Targeted next-generation sequencing-based sequencing of cell-free DNA in cerebrospinal fluid uncovers cancer-specific mutations in patients with brain cancer using a widely available panel

TL;DR

This study shows that sequencing cell-free DNA in cerebrospinal fluid can detect cancer mutations in brain tumor patients, offering a less invasive diagnostic option.

Contribution

The study demonstrates the clinical feasibility of using a widely available sequencing panel on CSF cfDNA for brain tumor diagnosis.

Findings

Sequencing was successful in 55% of CSF samples and identified more tumor-specific mutations compared to blood.

CSF sequencing aided diagnosis in two high-risk cases where surgery was not possible.

Tumors closer to the CSF space and larger in size were more likely to yield successful sequencing results.

Abstract

The gold standard for the diagnosis of brain tumors comprises neurosurgical biopsies or resections for tissue acquisition. Depending on the anatomical location of the tumor, this might not always be feasible. The aim of this study was to assess whether targeted sequencing of cell-free DNA (cfDNA) could be applied in a clinical setting for the detection of cancer-specific mutations in cerebrospinal fluid (CSF) and blood using a widely available panel, investigating if this approach would facilitate brain tumor diagnosis.

Patients with newly suspected or confirmed CNS tumors or suspected leptomeningeal metastasis were included. CSF and blood were sampled for each patient. Using next-generation sequencing (NGS), targeted sequencing was performed on cfDNA. If available, matched tissue samples were also sequenced using the same gene panel.

We investigated 58 samples from 29 patients. Sequencing was successful in 16 CSF samples (55%) and 29 blood samples (100%). On average, more tumor-specific mutations were found in CSF compared to blood. Patients in whom cfDNA sequencing in CSF was successful had tumors with a significantly shorter distance to the CSF space and were larger in size. In 2 cases, where surgeries were not considered because of high surgical risk due to the anatomical locations of the lesions, the diagnosis was aided by CSF sequencing results. One primary CNS lymphoma and 1 H3 K27M diffuse midline glioma were identified, respectively.

NGS-based targeted sequencing in CSF has the potential to facilitate brain tumor diagnosis in patients presenting with new cerebral lesions.

Tables

Table 1.. Clinical characteristics of included patients

	Overall (N = 29)
Age (years)	Mean: 55.4
Sex
Male	15 (51.7%)
Female	14 (48.3%)
Disease entity
AEG	1 (3.4%)
Astrocytoma, high grade	1 (3.4%)
Breast cancer	5 (17.2%)
Germinoma	2 (6.9%)
Glioblastoma	5 (17.2%)
LPD	1 (3.4%)
Lung cancer	3 (10.3%)
Lymphoma	1 (3.4%)
Melanoma	4 (13.8%)
non-neoplastic	3 (10.3%)
Unknown	3 (10.3%)
Blood analysis
Failed	0 (0%)
Successful	29 (100%)
CSF analysis
Failed	13 (44.8%)
Successful	16 (55.2%)

Full text

Brain cancer is still one of the most challenging malignancies. An accurate diagnosis is important to select the most appropriate treatment for each patient and is based on histopathological and molecular analysis of tumor material according to the 2021 WHO classification of CNS tumors.1 Currently, this requires tissue sampling of the tumor. However, surgical procedures like biopsies or resections have a perioperative risk and specific limitations in the CNS. Stereotactic biopsies are associated with an inpatient mortality of 2.8%.2 Especially in cases where the tumor is located in eloquent regions like the brain stem, a biopsy might not be considered feasible.

