Next-generation sequencing of cerebrospinal fluid for clinical molecular diagnostics in pediatric, adolescent and young adult brain tumor patients

Abstract

Background

Safe sampling of central nervous system tumor tissue for diagnostic purposes may be difficult if not impossible, especially in pediatric patients, and an unmet need exists to develop less invasive diagnostic tests.

Methods

We report our clinical experience with minimally invasive molecular diagnostics using a clinically validated assay for sequencing of cerebrospinal fluid (CSF) cell-free DNA (cfDNA). All CSF samples were collected as part of clinical care, and results reported to both clinicians and patients/families.

Results

We analyzed 64 CSF samples from 45 pediatric, adolescent and young adult (AYA) patients (pediatric = 25; AYA = 20) with primary and recurrent brain tumors across 12 histopathological subtypes including high-grade glioma (n = 10), medulloblastoma (n = 10), pineoblastoma (n = 5), low-grade glioma (n = 4), diffuse leptomeningeal glioneuronal tumor (DLGNT) (n = 4), retinoblastoma (n = 4), ependymoma (n = 3), and other (n = 5). Somatic alterations were detected in 30/64 samples (46.9%) and in at least one sample per unique patient in 21/45 patients (46.6%). CSF cfDNA positivity was strongly associated with the presence of disseminated disease at the time of collection (81.5% of samples from patients with disseminated disease were positive). No association was seen between CSF cfDNA positivity and the timing of CSF collection during the patient’s disease course.

Conclusions

We identified three general categories where CSF cfDNA testing provided additional relevant diagnostic, prognostic, and/or therapeutic information, impacting clinical assessment and decision making: (1) diagnosis and/or identification of actionable alterations; (2) monitor response to therapy; and (3) tracking tumor evolution. Our findings support broader implementation of clinical CSF cfDNA testing in this population to improve care.

cell-free DNA, cerebrospinal fluid (CSF), circulating tumor DNA, liquid biopsy, NextGen sequencing, pediatric brain tumors

Key Points

• We detected tumor-derived mutations in nearly 50% of cerebrospinal fluid (CSF) samples from pediatric/adolescent and young adult primary brain tumor patients.

• CSF cell-free DNA detection was associated with the presence of disseminated disease.

• CSF liquid biopsy improved diagnosis, clinical care, and decision making.

Importance of the Study

Accurate histological and molecular diagnosis of central nervous system tumors requires access to tumor tissue. In many clinical circumstances, however, tissue sampling may be difficult or impossible, especially in pediatric patients. Therefore, an urgent need exists to develop less invasive diagnostic tools, that is, “liquid biopsy”. Here, we report our experience with the implementation of cerebrospinal fluid (CSF)-based liquid biopsy into the care of pediatric brain tumor patients. We were able to detect tumor-derived cell-free DNA (cfDNA) from 21/45 patients (46.6%) across twelve histopathological subtypes. We identified three main areas where patient care was enhanced through CSF cfDNA detection: (1) establishing a diagnosis and/or identification of actionable genomic alterations in patients with surgically inaccessible tumors or nondiagnostic tumor biopsy, (2) assessing response to therapy, and (3) tracking tumor evolution. Our findings support broader implementation of clinical CSF cfDNA testing in this population.

In the pediatric, adolescent and young adult (AYA) population, central nervous system (CNS) tumors are one of the leading causes of cancer-related death. Based on anatomical location, sampling of tumor tissue for diagnostic purposes may carry significant risks or not be feasible. As a result, less invasive sampling methods to help establish a diagnosis, that is, “liquid biopsy,” are needed. Recent technological advances have led to the development of “liquid biopsy” assays predominantly in a research setting, which detect circulating tumor cell-free DNA (cfDNA) in blood, cerebrospinal fluid (CSF), or other body fluids. Tumor cfDNA is fragments of DNA released by tumor cells that then circulate in body fluids. Analysis of cfDNA across cancer types has shed light on the genomic profile of the tumor and allowed for the monitoring of tumor evolution.^1,2 We and others have demonstrated that circulating tumor cfDNA can be detected in the CSF of patients with primary and secondary brain tumors.^2–7

Early detection of minimal residual or recurrent disease (MRD), prior to definitive clinical or imaging progression, is invaluable in tailoring therapy and improving outcomes in subsets of hematological malignancies and solid tumors. In addition, “liquid biopsy” assays may track clonal and genetic evolution of cancer cells and identify individualized molecular targeted therapy.⁸ Recently, Liu et al. demonstrated that low-coverage whole-genome sequencing on CSF cfDNA could be used to detect MRD in medulloblastoma patients, and that persistent MRD positivity during or at the end of treatment was highly predictive of tumor recurrence.⁹ Given the rarity of serial sampling of tumor tissue in the pediatric and AYA brain tumor population, our understanding of molecular evolution in response to therapy remains limited.

