Introduction

Neoplasms of the central nervous system (CNS) represent the most common form of solid tumor of childhood-onset [1]. Advances in diagnostics, neurosurgical techniques, and the application of multimodal adjuvant therapy have led to improved patient survival over the past decades [2, 3]. Nonetheless, mortality related to pediatric CNS tumors remains the highest among all childhood cancers, and survivors frequently experience long-term toxicities such as neurocognitive and endocrinologic deficits [4]. Unlike in patients with leukemia, where markers of molecular residual disease (MRD, previously minimal residual disease) based on patient-specific disease clone allows the objective and sensitive determination of disease status, current tracking of response for CNS tumors relies predominantly on radiographic studies supplemented by CSF cytologic evaluation [5, 6]. Serial MRD measurements effectively identify good responders who demonstrate early disease clearance, and also poor responders whose disease remains refractory despite therapy. Molecular relapse can be detected prior to clinical/hematologic relapse, which allows salvage therapy to be initiated at a lower disease burden. Therefore, there is an urgent need for developing equivalent biomarkers for childhood CNS tumors.

The past decade has witnessed rapid advances in our understanding of oncogenic mechanisms that underly CNS tumors [7]. This has been made possible by the increasing availability of array- and massive parallel sequencing-based–platforms for studying the tumor genome, transcriptome, and epigenome [8,9,10,11,12,13,14,15,16,17]. These techniques have facilitated the unveiling of tumorigenic driver alterations, intertumoral disease heterogeneity, as well as prognostically important molecular signatures. The 2021 World Health Organization CNS Tumor Classification, for example, has refined its definition of entities based on characteristic molecular events and emphasized the utility of methylation arrays in integrative diagnostics [18]. In parallel, there has been a surge of studies that demonstrate the feasibility of detecting circulating tumor markers as biomarkers [19,20,21,22]. These include circulating tumor cells (CTCs), cell-free DNA (cfDNA), cell-free RNA (cfRNA, including mRNA and small RNAs), and extra-cellular vesicles (EVs) that can be detected in various body fluids. Potential applications in non-invasive diagnostics, response monitoring, and even as tools for cancer screening are topics for ongoing studies, with increasingly mature data being generated from adult-onset carcinomas. Coupling these breakthroughs in oncology research, it is likely that novel biomarkers will become available for use in the pediatric clinical setting for patients with CNS tumors to complement the current strategies for tumor diagnosis and monitoring. Here, we review the current status and notable challenges in the use of cfDNA profiling for childhood CNS tumors.

cfDNA in clinical medicine

Among the aforementioned circulating tumor markers, cfDNA carries the advantages of being relatively stable ex vivo and technically less demanding in isolation [19]. In contrast to high molecular weight genomic DNA (gDNA), cfDNA typically measures 150–200 base pairs and is thought to be released after non-random fragmentation during the apoptotic process, necrosis, or through active secretion by live cells [23,24,25]. The physiological presence of cfDNA has been best studied in blood and is contributed predominantly by the hematopoietic system [26]. Extra-cellular DNA fragments in circulation were first described in 1948 [27], predating the publication on DNA structure by Watson and Crick [28]. Subsequently, levels of plasma cfDNA have been shown to be elevated during physiologic and pathologic processes such as pregnancy, autoimmunity, trauma, infection, and cancer [27, 29,30,31,32]. Using radioimmunoassay, cfDNA concentration was determined to be higher in patients with cancer than those without, with higher levels in those with metastasis and reduction in levels with treatment [32]. Nonetheless, significant overlap was observed in cfDNA concentration between patients and healthy individuals; however, at that time, delineation between cfDNA from healthy and cancerous cells was impossible.