Recent technological advances have demonstrated the ability to detect cancer-specific mutations in cell-free DNA (cfDNA). Cell-free DNA is shed into the blood or cerebrospinal fluid (CSF) when cells die.3 Previous studies used digital droplet PCR to detect hot-spot mutations like the isocitrate dehydrogenase 1 mutation in cfDNA in blood or other tumor-defining mutations like the H3 K27M mutation for diffuse midline glioma in serial CSF samples.4^,^5 In a multiplexed approach, selected genes, including MYD88, were analyzed for the detection of CNS lymphoma in CSF.6 Other groups have created in-house next-generation sequencing (NGS) panels for the molecular-guided classification of CNS tumors based on single nucleotide variants, copy number variations, or detection of insertions and deletions in the CSF.7-10 Compared with peripheral blood, CSF is the preferred analyte due to its proximity to the tumor in patients with brain cancers.11

The FoundationOne Liquid CDx (F1LCDx) is an NGS-based gene panel for genomic profiling of tumor-derived cfDNA in blood.12^,^13 However, the applicability to brain tumors and CSF is unknown. In the present study, we investigated the performance of this NGS-based gene panel for the genomic profiling of cfDNA extracted from CSF and blood in a clinical setting in patients with cerebral lesions.

Methods

Patient Cohort

Patients with a new cerebral lesion as examined by MRI or suspicion of leptomeningeal metastasis or known cerebral tumors with a clinically indicated CSF analysis were approached for informed consent from 2023 to 2025. Written consent was obtained before the sampling of blood or CSF. The study was conducted in accordance with the World Medical Association Declaration of Helsinki, and approval by the Institutional Review Board was obtained (BASEC-Number: 2021-00652). Exclusion criteria were age under 18 years or contraindications to lumbar puncture (eg anticoagulation, platelet count <50 G/l, risk of herniation).

CSF and Blood Collection

Additional CSF was collected at the occasion of a clinically indicated lumbar puncture according to the standard operating procedure of the Department of Neurology at the University Hospital Zurich. For both blood and CSF, the same cfDNA collection tubes were used (CE-IVD Cell-Free DNA Collection Tube, Roche, Basel, Switzerland). For each patient, a minimum of 2.5 mL and a maximum of 10 mL CSF was collected. Peripheral blood was obtained via venipuncture collecting 16 mL. As per clinical routine, CSF chemistry (glucose, lactate) and CSF/blood albumin ratios were measured. Blood was centrifuged the same day of venipuncture at 2000g for 15 min to separate plasma and was stored at 4 °C.

DNA Extraction

For isolation of cfDNA from blood and CSF, the MagMAX Cell-Free DNA Isolation Kit (ThermoFisher, Waltham, MA) on a KingFisher system (ThermoFisher, Waltham, MA) and a single extraction method were used according to the manufacturer’s protocol. For genomic DNA extraction from formalin-fixed paraffin-embedded tissue, the Maxwell RSC DNA FFPE Kit (Promega) on a Maxwell RSC instrument (Promega) was used following the manufacturer’s protocol.

Molecular Analysis

Liquid biopsy specimens from plasma and CSF were analyzed using a research-use-only assay based on F1LCDx (Foundation Medicine, Inc., Cambridge, MA), an NGS-based in vitro diagnostic assay. This assay employs the whole-genome shotgun method for library construction and targeted high-throughput hybridization-based capture technology to detect and report alterations in 324 genes, as well as ctDNA tumor fraction and genomic signatures, including microsatellite instability and blood tumor mutational burden. Following hybrid capture, the libraries were sequenced with deep coverage using the Illumina NovaSeq 6000 platform. It is important to note that this assay has not been designed for CSF analysis, and the application to CSF samples in the present study does not represent the intended use of the F1LCDx test. Bioinformatic analysis, including detection of genomic alterations, that is, base substitutions, indels, selected copy number variants (gene amplifications and homozygous deletions), and selected genomic rearrangements as well as biomarkers, was performed using a proprietary customized analysis pipeline developed by Foundation Medicine, Inc. Successful sequencing was defined, among other criteria, as achieving a predefined minimum coverage threshold across the panel, ensuring sufficient depth for reliable variant calling. Fragment length of cfDNA was calculated from mapped sequencing data as the insert size, that is, the length of the DNA fragment between adapters. The details on the sequencing and analytical validation of the assay have been published previously.12 Clinical interpretation of sequencing results was restricted to established brain tumor mutations, excluding isolated variants in genes frequently associated with clonal hematopoiesis.

