Targeted Analysis of Cell-free Circulating Tumor DNA is Suitable for Early Relapse and Actionable Target Detection in Patients with Neuroblastoma

This article is featured in Highlights of This Issue, p. 1743

The molecular landscape of solid tumors is currently assessed using DNA and/or RNA extracted from tissue samples. Endogenous and exogenous pressures, however, cause these molecular profiles to dynamically evolve over time. Competition among the heterogeneous genetic background of multiple subclonal populations and stress exerted on tumor cells by conventional cytotoxic chemotherapy and targeted therapies represent major processes driving tumor evolution (1). Increasing evidence suggests that recent technical advances improving sensitivity and accuracy of detection and characterization of total cell-free DNA (cfDNA), RNA, and/or circulating tumor cells in liquid biopsies could allow clinicians to noninvasively monitor tumor evolution by multiple longitudinal testing (2). Liquid biopsy–based diagnostic approaches have begun to be incorporated into routine disease monitoring for the first cancer entities afflicting adult patients (3, 4). A large number of clinical trials currently evaluate circulating tumor DNA (ctDNA) diagnostics for further adult cancer entities, addressing a variety of observational and interventional research questions, including using liquid biopsy–based findings to prompt therapeutic actions (5). Liquid biopsy applications for pediatric oncology lag behind their adult counterpart, with predominantly retrospective proof-of-concept studies in small cohorts so far (6).

The pediatric tumor, neuroblastoma, originates from neuroectodermal progenitor cells and is the most frequent extracranial solid tumor in infancy and childhood (7). Approximately half of all newly diagnosed neuroblastomas are designated high-risk for relapse (8). Multimodal therapy including induction, surgery, high-dose chemotherapy followed by autologous stem cell rescue, radiation, and anti–GD2-directed mAb-based immunotherapy provokes a good initial response. However, minimal residual disease with few disseminated resistant tumor cells frequently causes tumors to relapse (9). Treating refractory or relapsed neuroblastoma remains challenging (10). Despite advances made in international efforts, minimal residual disease monitoring must be improved and therapy resistance must be detected earlier at any time during therapy. Monitoring minimal residual disease in peripheral blood or bone marrow supports more frequent diagnostic surveillance and may have the power to detect chemotherapy-resistant clones before they leave the bone marrow niche, where they can arise even years after initial diagnosis. If actionable targets are identified in liquid biopsies used to monitor patients, second-line targeted treatment options could be considered much nearer to detecting therapy nonresponse.

That molecular features determine neuroblastoma aggressiveness and risk for relapse is well documented (10), adding MYCN amplifications and ALK mutations or amplifications, among others, to the clinical risk factors (11). Molecular factors have classically been assessed from the single diagnostic primary tumor biopsy. However, recent publications demonstrate clonal and subclonal heterogeneity in neuroblastomas at diagnosis (12) and branched clonal evolution with increasing molecular heterogeneity at relapse (13), supporting the necessity to monitor clonal evolution for optimal personalized care. These revelations may also explain why existing DNA- and mRNA-based molecular classifiers (14–16) do not sufficiently predict differential survival and heterogeneous outcomes in patients with high-risk disease. MYCN amplification occurs in approximately 25% of neuroblastomas and is a strong predictive biomarker for unfavorable patient survival (17). Recent data suggest MYCN amplification can exist at the (sub)clonal level, necessitating biosampling procedures and technologies capable of detecting these cell populations (18, 19). Activating mutations in the anaplastic lymphoma kinase (ALK) gene occur in 8% of neuroblastomas, the most frequent causing the F1174 L and R1275Q substitutions in the receptor tyrosine kinase domain (20–23). ALK-driven neuroblastomas often develop relapses that may have expanded from a single ALK mutant clone (24) and are frequently resistant to chemotherapy and radiotherapy (9, 20, 21). Activating ALK mutations or amplifications have become the first target in neuroblastoma that is directly druggable by small-molecule inhibitors as a personalized medicine approach (25, 26), necessitating continuous molecular monitoring in patients with neuroblastoma for potential (re)emergence of ALK mutant or amplified clones.