In the pediatric and AYA population, diffuse midline glioma (DMG) is one of the most challenging brain tumors. The disease-defining mutation H3 K27M has been detected in a subset of banked CSF specimens using PCR, Sanger sequencing,¹⁰ and NextGen Sequencing.¹¹ Panditharatna et al. analyzed liquid biome specimens from 84 subjects (48 DMG patients and 36 controls without CNS disease) enrolled on a clinical trial (plasma samples) or who were part of a brain tumor biorepository (CSF samples). Using digital droplet PCR (ddPCR), histone 3 mutant alleles were detected in 75% of the CSF samples collected at diagnosis, 66.7% during treatment, and 90% at autopsy.¹² More recently, Escudero et al. demonstrated that in medulloblastoma patients, molecular subgroup classification and risk stratification were feasible and genomic evolution could be tracked when the CSF cfDNA had a high yield.¹³ The majority of this work has been performed on stored samples and on a research basis.

To our knowledge, real-time implementation of CSF cfDNA testing in the clinical setting has not yet been previously described. We hypothesized that integrating CSF cfDNA evaluation into clinical care would enhance and refine diagnosis, prognosis, and treatment. We established a pediatric and AYA neuro-oncology liquid biopsy program at Memorial Sloan Kettering Cancer Center (MSK) and collected CSF samples when clinically indicated via lumbar puncture (LP) or ventricular access device (VAD; Ommaya reservoir), as well as during select neurosurgical procedures (eg, tumor resection and CSF diversion). Fresh CSF was submitted for cfDNA extraction and next-generation sequencing using the Memorial Sloan Kettering-Integrated Molecular Profiling of Actionable Cancer Targets (MSK-IMPACT) assay, which is Clinical Laboratory Improvement Amendments (CLIA) certified by the New York State Department of Health for CSF cfDNA profiling, and results were reported to both clinicians and patients/families with approximately 2–3 weeks turnaround time. Here, we report our experience over a 30-month period (May 2018 to November 2020) with 45 patients (64 samples) across twelve different tumor entities.

Methods

Patients

Our study included CSF and tumor samples from 45 unique patients with pediatric CNS tumors seen at Memorial Sloan Kettering Kids (MSK Kids) between May 2018 and November 2020. All patients signed informed consent under protocols approved by the MSK Institutional Review Board. All CSF samples were collected as part of clinical care, with results reported to both clinicians and patients/families.

Magnetic Resonance Imaging

All patients underwent brain magnetic resonance imaging (MRIs) as indicated by standard of care with standard sequences including axial T1-weighted, T2-weighted, fluid-attenuated inversion recovery, and contrast T1-weighted images. The MRIs closest to the CSF collection date were reviewed by a board-certified radiologist with a certificate of added qualification in neuroradiology who was blinded to the CSF cfDNA results to determine presence or absence of leptomeningeal tumor dissemination on imaging.

CSF Collection

All CSF collections were performed either as part of standard clinical care, such as disease staging, or to follow-up on previously positive cfDNA results. Samples were submitted for clinical tests including cell profile, protein, glucose, cytology as clinically indicated. Aliquots ranging from 0.4 to 12 ml from the same procedure were collected in sterile body fluid containers or Streck Cell-Free DNA BCT tubes and submitted for molecular studies. Specimens received in sterile containers were processed immediately upon arrival, while those in Streck tubes were stored at room temperature and processed within 24 h of receipt. A blood sample was also obtained from all patients as matched normal control for sequencing whenever one had not been previously obtained for tumor sequencing.

Isolation of CSF cfDNA

Each CSF sample was separated from its cellular components using double centrifugation (10 min at 1,600 × g, 10 min at 3,000 × g at 22°C). cfDNA was manually extracted from the corresponding fluid compartment using MagMax cfDNA Kits per instruction (Thermo Fisher Scientific, MA).