Tumor Radiological Characteristics

To approximate the tumor size, the product of diameter (POD, largest perpendicular diameter) was calculated using contrast-enhanced, T1-weighted MRI. The MRI scan closest in time to the lumbar puncture was selected for analysis. If the cerebral lesion did not show any contrast enhancement, the patient was not further analyzed regarding tumor size.

The distance from the tumor to the CSF space was defined as the shortest distance of the contrast-enhancing tumor to the CSF on contrast-enhanced, T1-weighted MRI. Patients with newly diagnosed brain lesions, which were diagnosed later as non-neoplastic, were excluded for this analysis.

Statistical Methods

All statistical analysis were performed in GraphPad Prism (Version 10.3.1). Normal distribution was checked using Shapiro–Wilk test for group-wise comparisons of continuous variables. For normally distributed variables, we used a 2-sided t-test. For non-normally distributed variables, we applied the non-parametric Mann–Whitney U-test, unless otherwise stated. Statistical significance was set to P < .05.

Results

NGS-Based cfDNA Sequencing in CSF is Feasible

Twenty-nine patients with new cerebral lesions, a suspicion of leptomeningeal metastasis, or known cerebral tumors were included (Figure 1A; Table 1; Table S1). Sequencing of cfDNA in CSF was successful in 16 patients (55%) (Figure 1B). In contrast, in blood, it was successful in all samples (N = 29, 100%). To investigate confounding variables that could explain the success of cfDNA sequencing in CSF, we considered several factors, such as total CSF volume retrieved, DNA amount, or tumor size. Regarding collected CSF volumes, there were no significant differences between the groups with successful sequencing versus failed sequencing (P = .74). The mean volume was 7.03 ± 2.11 mL in the successful sequencing group and 6.79 ± 1.4 mL in the failed sequencing group (Figure 1C). While the extracted DNA amount was not significantly different (9.01 ± 8.14 ng vs. 6.47 ± 3.71 ng, P = .56), it should be noted that the accepted cfDNA input for the assay is specified as 20 ng. Only after hybrid capture, the DNA yield in the successful sequencing group was higher than that in the failed sequencing group (P = .003, Figure S1A). Notably, patients for whom the sequencing of cfDNA from the CSF was successful had a contrast-enhancing tumor closer to the CSF than those for whom sequencing was not successful (0.39 ± 1.07 mm vs. 3.13 ± 2.64 mm, P = .0008) (Figure 1D). In addition, the tumor size approximated by the POD was larger in the successful sequencing group (899.1 ± 915.6 mm^2^ vs. 148.3 ± 99.48 mm^2^, P = .0027, Figure 1D). Furthermore, glucose levels, lactate levels, and the albumin ratio were not significantly altered between these 2 groups (Figure S1C).

*Targeted NGS of cfDNA in CSF versus peripheral blood. (A) Workflow of the presented study. (B) Ratio of successful cfDNA sequencing from CSF. (C) Left: CSF volume in mL after lumbar puncture for the NGS successful versus NGS failed groups (P = .74). Right: DNA yield in ng for these groups (P = .56). (D) Left: Distance to CSF in mm measured by contrast-enhancing tumor using contrast-enhanced, T1-weighted MRI (P = .0008). Right: Tumor size measured by product of diameter (POD) in mm2 of contrast-enhancing lesion on T1-weighted MRI (P = .0027). NGS, next-generation sequencing; cfDNA, cell-free DNA; CSF, cerebrospinal fluid **<0.01 **<0.001.