Several studies most recently demonstrated that pediatric patients with solid tumors including neuroblastomas have blood ctDNA levels detectable with next-generation sequencing approaches (27–30). Optimized preanalytic sample processing allowed reliable assessment of copy-number variations, segmental chromosomal changes, and single-nucleotide variants (SNV) in ctDNA from children with neuroblastoma, hepatoblastoma, and sarcoma (30, 31). Chicard and colleagues reported the feasibility of genomic copy-number profiling of blood-based cfDNA from 70 patients with neuroblastoma using a molecular inversion probe-based OncoScan array (27). In a follow-up study, they combined whole-exome sequencing with deep-coverage targeted sequencing to investigate sequential liquid biopsy samples from 19 patients with neuroblastoma and characterize patterns of clonal evolution (28). Scientific laboratories across the globe including ours established multiplexed detection of MYCN and ALK amplifications and ALK hotspot mutations by droplet digital PCR (ddPCR) using total cfDNA purified from biofluids and genomic DNA extracted from tumor tissue as input materials (27, 30, 32–34). This study investigated the potential of monitoring ctDNA-based markers for advanced molecular longitudinal disease monitoring and actionable target identification using ddPCR protocols applicable for the routine clinical setting.

Blood plasma, bone marrow plasma, and cerebrospinal fluid (CSF) samples were collected together with matched formalin-fixed paraffin-embedded or snap-frozen tumor samples (local ethics approval: EA2/055/17) from 31 patients with stage M, high-risk neuroblastoma according to the International Neuroblastoma Risk Group (Table 1; ref. 11). Peripheral blood and bone marrow were uniformely collected in EDTA tubes without the addition of stabilizers. CSF was sampled in sterile 10 mL polypropylene screw-cap tubes. Median patient age was 33.3 months (min–max: 2.1–169.0 months; Supplementary Table S1). All patients were treated at the Charité – University Medicine Berlin and registered in the German NB2004 Trial (EudraCT 20661) or the NB 2016 Registry (Supplementary Table S2). Informed written patient/parent consent was obtained during trial/registry participation. White blood cells served as a source for germline DNA. Blood plasma was collected (local ethics approval: EA2/131/11) from 25 pediatric patients with nonmalignant conditions and a median age of 72.4 months (min–max: 21.6–244.4 months) as comparative controls. Likewise, bone marrow plasma was collected (local ethics approval: EA4/132/17) from 24 healthy individuals with a median age of 23.2 years (min–max: 4.5–46.4 years). Archived surplus bone marrow plasma from 28 pediatric patients with acute lymphoblastic leukemia (ALL) collected at initial diagnosis was also investigated as a non-neuroblastoma control. All patients with ALL were treated within the AIEOP-BFM ALL 2009 (EudraCT 2007-004270-43) or AIEOP-BFM ALL 2017 trials (EudraCT 2016-001935-12) and had a median age of 5.7 years (min–max: 1.9–17.4 years). Informed written patient/parent consent was obtained during trial participation. All studies involving human subjects were conducted in accordance with the Declaration of Helsinki. Peripheral blood and CSF were centrifuged at 1,900 × g for 7 minutes to separate plasma or remove cell debris (32). Bone marrow was centrifuged at 450 × g for 7 minutes to separate plasma from cells. The average time interval from collecting the blood, bone marrow, or CSF sample to separating plasma or removing cell debris was 1 hour (interquartile range, 0.5–1.9 hours). All plasma and CSF samples were centrifuged a second time at 3,250 × g for 10 minutes to remove cell debris before storage at −80°C.