Targeted Capture and Sequencing

All CSF cfDNA samples were subjected to molecular analysis using Memorial Sloan Kettering-Integrated Molecular Profiling of Actionable Cancer Targets (MSK-IMPACT) assay which is a Food and Drug Administration (FDA)/New York State Department of Health (NYSDOH) approved assay which captures all protein-coding exons of 468 cancer-associated genes and select introns.¹⁴ This assay has also been validated and approved by NYSDOH for use on cfDNA derived from CSF.¹⁵ Each CSF sample was paired with a genetically matched normal (germline) sample. Input cfDNA samples were not sheared prior to library preparation. The results for each CSF sample were compared to all available prior tumor sequencing results from the patient.

Our criteria for positive mutation calls and thresholds have been previously described.^14,15 Briefly, we use a two-tier system for mutation calling. Tier 1 are mutations that are considered hotspots based on extensive evidence in COSMIC, TCGA, and other literature sources. All other variants that do not occur in hotspots are considered Tier 2. Threshold for mutation calling is ≥2% variant allele frequency (VAF); mutant reads (AD) ≥ 8, Total Depth of Coverage (DP) ≥ 20 and ≥ 5% VAF; AD ≥ 10; DP ≥ 20 for tier 1 and 2, respectively.

We use a threshold of 150× on mean sample coverage to determine if a sample is sequenced to sufficient depth to allow the detection of mutations at this lowest limit. Samples are flagged as being at increased risk of false negatives if the mean coverage is below this level. In such cases, we can confirm any mutation that is present as a true positive but we cannot rule out a false negative result.

Given that cfDNA samples from the CSF are biologically enriched to contain a large proportion of DNA derived from tumor and not normal cells, even samples with very low coverage may still be positive for mutations. With coverages as low as 50× and even below, mutations that are present at 20% VAF can be detected with high confidence.

A sample was defined as “cfDNA positive” if there is at least one somatic variant detected on the MSK-IMPACT panel. These “calls” are based on the cutoffs described above which were established in the clinical pipeline for MSK-IMPACT. Negative samples in this cohort consisted of samples that were (1) negative (as defined by at least 50× sequencing coverage without detected somatic variants) and (2) technical failures when no variants were detected, and coverage was less than 50×. These represent our previously published criteria outlined in Bale et al.¹⁵ Quantification of cfDNA did not play a role in determining sample positivity.

A detailed technical description of the MSK-IMPACT assay can be found on the FDA’s website: https://www.accessdata.fda.gov/cdrh_docs/reviews/den170058.pdf

Statistical analysis

Patients’ and samples’ characteristics were described by median and range if continuous, or frequency and percentages if categorical. Characteristics were compared between positive and negative CSF cfDNA using Mann-Whitney-Wilcoxon (if continuous) or Fisher’s exact (if categorical) tests. At the sample level, samples were considered independent since the large majority of patients had only one sample, sample characteristics were only correlated to sample positivity (not overall patient characteristics), and only nonparametric tests were used.

Results

Patients

Sixty-four CSF samples from 45 unique pediatric and AYA patients with primary CNS tumors were obtained as part of clinical care. Twenty-five patients were pediatric (age < 15), and 20 patients were AYA (age 15–40; Table 1). 46/64 samples (71.9%) were collected via LP or VAD (Ommaya reservoir), 13/64 (20.3%) were collected during ventriculo-peritoneal (VP) shunt or VAD placement, and 5/64 (7.8%) were collected surgically.

Table 1.

Clinical Correlates of CSF cfDNA

Patient Characteristics (n = 45)
Demographics
Median age [years] (range)	14.4 (8 months–40 years)
Pediatric [n] (%)	25 (55.5)
Adolescent/Young Adult (AYA)	20 (44.4)
Male	30 (66.7)
Female	15 (33.3)
Pathological diagnosis
High-grade glioma	10 (22.2)
Medulloblastoma	10 (22.2)
Pineoblastoma	5 (11.1)
Low-grade Glioma	4 (8.8)
Diffuse leptomeningeal glioneuronal tumor	4 (8.8)
Retinoblastoma	4 (8.8)
Ependymoma	3 (6.6)
Other	5 (11.1)
Number of samples per patient
One sample	35 (77.8)
Two samples	6 (13.3)
Three or more samples	4 (8.9)
Sample Characteristics (n = 64)
Disease stage at the time of CSF collection
Newly diagnosed	16 (25.0)
Obtained at recurrence	21 (32.8)
Obtained during treatment	17 (26.6)
Obtained during surveillance	10 (15.6)
Disease status at the time of CSF collection
Disseminated disease	27 (42.1)
Localized disease	24 (37.5)
No evidence of disease	13 (20.3)
Prior therapy
Prior surgery	58 (90.6)
Prior treatment	46 (71.9)
Disseminated disease at the time of CSF collection
Leptomeningeal disease (LMD) on imaging	26 (40.6)
Positive CSF cytology	13 (20.3)
LMD on imaging or positive CSF cytology	27 (42.2)

Patient-level and sample-level characteristics are summarized.