In the 1990s, landmark studies further supported the potential of cfDNA profiling in the field of oncology by demonstrating feasibility in genotyping oncogenic mutations such as NRAS mutations in myelodysplastic syndrome or acute myeloid leukemia and KRAS mutations in pancreatic or colorectal cancers [33,34,35]. Nonetheless, these findings have garnered limited interest, and the subsequent decade of cfDNA research has focused on development of the revolutionary non-invasive pregnancy testing, following the demonstration of fetal cfDNA in maternal serum [36]. Non-invasive prenatal testing has since become the standard of care, allowing early and risk-free determination of germline chromosomal aneuploidies and other genetic syndromes [37]. Additional applications include the use of microbial cfDNA for rapid diagnosis of pathogens [38, 39] and as a potential diagnostic or prognostic indicator for patients with stroke as well as myocardial infarction [40,41,42]. With advances in the sensitivity, throughput, and efficiency of sequencing platforms, as well as our knowledge on the genomic landscape of common human cancers [43, 44], there has been a resurgence in the interest in studying the somatic cancer signature carried by tumor-derived cfDNA or circulating-tumor DNA (ctDNA). The capability to reconstruct tumor genomes through non-invasive profiling of ctDNA opens up new avenues for cancer management and research.

The era of liquid biopsies

Liquid biopsy was originally coined by Pantel and Alix-Panabières in 2010 to the study of CTCs in blood samples and disseminated malignant cells in the marrow of cancer patients [45]. Since then, the term has been more liberally adopted in reference to the study of circulating tumor material, contrasting such approaches with the gold standard of tissue biopsy [21]. This encompasses the use of various body fluids, such as urine and cerebrospinal fluid, in addition to blood, as well as an array of wet-lab strategies to profile respective tumor cell metabolites and the associated bioinformatic techniques required to interpret the vast amount of data generated. Somatic alterations in tumor DNA, such as single nucleotide variants and small indels, copy-number changes, focal amplifications, loss of heterozygosity, and methylation patterns can be recapitulated in the study of ctDNA. The evidence stemming predominantly from adult-onset carcinomas has established the complementary role of liquid biopsy in the clinical setting [46, 47]. Despite the fact that surgical resection remains an integral component of treating solid tumors, allowing access to tumor tissue for histologic and molecular evaluations and studying circulating tumor material has some advantages. First, liquid biopsy can be performed non-invasively in a serial manner for the assessment of treatment response, thereby effectively representing a patient-specific tumor marker that might supersede current biochemical markers. Second, assessment of tumor genotype in the relapsed setting, which might not be pursued if a tissue biopsy is required, yields critical information for the identification of molecular targets and mutational burdens that might evolve with prior therapy. Third, assessment of tumor genomics through tissue biopsy is hampered by intra-tumoral spatial heterogeneity, which on the contrary might be effectively captured by the study of circulating tumor material. Furthermore, in cases where tumor resection or sampling is either extremely risky or does not add therapeutic benefits, liquid biopsy may circumvent the need for surgical procedures done purely for diagnostic purposes.

The value of cfDNA profiling has been exemplified by recent prospective studies on adult-onset oncologic conditions such as lung, breast, colon, gastric, and bladder cancers [47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62]. These protocols incorporate serial targeted tracking of somatic oncogenic alterations in plasma cfDNA as a molecular tumor marker, allowing response monitoring, surveillance for relapse, prognostication, and prediction of response to immunotherapy or targeted agents. In 2016, the first liquid biopsy assay was approved by the Food and Drug Administration (FDA) [63]. The Cobas assay (Roche) adopts real-time PCR for the detection of deletions and hotspot mutations in EGFR, which in turn confer sensitivity to the tyrosine kinase inhibitor erlotinib in patients with metastatic non-small cell lung cancer. This was followed by approval of two NGS-based plasma cfDNA assays (Guardant360, Guardant Health and FoundationOne Liquid CDx, Roche) in 2020, focusing on the identification of somatic alterations in patients who might benefit from targeted therapies as “companion diagnostics.” Expanding beyond the roles in patients with an established cancer diagnosis, further studies are underway to examine the cost-effectiveness of using cfDNA profiling as a cancer screening tool in asymptomatic individuals [64,65,66]. Few commercial assays are under development (CancerSEEK [ClinicalTrial.gov identifier: NCT04213326], GRAIL - [ClinicalTrial.gov identifier: NCT03085888]), focusing on early detection.