Overall response to treatment was assessed in line with revised International Neuroblastoma Response Criteria (35). In brief, overall response integrated tumor response in the primary tumor, soft tissue, bone metastases, and bone marrow. Primary and metastatic soft-tissue sites were assessed using the RECIST and ¹²³I-meta-iodobenzylguanidine (MIBG) imaging or ¹⁸F-fluorodeoxyglucose positron emission magnetic resonance imaging (¹⁸F-FDG-PET-MRI) for MIBG-nonavid tumors. Cytology and GD2 immunocytology were assessed in bone marrow cytospins. Tumor marker assessment was performed during routine clinical diagnostics and included blood levels of neuron-specific enolase, lactate dehydrogenase, and ferritin as well as urine concentrations of the catecholamine metabolites, homovanillic acid, and vanillylmandelic acid. Overall response was defined as complete, partial, or minor response or stable or progressive disease (35).

Genomic DNA was extracted from tumor tissues using the Qiagen Puregene Core kit A (Qiagen) or the QIAamp DNA Mini kit (Qiagen) according to manufacturer's instructions, and quantified on a Qubit 2.0 fluorometer (Life Technologies). Fragmentation was achieved by 5U of AluI or HaeIII restriction enzyme (New England Biolabs) added to each ddPCR reaction (32). Thawed blood or bone marrow plasma and CSF samples were centrifuged at 2,000 × g for 5 minutes to clear debris, then supernatants were centrifuged at 20,000 × g for 5 minutes. cfDNA was purified from a minimum of 120 µL stored samples using the QIAamp Circulating Nucleic Acid kit (Qiagen), then concentrated to 50 µL using the DNA Clean and Concentrator-5 kit (Zymo Research), both according to manufacturers’ directions. Total cfDNA was quantified using the cfDNA ScreenTape assay (Agilent) and Agilent 4200 TapeStation System according to manufacturer's instructions (33). DNA fragments between 100 and 300 bp were considered to be total cfDNA (36). The total cfDNA amount available for further analysis is summarized in Supplementary Table S3 for the different study populations. In total, 2.3% of all samples subjected to ddPCR contained a DNA input amount insufficient for a clear assay result.

The QX200 ddPCR System (Bio-Rad) was used to determine MYCN (2p24.3) and ALK (2p23.2-2p23.1) copy number and detect ALK p.F1174L (3522, C>A) and ALK p.R1275Q (3824, G>A) hotspot mutations with their corresponding wildtype sequences in duplex ddPCR assays as described previously (32, 33). Amplification of either the MYCN or ALK gene was defined as detecting ≥ 8.01 gene copies by ddPCR analysis, while 2.74 to 8.00 copies indicated a gene gain and 1.50 to 2.73 copies indicated the normal diploid gene contingent (32). In the background of plasma, 1 ng ctDNA is required to reliably quantify tumor-specific copy-number alterations. This limit of detection was determined by spiking 0.01 to 10 ng of sonicated genomic DNA from six neuroblastoma cell lines with varying MYCN amplification levels into plasma from pediatric patients with nonmalignant conditions (Supplementary Fig. S1). Briefly, the following T100 Thermo Cycler (Bio-Rad) programs were performed: (i) copy-number variation (CNV): denaturation at 95°C for 10 minutes, 40 cycles of 30 seconds at 94°C and 1 minute at 58°C, and final denaturation for 10 minutes 98°C and (ii) SNV: denaturation at 95°C for 10 minutes, 40 cycles of 30 seconds at 94°C and 1 minute at 62.5°C, and final denaturation for 10 minutes 98°C. Optimized primer and probe concentrations for CNVs and SNVs are summarized in Supplementary Tables S4 and S5, respectively. Target gene CNV and mutant allele frequency were analyzed using QuantaSoft Analysis software, version 1.7.4.0917 (Bio-Rad). All ddPCR assays contained appropriate non-template, positive and negative controls in each run to enable software to generate specific thresholds. The QuantaSoft Analysis software used for duplex ddPCR assays determined copy number by calculating the ratio of target molecule concentration, A (copies/µL), to the reference molecule concentration, B (copies/µL), times the number of reference species copies, N_B, in the human genome|\ ( {copy\ number\ = \ {{{\frac{A}{B}}}} \times {N_B}} )$ |⁠. False-positive rate and limit of detection for SNV analyses were calculated with Bio-Rad lookup tables in line with the model by Armbruster and Pry (37). False-positive rate was, in principle, calculated from two parameters, the number of false-positive droplets and the minimally required concentration of mutant target molecules. False-positive rate calculation was performed for each SNV protocol as described previously (33). A sample was scored as positive if both the number of droplets detecting the respective mutation and the concentration of mutant target molecules (copies/µL) were above the set thresholds (33).