() refer to percentages.

There was no statistical significance in the CSF cfDNA positivity rates of samples collected ventricularly as opposed to lumbar (P = .41). Histopathological subtypes included high-grade glioma (n = 10), medulloblastoma (n = 10), pineoblastoma (n = 5), low-grade glioma (n = 4), DLGNT (n = 4), retinoblastoma (n = 4), ependymoma (n = 3), and other (n = 5) (Table 1; Figure 1; Supplementary Table S1). Tumors were classified in accordance with the 2016 World Health Organization Classification of Tumors of the Central Nervous System.¹⁶

Fig. 1

Frequency of CSF cfDNA detection across various pediatric CNS tumor types.

Positive samples are shown in blue, negative samples in red.

The median patient age at the time of first CSF collection was 14.4 years (range 8 months–40 years) and 66.7% were male (Table 1; Supplementary Table S1).

Most samples were collected after initial treatment, that is, surgery (58/64), radiation therapy, and/or chemotherapy (46/64) (Supplementary Table S2).

cfDNA Yield, Genomic Coverage, and Sequencing Results

The volume of CSF obtained ranged from 0.4–12 ml with the great majority between 2.1 and 9.9 ml (10 samples were ≤2 ml and 5 samples were ≥10 ml; Supplementary Table S2). The median for both the CSF cfDNA positive and CSF cfDNA negative samples was 4 ml (P = .58; Supplementary Figure S2). Genomic coverage (sequencing depth) ranged from 2× to 1,368×. The coverage is related to the total DNA recovered from the sample. As expected, coverage was higher in the CSF cfDNA (+) samples (median = 381 × vs. 35×); P < .001; Supplementary Figure S2). The amount of CSF cfDNA extracted ranged from 0.00033 ng to 959.75 ng. We were able to obtain positive sequencing results (KIAA1549-BRAF fusion, patient #18) with input as low as 0.088 ng (total input CSF cfDNA). The concentration of CSF cfDNA in the mutation positive samples was significantly higher compared to the samples that were negative on MSK-IMPACT (median was 1.18 ng/ml vs. 0.067 ng/ml, P < .001; Supplementary Figure S2).

Somatic alterations were detected in 30 samples (46.8%). When accounting for serial testing, somatic alterations were detected in at least one sample per unique patient in 21/45 patients (46.7%). CSF positivity rates were as follows: pineoblastoma 5/5; DLGNT 4/4; high-grade glioma 7/10; medulloblastoma 3/10; low-grade glioma 0/4; retinoblastoma 1/4. Numbers were too limited to determine frequency rates across disease types, and this will be a focus of future research efforts (Figures 1 and 2).

Fig. 2

Oncoprint of tumor and CSF mutations across 21 pediatric patients with primary CNS tumors and positive CSF cfDNA.

Shown are the most frequent genomic alterations including single nucleotide variants (SNVs), copy number alterations (CNAs), and structural variants (SVs).

Timing of Collections and Clinical Characteristics

We evaluated whether timing of sample collection during disease course impacted the likelihood of cfDNA detection. Samples were divided into four categories: obtained (1) at diagnosis; (2) at recurrence; (3) during active treatment; (4) during surveillance. Within each category, we evaluated the number of samples that were CSF cfDNA (+) versus cfDNA (−). Rates of positivity were 8/16 (50%) at diagnosis, 12/21 (57%) at recurrence, 7/17 (41%) during treatment, and 3/10 (30%) during surveillance. There was no significant difference (P = .44) in the rates of CSF positivity across the different timepoints (Table 1; Supplementary Tables S2 and S4).

The most predictive factor associated with CSF cfDNA positivity was disseminated disease status at the time of collection. CSF cytology was performed on 56/64 samples; of these, thirteen samples were positive on cytology, twelve of which were also positive for CSF cfDNA (P = .001). Among the 26 samples taken from patients with leptomeningeal disease on imaging, 21 also had positive CSF cfDNA (Table 1; Supplementary Tables S2 and S4).