Unique challenges for patients with primary CNS tumors

The application of cfDNA profiling for patients with primary CNS tumors requires special considerations (Fig. 1). First and foremost, the concentration of cfDNA derived from CNS tumors is much lower than that from other solid tumors [67, 68]. This appears to be influenced by tumor burden and the extensiveness of the tumor–CSF interface. This poses practical challenges for downstream processing, such as the generation of NGS libraries with adequate complexity. Nevertheless, CSF has been shown to be enriched with tumor-derived cfDNA in comparison with other body fluids such as plasma or urine in these patients [67, 69, 70]. This is likely the result of the blood-brain barrier, but could also be a dilutional effect from non-tumor derived cfDNA physiologically present in the systemic circulation. Importantly, profiling of cfDNA from the centrifuged CSF supernatant has been demonstrated to be superior to the study of gDNA from the cell pellet in detecting tumor alterations [71]. There is thus a need to determine how CSF collection could be systematically incorporated into the management and monitoring of patients with CNS tumors. Beyond intra-operative sampling and collection through ventricular shunts or reservoirs, lumbar puncture (LP) is a minimally-invasive procedure for the collection of CSF. Despite the relative intrusiveness, LPs are routinely performed for staging and disease evaluation for childhood CNS tumors, in particular CNS embryonal tumors [72]. Cytologic examination is associated with relatively low yield, and the availability of cfDNA profiling thus offers an opportunity to optimize the use of such previous clinical samples. Several CNS tumor-focused studies have emerged to testify the feasibility and utility of the study of cfDNA in CSF [73,74,75,76,77,78,79].

Fig. 1: Summary of technical considerations.
figure 1

The application of cell-free DNA profiling for children with primary central nervous system tumors requires the optimization of sample handling and experimental pipeline, inclusive of wet and dry lab components.

Full size image

Diffuse glioma is the most common malignant brain tumor in adults. Somatic EGFRvIII truncating mutant, an oncogenic event in glioblastoma, was first shown to be detectable in plasma cfDNA by PCR amplification, albeit in a small proportion of participants (3 of 13) [75]. Adopting Guardant360, which targets the coding sequence of 54 cancer-associated genes from plasma cfDNA, one or more alterations were detected in 9 of 33 patients with glioblastoma [76]. With comparative studies showing better performance of CSF over plasma for CNS tumors, subsequent studies focused on the profiling of cfDNA derived from CSF samples. Martínez-Ricarte and colleagues designed amplicon sequencing and digital droplet PCR (ddPCR) assays targeting 7 genes (IDH1, IDH2, TP53, TERT, ATRX, H3F3A, and HIST1H3B) that are key to the classification of diffuse gliomas [79]. Paired tumor-CSF analysis in 20 patients with diffuse gliomas showed that tumor-specific mutations can be captured in 17 of them (inclusive of all 10 with glioblastoma). In addition to the study of mutations, low-coverage whole-genome sequencing (lcWGS) at <0.4× was sufficient to study somatic copy-number variations (CNVs) from CSF-derived cfDNA, which was discernable in 5 of 13 patients with diffuse glioma [77]. Based on the MSK-IMPACT targeted-sequencing assay, Miller and colleagues further demonstrated the value of longitudinal cfDNA profiling in 85 patients with gliomas [78]. Tumor-derived cfDNA was detectable in CSF in 49% of patients and correlated with disease burden and prognosis. Genomic evolution resulted in divergence between alterations observed in tumor-CSF pairs. With disease cure being largely impossible for patients with glioblastoma, it is likely that cfDNA analysis would be of greatest value for complementing histologic evaluation in the understanding of therapeutic vulnerability and mechanisms of resistance.

Technical considerations in cfDNA studies

Until recently, the available technology did not allow the robust detection of tumor-derived cfDNA, in particular for patients with CNS tumors. As alluded to previously, this may be attributed to inherent properties such as the low quantities of cfDNA, its fragmented nature, and contamination by normal cell-derived cfDNA and gDNA. Apparently trivial mishaps such as traumatic LPs could introduce non-tumor cell-derived cfDNA present in the systemic circulation, which may then interfere with the study of tumor-derived cfDNA in CSF due to the similarity in fragment sizes. Much work remains to be conducted for the optimization and standardization of cfDNA profiling as a clinical assay [80,81,82,83,84]. This includes considerations for the initial steps of cfDNA purification (preferred source biofluid, method of sample collection and storage, sample pre-processing, procedure for cfDNA extraction, quality control measures) and the downstream wet and dry lab analytical pipelines for detection of tumor signatures.