The nonparametric Mann–Whitney U test evaluated the significance of differences between total cfDNA concentrations in patient cohorts. The statistical relationship between two variables was calculated using Pearson correlation coefficient. All tests were conducted using GraphPad Prism version 7.0 (GraphPad Software). P values below 0.05 were considered significant.

Targeted sequencing data have been deposited into the European Genome-phenome Archive under accession number EGAS00001006027 (https://www.ebi.ac.uk/ega/home). Original ddPCR data generated in this study are available upon request from the corresponding author.

Circulating cfDNA was previously reported to be present at higher levels in blood plasma from patients with cancer compared with healthy individuals (38). We set out to validate this in blood plasma collected at initial and/or relapse diagnosis prior to starting systemic therapy in 31 patients with high-risk neuroblastoma (Supplementary Tables S1 and S2). Control blood plasma samples were available as residues from routine endocrinological diagnostics from 25 children. Median cfDNA levels were 68-fold higher in blood plasma from patients with high-risk neuroblastoma compared with controls at initial diagnosis (Fig. 1A). Similarly, cfDNA levels in blood plasma from patients with high-risk neuroblastoma at relapse diagnosis were 5-fold higher compared with controls (Fig. 1A). The concentration of circulating cfDNA varied considerably more at initial diagnosis than at diagnosis of relapse in the study cohort, and the median total cfDNA concentration at initial diagnosis was 13-fold higher than at relapse diagnosis (Fig. 1A). Segregating the study cohort according to the age at diagnosis, the tumor MYCN, chromosome 1p36, or ALK status demonstrated no differences in total cfDNA concentration (Fig. 1B; Supplementary Fig. S2). We next compared cfDNA concentrations purified from bone marrow plasma collected from healthy individuals or pediatric patients at initial diagnosis of ALL or high-risk neuroblastoma with infiltrated bone marrow. Bone marrow–derived cfDNA concentrations in patient groups with neuroblastoma and ALL were similar, and on average up to 5-fold higher compared with controls (Fig. 1C). A comparison of blood and bone marrow–derived cfDNA levels demonstrated a strong correlation between both compartments, suggesting either a high correlation between systemic disease burden and disease activity in the bone marrow niche or a preanalytic dilution of the bone marrow through peripheral blood during the sampling process (Fig. 1D). Additional datasets including single-cell analyses from the bone marrow niche are necessary to reliably interpret ctDNA surveillance from bone marrow plasma and its clinical diagnostic potential. We performed a data meta-analysis to investigate the relationship between cfDNA amount shed into peripheral blood and established parameters for disease activity, cellular turnover, and systemic disease burden from blood (neuron-specific enolase, lactate dehydrogenase activity, and ferritin), urine (the catecholamine metabolites, homovanillic acid, and vanillylmandelic acid), and the scoring results from diagnostic MIBG imaging. These analyses demonstrated that cfDNA levels in patients with high-risk neuroblastoma closely correlated at both initial and relapse diagnosis with neuron-specific enolase levels and lactate dehydrogenase activity but not with ferritin levels or catecholamine metabolite excretion in the urine (Fig. 2A and B; Supplementary Fig. S3). MIBG-derived Curie scores (39) correlated with total cfDNA in blood plasma at initial diagnosis but not diagnosis of relapse (Fig. 2C). MIBG scores according to SIOPEN (40) correlated with blood plasma cfDNA levels at initial diagnosis in the bone compartment but not in any other constellation (Fig. 2D). Altogether, total cfDNA concentrations are higher in blood and bone marrow plasma from patients with high-risk neuroblastoma than in healthy controls, and the extent of ctDNA shed into the blood most closely correlated with markers for neuronal activity and high cellular turnover.