Overall, patients with disseminated disease at the time of CSF collection were far more likely to have detectable CSF cfDNA than patients with localized disease or no evidence of active disease: 22/27 versus 4/24 versus 4/13 samples respectively; P < .001 (Supplementary Tables S2 and S4). There was no statistically significant difference in positivity rates between lumbar versus ventricular CSF samples (Supplementary Table S4).

A wide range of molecular alterations was noted including single nucleotide variants, copy number alterations (CNAs), structural arrangements encompassing several disease-defining mutations, such as KIAA1549-BRAF fusions and chromosome 1p deletions (defining molecular alterations in DLGNTs), as well as a PTCH1 and CTNNB1 mutation in a patient with a Wnt-activated, TP53 wild-type medulloblastoma (Figure 2; Supplementary Table S3).

Of the 21 patients with detectable tumor-derived cfDNA, 15 had matching tumor tissue biopsies from diagnosis that were successfully sequenced, three patients’ tumors failed sequencing, and three surgeries were performed at outside institutions and samples were not submitted to MSK (Supplementary Table S3). Shared mutations between tumor tissue and CSF were seen in 14/15 cases (Figures 2 and 3; Supplementary Figure S1 and Supplementary Table S3). The one patient without shared mutation had repeated CSF cfDNA sequencing at two time points with two shared mutations between them (see Patient #01 in Supplementary Figure S1). Additionally, in 8/15 patients the CSF revealed private mutations not seen in the tumor representing either spatial heterogeneity or genomic evolution (Figure 3; Supplementary Figure S1; Supplementary Tables S3 and S5). We performed a concordance analysis across the 15 patients with matched tumor/CSF pairs to directly compare the mutational profiles. In instances in which there were multiple tumor and/or CSF samples that were profiled, the ones obtained in closest proximity to one another were selected. In the three disease types where, multiple samples were available (pineoblastoma, high-grade glioma, and medulloblastoma) we found that the shared mutation rate was 10/35 (28.6%) in pineoblastoma; 18/56 (32.1%) in high-grade glioma; and 7/33 (21.2%) in medulloblastoma (Figure 3; Supplementary Table S5).

Fig. 3

Concordance of mutations across tumor tissue versus CSF across matched pairs.

Shown is the frequency per patient of shared versus tissue-only or CSF-only mutations in matched tumor tissue—CSF sample pairs (N = 15 patient). Blue, shared; red, tissue-only; gray, CSF-only. Concordance analysis was performed with the tumor tissue and the CSF sample obtained in closest temporal proximity to each other.

There were three general categories where CSF cfDNA testing provided additional relevant diagnostic, prognostic, and/or therapeutic information, and impacted clinical assessment and decision making. Representative details for each category are provided below.

Category 1: Use of cfDNA to Establish an Initial Diagnosis and/or Identify Actionable Genomic Alterations in Patients With Surgically Inaccessible Tumors or Nondiagnostic Tumor Biopsy

In three instances, patients presented with symptoms of hydrocephalus and leptomeningeal enhancement on MRI scan (Supplementary Figure S3). We are describing one patient in detail.

Patient #19 (

Supplementary Figure S3)—A 3-year-old boy with progressive recurrent emesis over a 6-month period presented with altered mental status and bilateral lower extremity weakness. Based on MRI findings of communicating hydrocephalus and diffuse leptomeningeal enhancement and a positive CSF enterovirus antigen test, he was diagnosed with viral encephalitis. Due to persistent hydrocephalus, a VP shunt was placed 6 months later at which time the patient developed visual deterioration. Two years after the initial presentation, he underwent a biopsy of the leptomeninges, but pathology was inconclusive and clinical molecular testing failed, presumably due to low tumor cell content. His vision continued to gradually worsen in the next 18 months. As a final attempt to obtain a diagnosis, we performed a LP for CSF cfDNA testing, which revealed a KIAA1549-BRAF fusion consistent with the diagnosis of DLGNT. CSF cytology was negative. Based on this information, molecular targeted therapy with a MEK inhibitor was initiated and the patient remains stable both clinically and on imaging 3 months into treatment.

Category 2: Use of cfDNA to Monitor Tumor Response to Therapy

Several patients underwent clinical CSF collection at multiple timepoints, enabling us to track the relationship between CSF cfDNA status and disease status.