Most benchmarking studies and consensus recommendations have been derived for cfDNA extraction from blood samples. Maas and colleagues recently published their rigorous evaluation of pre-analytical factors for liquid biopsy, including the use of CSF as input material [80]. Compared to plasma, CSF specimens are often more restricted in volume, relatively acellular, and carry lower amounts of cfDNA in the physiological state. Comparison among a few cell-stabilizing tubes (Norgen, PAX, Streck) that allow storage at room temperature, and the use of Eppendorf tubes for storage without preservatives at 4 °C suggest highest recovery and purity of cfDNA in Norgen tubes (supplemented with phosphate buffered saline per protocol), followed by Eppendorf tubes after a 7 day duration. Such findings are particularly relevant when study logistics require= shipment between institutions, thereby precluding immediate centrifugation and banking. For cfDNA isolation, silica membrane columns have been shown to be superior in cfDNA recovery to magnetic bead–based protocols, with the latter having the advantage of being automatable [85, 86]. The performance also varies among silica column kits, given additional practical differences such as hands-on time and limiting sample volume. In terms of quality control, fluorometric measures could be applied (Qubit preferred to PicoGreen), but qPCR-based measurements would be superior in case of lower cfDNA concentrations, as in the case of CSF-derived material [87]. Using primers amplifying ALU sequences at two amplicon sizes (115 bp and 247 bp), qPCR could also offer information on the ratio between short-fragment cfDNA and high-molecular-weight gDNA, which otherwise should be profiled using capillary electrophoresis. In samples wherein gDNA contamination is significant and dependent on the assay to be adopted downstream, magnetic bead-based size selection may be adopted to reduce the proportion of large DNA fragments.

Unlike the pre-analytic components, sequencing assays for cfDNA will likely evolve at a much quicker pace (Table 1). Tolerance to low quantities and the fragmented nature of cfDNA is key to the success of building sequencing libraries, and the use of molecular barcodes allow PCR artefacts to be discerned from genuine somatic mutations occurring at low allele frequencies [88]. The application-specific to pediatric CNS tumors is further reviewed in the next section.

Table 1 Comparison of molecular assays for detection of tumor signature in cell-free DNA.
Full size table

Experience in pediatric CNS tumors

The spectrum of childhood CNS tumors differs from its adult counterparts. CNS embryonal tumors, diffuse midline glioma, and pilocytic astrocytoma, for example, represent entities that are predominantly seen in young patients [89]. In addition, with the tendency of CNS embryonal tumors (including medulloblastoma) to metastasize along the leptomeninges, higher concentration of tumor-derived cfDNA is expected in the CSF. Despite its heterogeneity, the overall outcomes of children with CNS tumors are superior to those of adults, where glioblastoma and extra-axial solid tumor metastasis to the CNS predominate. This, together with the need to mitigate chronic health deficits associated with adjuvant chemo-irradiation, suggests greater potential for the introduction of a molecular biomarker in the stratification of patients according to treatment response. There is a growing body of literature to support the application of cfDNA profiling for childhood brain tumors (Table 2) [68, 73, 79, 90,91,92,93,94,95,96,97,98]. The evidence is summarized by the disease entity below.

Table 2 Summary of key published cfDNA studies for CNS tumors in children.
Full size table

Diffuse midline glioma

Diffuse midline glioma, previously referred to as diffuse intrinsic pontine glioma, is a catastrophic diagnosis that is almost always associated with universal lethal outcome. This tumor is characterized by infiltrative nature and involvement of vital CNS structures, and is not amenable to meaningful surgical resection. While in most cases the diagnosis can be made by a combination of typical clinical and radiographic features, histologic and molecular confirmation remains relevant in a subset of patients with atypical features and in the context of clinical trial enrollment [99]. Surgical biopsy, while feasible, carries an inherent risk and can be conducted only in centers with adequate volume and expertise. Furthermore, rebiopsy is not typically pursued at the time of relapse and MRI features might not be distinguishable between treatment effects and disease progression. The majority of diffuse midline gliomas are driven by somatic hotspot histone H3 alterations in the H3.3 or H3.1 genes, offering molecular markers that can be used as a circulating-tumor signature [100, 101].