Multiregion tumor sequencing has shown that the multiple genomic aberrations detected vary within the tumor, demonstrating intratumor heterogeneity (41). Clonal evolution of mutations and copy numbers was also recently reported in neuroblastomas (42). To test whether ctDNA-based diagnostics can detect intratumor heterogeneity not reflected in single biopsies, we compared marker profiles in matched blood and tumor tissue samples. The first blood sample collected from each patient at initial and/or relapse diagnosis was used for comparison, demonstrating that MYCN copy number was generally strongly correlated between cfDNA and tumor tissue (Fig. 3A; Supplementary Table S6). In 2 patients, a tumor MYCN amplification or gain were not detected in cfDNA purified from the first blood sample collected (Fig. 3A; Supplementary Tables S7 and S8). In 1 patient, a tumor MYCN amplification was detected as a gain in cfDNA (Fig. 3A), most likely due to dilution of the high-level MYCN copy-number signal through background DNA signals released from healthy tissues into circulation. Strikingly, cfDNA analysis detected a MYCN amplification in 3 patients classified as tumor MYCN diploid (Fig. 3A; one of these cases exemplarily shown in Supplementary Fig. S4A; Supplementary Table S9). Similarly, a MYCN amplification was detected in cfDNA from 2 patients whose tumor analysis detected a MYCN gain. These data indicate that cfDNA analysis detected MYCN-amplified tumor clones or subclones not reflected in the tumor biopsy used to characterize molecular disease. Comparing ALK copy number in cfDNA and tumor tissue produced a similar picture. A tumor ALK gain was not detected in the first blood sample from 4 patients, but detected in subsequent samples (Fig. 3B; Supplementary Table S6). Vice versa, an ALK gain was measured in cfDNA that was not detected in the tumor samples available from 13 patients at initial and/or relapse diagnosis (Fig. 3B). Detection of ALK p.F1174L and ALK p.R1275Q hotspot mutations were strongly concordant between cfDNA and tumor tissue. Tumor mutant allele frequencies were below 1% in 3 patients, and not detectable in cfDNA purified from the first blood sample collected (Fig. 3C and D; Supplementary Table S6). In one of these cases, the ALK p.F1174L mutation became detectable in all follow-up blood samples (Supplementary Fig. S4A; Supplementary Table S9). The ALK p.R1275Q mutation detected in cfDNA from 1 patient was neither detected in the initial tumor biopsy nor resected tumor tissue from this patient (Fig. 3D). Altogether, these findings support that neuroblastoma in a patient is spatially genetically heterogeneous and that this clonal heterogeneity is reflected in circulating cfDNA. This finding has implications for ALK inhibitor therapy selection.

We next evaluated whether disease could be longitudinally monitored in blood and bone marrow plasma collected from patients with high-risk neuroblastoma during treatment course. Molecular analysis of tumor tissue from patient B50 identified a MYCN amplification and an ALK gain (Supplementary Fig. S5A; Supplementary Table S7). Both markers were employed to retrospectively analyze MYCN and ALK CNVs in blood and bone marrow plasma collected during induction chemotherapy, surgery, and high-dose chemotherapy followed by autologous stem cell rescue and immunotherapy (Fig. 4A). INRC response assessment determined a complete remission in patient B50 on day 220 after admission to the Charité (Fig. 4A). Likewise, MYCN and ALK copy numbers in blood-derived cfDNA were normal (Fig. 4A). On day 247, six copies of the MYCN amplicon were detected in blood-derived cfDNA while ALK copy numbers remained normal (Fig. 4A). Routine restaging on day 350 in the patient (clinically inapparent disease) revealed an MIBG-positive relapse localized to two independent sites in lower extremity bones (Fig. 4A). A biopsy obtained from one lesion contained MYCN-amplified and ALK-diploid neuroblastoma cells (Supplementary Fig. S5A; Supplementary Table S7). An overall increase in cfDNA levels was observed in parallel, which was in line with median blood cfDNA levels in patients at relapse diagnosis (Fig. 4A). MYCN copy numbers further increased to 15, and three copies of the ALK gene were detected (Fig. 4A). While standard bone marrow diagnostics detected no active disease, cfDNA purified from bone marrow plasma detected between six and 11 MYCN copies (Fig. 4A). Follow-up analysis on day 364 showed that MYCN copy numbers had increased to 26 in blood-derived cfDNA (Fig. 4A). Patient B50 reached a second complete remission through relapse therapy, and MYCN copy numbers normalized over time (Supplementary Table S7).