Patient #03 (

Supplementary Figure S4)—A 13-year-old girl was diagnosed with metastatic pineoblastoma and extensive leptomeningeal disease at diagnosis. At the time of diagnosis CSF cytology was positive, CSF cfDNA was detected at very high concentrations (27.45 ng/ml) and showed the following mutations: CREBBP P692T, DROSHA Q820*, PTPRS P359H. The leptomeningeal disease from the thoracic spine was biopsied and shared the CREBBP, DROSHA, and PTPRS mutations in addition to a private CREBBP loss.

She received standard therapy with craniospinal irradiation and adjuvant chemotherapy. End of treatment disease evaluation including MRI brain, spine, and CSF cytology showed no evidence of disease, however, CSF cfDNA remained positive, revealing the same mutations seen at diagnosis, however with a significant decrease in cfDNA concentration (27.45 ng/ml → 1.72 ng/ml).

The persistent presence of cfDNA in the CSF prompted closer clinical follow-up and subsequent CSF sampling (two, four, and eight months after completion of treatment). For the first 6 months after completion of treatment, while cytology and MRIs continued to be negative, CSF cfDNA remained positive showing the same mutational profile on each sample. The CSF cfDNA concentration (ng/ml) of the samples continued to decrease and was finally undetectable on her most recent collection (27.45 → 1.72 → 0.18 → 0.08 → 0.06 → 0), and she remains in clinical disease remission 34 months from initial diagnosis (Supplementary Figure S4).

Further examples of tracking MRD during and after treatment, as well as at time of recurrence, are illustrated in Supplementary Figures S5A–C.

Category 3: Use of CSF cfDNA to Monitor Tumor Evolution

Patient #08 (

Supplementary Figure S6)—A 17-year-old boy presented with episodes of déjà vu sense, smell alterations, and nausea followed by an R hemianopsia and was found to have a 5 cm enhancing mass in the medial temporal lobe and the left basal ganglia. He underwent gross total resection which revealed DMG, H3 K27M mutant. MSK-IMPACT was performed and showed the following mutations: ATRX V1550E, PPM1D L484*, and a NTRK2 AMP (in addition to the H3F3A K27M mutation). He was treated with chemoradiotherapy. Four months after completion of therapy, MRI brain revealed disease progression with possible leptomeningeal disease. LP was performed and revealed pleocytosis with elevated protein level, cytology was negative. CSF cfDNA showed the following shared mutations with his initial tumor: H3F3A K27M, ATRX V1550E, PPM1D L484*. Additionally, his tumor had evolved, and the following new mutations were identified in the CSF: PIK3CA H1047R, EZH1 V599M, PPRD V1029I, and an EVD1 rearrangement. He received Tumor Treating Fields therapy and bevacizumab, but his condition rapidly worsened and he passed away.

While the number of patients with each rare disease type remains limited, these data show the potential of CSF-based cfDNA testing to impact clinical decision making. Overall, in our cohort, 33.33% of the patients (7/21) directly benefited from CSF cfDNA analysis (Supplementary Table S1)

Discussion

CSF-based cytopathology is commonly used for disease staging, and typically performed soon after the initial surgery or biopsy of a malignant primary CNS tumor. As a purely morphologic diagnostic tool, cytopathology has limited sensitivity and specificity, and generally does not allow for molecular testing. Given these limitations, CSF-based liquid biopsy has emerged as a novel molecular diagnostic tool, first in the research setting.¹⁷ At MSK, we recently began integrating CSF cfDNA sequencing into routine clinical care of brain tumor patients. Here, we summarize our experience of the successful implementation of CSF-based liquid biopsy program into the care of pediatric and AYA brain tumor patients.

We have identified three main categories in which detecting CSF cfDNA enhanced our understanding of tumor biology and impacted clinical decision making: (1) establishing a diagnosis and/or identification of actionable genomic alterations in patients with surgically inaccessible tumors or nondiagnostic tumor biopsy, (2) monitoring tumor response to therapy, and (3) tracking tumor evolution.

According to prior studies, the likelihood of detecting cfDNA in the CSF is influenced by several factors including tumor type, tumor burden, presence of leptomeningeal disease, proximity to the ventricular system, timing of the test, and site of sampling.^2,5,11 In line with these data, our patients who had leptomeningeal disease at the time of CSF collection were far more likely to have detectable CSF cfDNA (81.5% vs. 18.5%). This number increased to 92.3% when cytopathology was positive. We did not detect any difference in positivity between samples obtained via an Ommaya reservoir tap, LP, or ventricular sampling during surgery.