cfDNA Sequencing Was More Informative in CSF than in Blood in This Cohort

We sequenced CSF and blood for each patient. The average length of cfDNA fragments was 167 bp in blood, whereas the signal from cfDNA from CSF showed several peaks, with the most frequent at 167 and 143 bp (Figure S2). On average, we found 17 mutations, including known pathological mutations and variants of uncertain significance, in the CSF compared to 8 in the blood (Figure 2A). In the CSF, 3 known pathological mutations were found on average compared to 1 in the blood. Next, we determined which genetic profiles were revealed by NGS-based cfDNA sequencing in patients with glioblastoma and compared the liquid-based sequencing to tissue-based sequencing results. For 3 patients, we had matched tumor tissue available (Figure 2B). Most mutations found within the tissue were also found in cfDNA in the CSF, but not in the blood. Importantly, a tissue-confirmed BRAF V600E mutation was only detected in the CSF, but not in the blood. Other tissue alterations like CDKN2A loss, CDKN2B loss, or MTAP loss were neither detected on cfDNA in CSF nor in the blood. Twenty-two known pathogenic mutations were found in tissue and CSF combined, of which 12 (55%) were shared between tissue and CSF (Figure 2C). In contrast, only 4 of 20 (18%) known pathogenic mutations were shared between tissue and blood in these 3 glioblastoma patients.

Sequencing results of cfDNA in CSF and blood and comparison to tissue. (A) Left: Mutations found in blood and CSF (known pathological mutations and variants of uncertain significances [VUS]). Each point represents 1 patient. Right: Reduced to known pathological mutations detected in serum and CSF. (B) For N = 3 patients with known glioblastoma-matched tissue was available. The plot shows which mutations were shared or private between tissue, CSF, and serum. White squares indicate that the alteration was not present in this patient. (C) Pie charts showing the distribution of shared or private mutations among all identified variants between each 2 of CSF, serum, and tissue of the same 3 patients with glioblastoma. In the lower pie chart of CSF and serum, the results of N = 5 patients with known glioblastoma are shown. cfDNA, cell-free DNA; CSF, cerebrospinal fluid.

cfDNA Sequencing in CSF Enables the Diagnosis of New Cerebral Lesions

Within this study, we also included patients with a suspicion of brain tumor. Patient 19 was a 74-year-old male patient presenting with progressive reduced alertness and confusion. The brain MRI revealed a diffuse contrast-enhancing parietal mass close to the ventricle (Figure 3A). Sequencing of cfDNA from CSF showed a MYD88 mutation with a variant allele frequency (VAF) of 18.37% (Figure 3B;  Figure S3). The mutation was not detected in cfDNA from blood. Due to the rapid decline of the patient, a biopsy was performed, which later confirmed the MYD88 mutation leading to the diagnosis of a primary CNS lymphoma (Figure 3C).

Sequencing of cfDNA in CSF enables diagnosis of brain cancer. (A) Contrast-enhanced, T1-weighted MRI of patient 19 and patient 36 with new cerebral lesions. (B) Results of cfDNA sequencing of patients 19 and 36 in CSF and blood. The known pathological mutations are marked in yellow, in white VUS. (C) Showing the results of tissue sequencing from a biopsy of patient 19. cfDNA, cell-free DNA; CSF, cerebrospinal fluid; VUS, variants of uncertain significance.

Patient 36 was an 88-year-old male patient. The MRI showed several contrast-enhancing lesions in the brain stem close to the fourth ventricle and the aqueduct (Figure 3A). A biopsy was not recommended by the interdisciplinary tumor board due to the associated risks of neurological deficits. Sequencing of cfDNA in the CSF revealed an H3 K27M alteration with a VAF of 58.82% (Figure 3B; Figure S4). This alteration was not detected in the blood. Given the imaging findings and the sequencing results, after interdisciplinary discussion, a diffuse midline glioma, H3 K27M altered, was the preferred diagnosis, and the patient underwent further treatment accordingly.

Discussion

Here, we report on the feasibility of applying an NGS-based gene panel for sequencing cfDNA from CSF and blood of brain tumor patients to reveal cancer-specific mutations. Compared to sequencing from blood, sequencing cfDNA from CSF was superior in detecting more mutations in patients with brain cancer, as reported in previous work.14

The proximity of brain tumors to the CSF relative to blood could be 1 reason accounting for this. The blood–brain barrier might hinder the transfer of cfDNA into the bloodstream. The success rate of 55% of CSF samples is in the range of what others have reported.9^,^15 We showed that successful sequencing in CSF was associated with a larger tumor size and a shorter lesion distance to the CSF.16-19 However, regarding the tumor size, there are conflicting data.7 How tumor size, distance to CSF, and other factors affect NGS performance needs to be characterized in future large prospective clinical trials. Alterations that were detected exclusively in blood but were absent in matched tissue and CSF might originate from clonal hematopoiesis.20