Targeted sequencing of the primary tumor from patient B22 demonstrated a partial ALK gain with approximately four copies (Supplementary Fig. S5B). This finding was utilized to retrospectively analyze blood-derived and bone marrow–derived cfDNA from this patient. The first sample was collected during routine follow-up after first-line treatment, when patient B22 had been in persistent first remission for 15.7 months, and revealed diploid ALK status and a comparatively high cfDNA level (Fig. 4B; Supplementary Table S10). The next follow-up sample, collected 10 days later, detected three ALK copies (Fig. 4B). Routine follow-up diagnostics performed on day 122 showed a relapse in the primary tumor region. Molecular analysis of biopsied tumor tissue collected on day 134 demonstrated persistence of the ALK gain (Supplementary Fig. S5B), and blood-derived cfDNA from day 132 confirmed the ALK gain (Fig. 4B). Patient B22 also reached a second complete remission, and ALK copy numbers normalized during relapse treatment (Supplementary Table S10). Radiographic evidence of relapse lagged 102 and 122 days behind ctDNA-based relapse detection in patients B50 and B22, respectively. Hence, targeted cfDNA diagnostics proved superior to all clinically established approaches for early relapse detection in 2 patients with high-risk neuroblastoma in our cohort.

We retrospectively longitudinally monitored 2 patients with high-risk neuroblastoma, who responded well to induction treatment and were in complete remission at the time of publication. Molecular analysis of tumor tissue identified a MYCN amplification and an ALK p.R1275Q mutation in patient B10 (Fig. 5A; Supplementary Table S11) and a MYCN amplification and an ALK p.F1174L mutation in patient B42 (Fig. 5B; Supplementary Table S12). Retrospective marker analysis in blood-derived and bone marrow–derived cfDNA longitudinally collected from both patients during therapy showed a molecular remission prior to day 100 of induction therapy in the liquid biopsy compartments, while standard imaging demonstrated active disease in a metastatic lesion (patient B10) and the primary tumor (patient B42), retrospectively (Fig. 5A and B). Whether rapid ctDNA tumor marker clearance correlates with favorable event-free survival will require prospective validation studies in large patient cohorts. The observation of rapid ctDNA marker clearance is contrasted by the sustained persistence of such markers in patients with refractory relapsed disease as exemplarily shown for patient B9. Molecular analysis of tumor tissue from patient B9 documented an ALK gain, which was consistently detected in blood-derived and bone marrow–derived cfDNA during second- and third-line treatment (Supplementary Fig. S4B; Supplementary Table S13). These findings demonstrate that therapy success is reflected in cfDNA-based longitudinal patient monitoring, but also that molecular markers are rapidly cleared from the blood after a patient responds well to treatment.