Since the sample collection for CSF cfDNA analysis was tied to clinically indicated procedures, the timing of testing was not always optimal for the detection of tumor-specific mutations in a given patient, that is, at the time of original diagnosis or recurrence. In our patient population, we only perform diagnostic LPs in a subset of entities as part of disease staging, including embryonal tumors (medulloblastoma, pineoblastoma, atypical teratoid/rhabdoid tumor), ependymoma, and germ cell tumors, generally within several weeks after surgical resection. The average length of tumor-derived DNA is about 145 bp and undergoes rapid degradation by nucleases.^18,19 Among our cohort, we observed only one patient with a medulloblastoma who had a positive CSF cfDNA sample at the time of diagnosis following a gross total tumor resection and having no evidence of leptomeningeal disease based on CSF cytology and imaging. This observation is consistent with recently published retrospective data on a clinical trial population, where CSF cfDNA testing was negative in 46% of non-metastatic patients after surgery.⁹

Spatial heterogeneity of transcriptional and genetic markers is a well-known feature of many pediatric CNS tumors.^20,21 Due to this spatial heterogeneity, analysis of multiple different specimens from the tumor can uncover patient-specific patterns of cancer evolution, while simultaneously exposing the pitfalls of making assessments and treatment decisions based on the histopathology and genomic analysis of a single sample. We believe that CSF-based liquid biopsy, which contains tumor-derived DNA from all parts of the tumor may reduce the risk of sampling bias.¹³ We were able to perform de novo mutation calling in the cfDNA samples using MSK-IMPACT, which is a distinct advantage over ddPCR and other narrow targeted techniques that require prior knowledge of the mutations present in the tumor. Technically, MSK-IMPACT is best utilized for short nucleotide variant and insertion/deletion calling. CNAs, structural variants and gene fusions, as well as microsatellite instability are also reliably called, although their detection is limited by sample purity. While MSK-IMPACT includes a relatively large panel of protein-coding exons of over 450 cancer-associated genes and select introns, detection is inherently limited to genomic alterations covered by the gene panel. Among the 15 patients with positive CSF cfDNA and matching tumor tissue, we identified 14/15 instances where there was a shared somatic mutation between the tumor and the CSF (Figure 3; Supplementary Table S3). Additionally, in 8/15 patients the CSF revealed additional mutations not seen in the tumor (Supplementary Figure S1; Supplementary Tables S3 and S5).

We believe that many of the private alterations in the tumor tissue and CSF are representing either tumor evolution or spatial heterogeneity. Wu et al. have shown that leptomeningeal metastases of medulloblastoma from a single human or mouse are genetically highly divergent from the matched primary tumor.²² Metastases arise from a restricted subclone of the primary tumor through a process of clonal selection, and we observed several examples of genomic evolution in our dataset. For example, a CSF sample from a pineoblastoma case (patient #01) revealed a private mutation of ATR (ataxia telangiectasia and Rad3-related), a key molecule of genome maintenance and DNA-damage response. In the high-grade glioma patient #08, CSF cfDNA showed an ETV1 fusion, which was not found in the tumor sample. ETV1 is a member of ETS (erythroblast transformation specific) family of transcription factors and a well-known oncogene. In the same CSF sample, we also found a PIK3CA mutation, which can be considered therapeutically actionable.

While we believe that the alterations seen in the CSF are reflective of genomic evolution, we recognize and acknowledge that there is limited sensitivity in the CSF to detect the full compilation of genomic alterations especially for copy number level alterations. Hence, we acknowledge that an “absent” alteration in the CSF may not be truly “absent.” We found that the assay is highly specific, and we have not found any suggestion of false positive mutations in the CSF. However, some alterations seen in the tumor tissue may be below the level of detection in the CSF due to limited amounts of CSF cfDNA.

In line with recent findings from monitoring MRD in CSF of medulloblastoma patients undergoing upfront therapy,⁹ residual MRD during and even after adjuvant chemotherapy in some patients does not necessarily imply disease recurrence, as illustrated in patients #01, #08, and #13. As a result, further research will be necessary to better delineate the predictive value of CSF cfDNA positivity at different time points in the context of different tumor types and therapies.