The feasibility of detecting histone mutations in CSF of children with diffuse midline gliomas was first demonstrated by Sanger sequencing and nested PCR using mutation-specific primers, with the choice of platform based on the quantity of cfDNA isolated [90]. Using ddPCR and customized genotyping assays, H3K27M mutations could be detected from CSF in a more sensitive manner, with concomitant quantification of allele frequencies [79, 91, 102, 103]. Panditharatna and colleagues applied the ddPCR assay on CSF and plasma from 48 patients with diffuse midline glioma [91]. This landmark study confirms the detectability of histone mutations in 87% of CSF samples from patients whose tumors are known to harbor such mutations. Comparison of paired CSF and plasma samples confirms a similar detectability but higher mutant allele frequency in CSF specimens. Such enrichment is further enhanced for CSF samples obtained from ventricular spaces. Based on serial plasma samples from children with diffuse midline glioma, cfDNA dynamics in terms of changes in mutant allele frequencies mirrored that of MRI-based response and subsequent progression. Importantly, the assay could also be multiplexed to allow detection of associated genomic targets of interest, such as the activating R206H mutation in ACVR1. The increasing evidence for spatial heterogeneity and clonal evolution with therapy implies that cfDNA assays need to incorporate a broader spectrum of genomic targets and events, which would be of value in longitudinal studies [92, 104, 105].

With further standardization of the experimental pipeline, including optimization of primer-probe designs, enhancing sensitivity by vacuum concentration of cfDNA and pre-amplification before ddPCR, and consideration of the ddPCR platform adopted [68, 103], cfDNA-based molecular analysis could possibly be undertaken in lieu of tissue studies in future patients and clinical trials for diffuse midline glioma [106].

Medulloblastoma

Medulloblastoma is the most common malignant CNS tumor of childhood [89, 107]. It represents a CNS embryonal tumor arising from the posterior fossa, with a tendency for leptomeningeal metastasis. Its intertumoral heterogeneity has been extensively studied by multi-omic studies, resulting in the recognition of four epigenomic groups (WNT-activated, SHH-activated, Group 3, Group 4) with differences in oncogenic mechanisms and clinical behaviors [107]. Unlike the case of diffuse midline glioma, patients with medulloblastoma often present with raised intracranial pressure requiring urgent upfront tumor removal, which is both therapeutic and prognostic (according to completeness of resection). Adjuvant therapy for medulloblastoma is integral to durable disease control and includes craniospinal irradiation and combination chemotherapy in non-infant patients and radiation-sparing chemotherapeutic strategies in younger children. Despite advances in tumor-directed therapies and supportive care over the past decades, progression-free survival has been achieved only in two-third of patients, with abysmal outcomes in those who experience relapse.

To mitigate toxicities associated with chemo-irradiation, treatment intensity is conventionally stratified by clinical risk features, namely, extent of resection and metastatic status, with risk-grouping strategies that incorporate tumor molecular features being evaluated in several ongoing clinical trials (SJMB12, ACNS1422, PNET 5) [11]. Response monitoring and disease surveillance after treatment completion are based on contrast-enhanced magnetic resonance imaging and complemented by CSF cytologic examination. cfDNA profiling offers an opportunity to build in a marker of MRD for response-based treatment personalization and for surveillance of early disease relapse. Nonetheless, the low disease burden after initial resection and the absence of recurrent mutations restrict the applicability of established liquid biopsy pipelines in patients with medulloblastoma.