Low cfDNA levels were postulated in other studies as the reason for the failure of analysis.9 More sensitive NGS panels have been developed to overcome low-input limitations.8 Here, the total amount of extracted cfDNA did not differ significantly between successfully sequenced and failed groups. Interestingly, a significant divergence in DNA yield after hybrid capture, with a trend already observable after library preparation, might indicate that the quality and integrity of cfDNA fragments might have influenced sequencing outcomes. As observed in our fragment size distribution analysis, blood-derived cfDNA showed a single peak at 167 bp, consistent with mononucleosomal DNA, while CSF cfDNA exhibited a broader range of fragment sizes, with a peak at 167 bp but additional smaller peaks (Figure S2). These differences in fragmentation patterns between blood and CSF align well with results from previous works on CSF obtained from lumbar puncture.21 Smaller or degraded cfDNA fragments may lead to reduced adapter ligation efficiency or lower PCR amplification efficiency. This could explain why DNA yield was significantly lower in failed sequencing samples after hybrid capture, even though extracted cfDNA amounts were similar across groups. Optimized low-input library protocols might improve the sequencing success rate. One further potential reason could be the abundance of cfDNA from healthy cells in blood, which may contribute to sequencing success while being non-informative compared to CSF in brain cancer patients. Further analysis to clarify the causes of these differences in sequencing success is needed.

While tumor-derived cfDNA in both blood and CSF is known to exhibit a different size distribution compared to non-tumor cfDNA, the exact mechanism that leads to cfDNA fragmentation is unclear.22-24 In previous work, mutations were more frequently detected in samples obtained via ventriculostomy compared to those collected by lumbar puncture.9 For larger tumors, lesions close to the ventricles, and when sampling near the time of surgery, more intact tumor DNA might reach the lumbar cistern, leading to higher detection rates.25 Beyond DNA properties, pre-analytical factors such as sample handling and storage or variability in lumbar puncture technique may have contributed to sequencing failure. Nevertheless, CSF is a low-protein matrix and therefore conveys lower potential interacting biomolecules than, for example blood, which leads to higher stability of several markers in CSF.

We tested the targeted sequencing of cfDNA in CSF and blood in a clinical setting. Thus, some cases turned out to be non-neoplastic (N = 3). For other patients, the final diagnosis is still pending due to a watch-and-wait strategy or inconclusive results. Nonetheless, we demonstrated that the proposed approach could add value to the diagnostic steps by revealing cancer-specific mutations, consistent with previous findings.26-28 In 2 cases, this led, combined with the imaging findings and after interdisciplinary discussion, to the diagnosis of a primary central nervous system lymphoma (PCNSL) and a diffuse midline glioma, H3 K27M-altered, respectively. In addition, we also detected a BRAF V600E mutation in the CSF of 1 glioblastoma patient, which is a druggable alteration.29 The ability to non-invasively detect actionable mutations could enhance treatment planning.

Major limitations of this study include the small sample size, heterogeneity of diagnoses, and enrichment for metastatic, non-CNS primary tumors. Further studies on homogeneous populations are necessary to validate clinical utility and to determine diagnostic accuracy. The NGS panel used in this study is not validated for CSF. Therefore, a larger cohort is needed in future studies to determine sensitivity, specificity, and limit of detection for sequencing of cfDNA in CSF. Moreover, matched sequencing of leukocyte-derived DNA should be performed in future studies to filter out mutations derived from clonal hematopoiesis.

Additionally, while our sequencing panel covered a broad range of oncogenic mutations, it is possible that other relevant genetic alterations were missed due to panel design constraints. In summary, NGS-based cfDNA sequencing in CSF should be considered if the cerebral lesion is close to the CSF. Moreover, in cases where invasive procedures such as biopsies pose a high risk, cfDNA sequencing from CSF may serve as a valuable alternative.

Supplementary Material

vdaf270_Supplementary_Data