Molecular profiling has transformed CNS tumor diagnostics and led the way to personalized therapeutic approaches. Incorporating precision medicine into the clinical care of pediatric neuro-oncology patients has had major impact, demonstrating its utility in about 40%–60% of patients according to institutional experiences.^23,24 Based on recent publications, fewer than 10% of the samples fail sequencing, most often due to insufficient DNA quantity and quality, resulting from low tumor cellularity or viability in the tissue.^23–25 We encountered four patients where genomic testing of available tumor tissue failed, but tumor-derived cfDNA was found in the CSF allowing to establish a diagnosis and identify targetable alterations (Figure 2; Supplementary Figures S1 and S3). Representative examples are four DLGNT patients in our series, where detection of the KIAA1549-BRAF fusion not only led to a diagnosis, but also revealed a molecular targeted therapeutic option in the form of a MEK inhibitor. This observation suggests that CSF cfDNA analysis is of particular value in tumors presenting with leptomeningeal dissemination, where biopsies often yield small, nondiagnostic samples with low tumor cellularity and purity.

Based on our initial experience, the evidence for the added clinical value of CSF-based liquid biopsy in pediatric/AYA neuro-oncology patients is compelling and supports its more widespread routine clinical implementation. We anticipate that ongoing technical advances in sequencing technologies are likely to further increase the sensitivity and therefore the clinical utility of this novel diagnostic tool in the near future.

Acknowledgments

We would like to acknowledge the MSK Kids Pediatric Translational Medicine Program (PTMP) and the Director of the PTMP, Dr. Neerav Shukla. We gratefully acknowledge the members of the Molecular Diagnostics Service in the Department of Pathology.

Funding

This work was supported in part by the Matthew Larson Foundation for Pediatric Brain Tumors through an IronMatt research grant awarded to AMM, the Marie-Josée and Henry R. Kravis Center for Molecular Oncology and the National Cancer Institute Cancer Center Core Grant P30 CA008748. We also thank Cycle for Survival, the National Brain Tumor Society (Defeat GBM Initiative) for providing research support (I.K.M.) and The Scarlett Fund for their support to the PTMP.

Conflict of interest statement. A.M.M., L.S., N.B., H.A., K.H., J.R., A.J.L., M.I.R.-S., O.Y., A.P., T.A.B., J.K.B., M.D., S.W.G., Y.K., J.P.G., M.M.S., S.H., A.M., M.F.B. have nothing to disclose. R.B. has received a grant and travel credit from ArcherDx, honoraria for advisory board participation from Loxo Oncology and Roche Dx andspeaking fees from Illumina. M.A. has received consulting and speaker fees for Biocartis, Invivoscribe, and AstraZeneca. I.J.D. serves as a consultant and on the advisory boards for Apexigen, AstraZeneca, Bristol Myers Squibb/Celgene, Day One, Fennec, QED Therapeutics, and Roche. K.K. is a paid consultant and has financial interest in YmAbs Therapeutics, Inc. S.F.S. has received consulting and personal fees from AstraZeneca and consulting fees from QED Therapeutics. I.K.M. has received honoraria from Roche, serves as a consultant or in an advisory role for Black Diamond Therapeutics, Debiopharm Group, Puma Biotechnology, Voyager Therapeutics, DC Europa LTD, Kazia Therapeutics, Novartis, Cardinal Health, Roche, and Vigeo Therapeutics, has received research funding from Amgen, General Electric, Lilly, Kazia Therapeutics, and has received travel accommodations from Voyager Therapeutics and AstraZeneca. M.A.K. received consulting and personal fees from AstraZeneca, Bayer, CereXis, QED Therapeutics, and Recursion Pharma.

Authorship statement. Study Design: A.M.M., L.S., I.K.M., M.A.K. Sample and/or data acquisition: A.M.M., L.S., N.B., K.H., H.A., J.R., A.J.L., M.I.R.-S., O.Y., A.P., T.A.B., J.K.B., R.B., M.A., M.D., I.J.D., S.W.G., Y.K., K.K., S.F.S., J.P.G., M.M.S., S.H., A.M., M.F.B., I.K.M., M.A.K. Data analysis: A.M.M., L.S., J.R., A.P., T.A.B., R.B., M.A., A.M., M.F.B., I.K.M., M.A.K. Manuscript writing, revisions and approval of the final version: all authors.

Author notes

These authors contributed equally to this work.