Wang and colleagues offered proof-of-principle by demonstrating the detectability of tumor-specific mutations in cfDNA obtained from intra-operative CSF for 6 medulloblastoma patients. using amplicon-based panel sequencing [73]. Such findings, however, could not be immediately translated into practice, as most diagnostic specimens are instead collected post-operatively and serially through LPs [72]. Similarly, Escudero and colleagues showed that tumor-specific mutations were detectable in cfDNA from pre-operative CSF by ddPCR in 10/13 medulloblastoma patients, in contrast to detecting them in only 1 of the matched blood samples [97]. Furthermore, whole-exome sequencing was adopted on tumor and CSF samples from 4 patients, indicating that most mutations present in the tumors could be detected in the corresponding CSF-derived cfDNA samples, with correlation in the mutant allele frequencies demonstrated. Further analysis of mutations on paired tumor and cfDNA samples from 9 patients indicated divergent profiles at recurrence, supporting genomic evolution with treatment and disease progression [98].

Recently, our group profiled serial CSF samples collected from 123 children with medulloblastoma treated on the risk-adapted trial SJMB03, aiming to study somatic chromosomal CNVs as a surrogate for tumor-derived cfDNA [51, 108]. By extracting cfDNA from 476 CSF samples (median volume at 1 ml) collected longitudinally after tumor resection through treatment and subsequent follow-ups, we confirmed low cfDNA concentration in the CSF of children with medulloblastoma. Nonetheless, genome-wide copy-number profiles could be derived through lcWGS and tracked through serial specimens as an MRD marker. Correlating with clinical and radiographic findings, the study suggested the utility of cfDNA profiling in assessing tumor burden, stratifying patients according to outcome, predicting disease relapse, and studying tumor evolution.

In addition to studying mutations and CNVs, methylation signatures carried by cfDNA could potentially be exploited as a marker to detect which are tumor and subgroup-specific and could potentially be exploited as an indicator for disease presence, using platforms such as whole-genome bisulfite sequencing, reduced representation bisulfite sequencing and anti–cytosine-5-methylenesulfonate immunoprecipitation sequencing (CMS-IP–seq) [96, 109].

Rare tumors and target-driven analysis

Anecdotal data are available on the use of cfDNA studies for other pediatric CNS tumors, with applicability dependent on the presence or absence of driver events, as well as concentration of cfDNA in CSF or plasma [108, 110, 111]. In our experience of studying diagnostic CSF samples for of children with CNS embryonal tumors, including pineoblastoma, atypical rhabdoid/teratoid tumor (ATRT), and embryonal tumor with multilayered rosettes, tumor-specific CNVs were detectable in 75% of cases [108]. In parallel, experience from MSKCC and Seattle Children’s Hospital suggested the feasibility of mutational profiling for CNS embryonal tumors including pineoblastoma and ATRT [110, 111]. On the other hand, cfDNA assays can be designed to capture specific hotspot mutations such as BRAF V600E, which is seen more frequently in pediatric low-grade gliomas, and might be applied in a histology-agnostic manner [94].

Future directions and conclusion

The study of cfDNA holds great promise in revolutionizing patient care in pediatric neuro-oncology. It is expected that more prospective therapeutic trials will include the use of cfDNA assays, thus yielding increasing evidence for its routine clinical use. This experience will also help determine optimal pre-analytical and analytical workflows, with specific relevance to CSF use based on performance, reproducibility, and versatility. Similarly, the dynamics of cfDNA levels in response to tumor-directed therapy and other systemic factors should be studied in an entity-specific manner. Novel strategies to enhance the shedding of cfDNA, such as by focused ultrasonography, might increase the detectability of tumor-derived material in blood samples [112]. Analogous to the epigenetic classifier based on tumor DNA, it is likely that a cfDNA-based reference for CNS tumors can be designed for non-invasive molecular diagnosis [8, 80, 96, 109]. In addition to studying mutations and CNVs, NGS data from cfDNA could lend further insight into nucleosomal occupancy and chromatin accessibility, which might prove to be further tools to delineate between tumor and non-tumor material, which can likely be achieved by machine-learning algorithms [113, 114].

In summary, studies on the liquid biome might redefine our approach to managing childhood CNS tumors as well as our understanding of the associated genomic landscape. Ultimately, an integrative liquid biopsy strategy that combines cfDNA studies with other circulating tumor analytes might prove to be the “Holy Grail” of non-invasive diagnostics.