We analyzed molecular markers in blood plasma and CSF derived from patient B35, who experienced neuroblastoma relapse in the central nervous system (CNS) between the fourth and fifth immunotherapy cycle. Tumor tissue was MYCN amplified at initial diagnosis (Supplementary Fig. S6), when > 8 MYCN copies were detected with high total cfDNA levels in blood (Fig. 6). Standard diagnostics documented complete remission prior to high-dose chemotherapy, and MYCN cfDNA levels normalized (Fig. 6). At relapse, a MYCN amplification was detected in tumor tissue and CSF-derived cfDNA (Fig. 6; Supplementary Fig. S6). Following neurosurgical intervention and start of intrathecal and systemic relapse treatment, MYCN copy numbers in the CSF dropped to normal diploid levels (Fig. 6). Analysis of blood-derived and bone marrow–derived cfDNA detected normal MYCN copy numbers at all timepoints except day 4 after neurosurgical intervention. At this time point, four MYCN cfDNA copies were detected, which may reflect manipulation associated with the surgical intervention (Fig. 6; Supplementary Table S14). Patient B35 reached a second remission that was associated with reduced cfDNA levels in the CSF to below detection thresholds. We provide proof of concept that molecular targets can be detected in CSF from patients with neuroblastoma.

Here we provide proof of principle that liquid biopsy–based targeted approaches employing ddPCR can be used for disease monitoring in patients with high-risk neuroblastoma. Circulating cfDNA in blood plasma contains representative neuroblastoma-derived genetic material capturing common genetic alterations in the MYCN and ALK oncogenes. This ctDNA surveillance retrospectively identified high-risk neuroblastoma relapses in patients with no other evidence of disease during consolidation therapy (patient B50, marker: amplified MYCN) or after primary treatment with curative intent during follow-up (patient B22, marker: ALK gain). This suggests that identifying recurrence in patients with no clinical, radiological, or currently routine tumor marker–based evidence of disease may become a key use of liquid biopsies in patients with high-risk neuroblastoma. Recurrences were identified with a median of 3.5 months before clinical evidence of disease, in line with ctDNA surveillance in patients with diffuse large B-cell lymphoma (43) and colorectal cancer (44). This is the first study documenting the superiority of cfDNA-based diagnostics using nonpatient-specific ddPCR assays for early molecular relapse detection in patients with high-risk neuroblastoma. Whether a switch to proactive second-line “rescue” therapies based on early detection of molecular relapse or insufficient first-line response can improve overall survival for patients with high-risk neuroblastoma remains to be investigated. However, studies on minimal residual disease activity in leukemias have sustainably changed management of these diseases (45) hinting at the potential of ctDNA surveillance for high-risk neuroblastoma.

Several studies demonstrate that CSF is also suitable to analyze ctDNA in patients with primary brain tumors, including high-grade glioma and medulloblastomas (46), and tumors metastasized to the brain (47). Only low ctDNA levels circulate in blood from these individuals, possibly due to the blood–brain barrier (46). In line with a previous case report employing a quantitative real-time PCR approach (48), we demonstrate that MYCN amplification can be detected in cfDNA purified from the CSF from a patient with an isolated leptomeningeal/intracerebral neuroblastoma relapse (patient B35). This patient received repeated intrathecal topotecan administered via a CSF reservoir, allowing longitudinal CSF collection for MYCN copy-number analysis. MYCN copy number decreased over time with response to therapy, suggesting that CSF-based ctDNA can be used to monitor patients with neuroblastoma CNS relapses. This is interesting because the number of neuroblastoma CNS relapses is expected to increase in coming years, because immunotherapies such as the GD2-targeting mAb cannot cross the blood–brain barrier, and this highly vulnerable patient population remains challenging to treat. MYCN amplification was detected in parallel in blood-based ctDNA from patient B35, with this signal also decreasing over time to reach normal values when secondary complete remission was achieved. Neurosurgery may, at least in part, have enabled the blood-based signal by disturbing the blood–brain barrier or the signal may have come primarily from the leptomeningeal relapse components. Proximity of neuroblastoma brain metastases to the cortical surface, in direct contact to the CSF, may also influence ctDNA shedding into this compartment and signal strength in ctDNA surveillance. Monitoring in further patients treated for neuroblastoma CNS relapses could illuminate the physiology behind ctDNA target source during surveillance in CSF and blood plasma to improve interpretation of ctDNA surveillance of CNS-metastasized disease. Genomic methodologies such as the ddPCR assays used in this study particularly make applications in clinical settings with pediatric patients, where only very low cfDNA levels are present, possible and have been shown to outperform classical quantitative real-time PCR approaches (49).

