequencing

NSCLC

Non-small cell lung cancer

OS

Overall survival

PGLYRP2

Peptidoglycan recognition protein 2

PRISMA

Preferred reporting items for systematic reviews and meta-analyses

PRM

Parallel reaction monitoring

qRT-PCR

Quantitative reverse transcription polymerase chain reaction

QUADAS-2

Quality assessment of diagnostic accuracy studies, version 2

SAA4

Serum amyloid A4

SmRNA-seq

Small RNA sequencing

Sn

Sensitivity

Sp

Specificity

TKT

Transketolase

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Data availability

Data extracted from included studies will be made available on Mendeley Data.

Declarations

Conflict of interest

The authors have no competing interests or disclosures.