It is widely accepted that tumor type, location, cell turnover, vascularity, and the presence and extent of circulating tumor cells and metastatic lesions influence blood ctDNA levels (46). In line with previous publications (27, 50), baseline characterization of total cfDNA in peripheral blood from patients at initial or relapse diagnosis of high-risk neuroblastoma prior to initiation of systemic therapy demonstrated significantly higher cfDNA levels than blood from healthy pediatric controls. The cfDNA concentrations detected, however, varied strongly among patients, and longitudinal total plasma cfDNA monitoring performed so far does not support a major role for total cfDNA monitoring for diagnostic neuroblastoma surveillance. Importantly, both MYCN/ALK CNVs and ALK SNVs could be identified even in samples yielding the lowest cfDNA levels.

Single-tissue biopsies do not fully mirror the spatial heterogeneity of stage M neuroblastoma as metastatic lesions are per se only very rarely biopsied. This study demonstrates that liquid biopsy–based diagnostics contributes to the identification of aggressive neuroblastoma cell clones harboring oncogenic gene amplifications or mutations that are not reflected in the tumor tissues available for molecular analysis. Specifically, the data suggest that neuroblastoma cell clones harboring alterations in the MYCN and ALK oncogenes are overrepresented in circulating tumor cells and/or metastases at distant sites. This observation supports a model in which neuroblastoma cells driven by MYCN and/or ALK are prone to migrate to and invade distant sites, thus, pointing toward intratumor heterogeneity and underlining the role of activated MYCN and ALK signaling pathway for neuroblastoma aggressiveness (20–23). In accordance with this model, this study also reports the codetection of ALK p.F1174L and ALK p.R1275Q mutations in plasma samples, but only the ALK p.F1174L mutation in the tumor tissue. The codetection of two ALK hotspot mutations in blood plasma but not the biopsy specimen from a patient with neuroblastoma has been previously reported (34), supporting clinical decision-making for targeted ALK inhibitor therapy based on ctDNA diagnostics. Altogether, this study identified genetic alterations exclusively in the ctDNA but not in tumor DNA in 10% of the patient cohort, which presents a novel finding compared with previous studies employing ddPCR approaches to study ctDNA (27, 30, 34).

The comparative analysis of ALK hotspot mutations in matched liquid biopsy–tumor tissue samples also demonstrates that mutant allele frequencies below 1% in tumor tissue are not detected in ctDNA diagnostics. This may be due to a very small amount of circulating tumor material, which is below the sensitivity threshold of the ddPCR technology applied. MYCN and ALK copy-number assessment in matched liquid biopsies and tumor tissue samples revealed loss of oncogene gain detection in the ctDNA diagnostics in few samples, most likely due to dilution effects mediated by background from DNA released from normal cells (51). Both observations highlight the importance of matched liquid biopsy–tumor tissue studies to capture the complete spectrum of genomic alterations driving malignancy through complementary analyses. The overall high concordance for MYCN and ALK copy numbers and ALK p.F1174L and ALK p.R1275Q allelic frequencies between blood-based ctDNA and tumor tissue in our patient cohort is in line with the mutational profiles detected in the KRAS, NRAS, PIK3CA, BRAF, and EGFR oncogenes in matched blood-based ctDNA-tumor tissue samples from patients with breast (46), colorectal (52), and lung cancer (53). Altogether, a plasma-positive test in the presence of tissue negativity is most likely due to the failure of the single-tissue biopsy to capture the intratumor heterogeneity of metastasized disease, and presents a strong argument for implementing dual tissue-plasma testing in patients with neuroblastoma.

The clinical potential for liquid biopsy–based molecular surveillance is just beginning to be appreciated for early diagnosis, prognostication, and identifying molecular relapses as well as assessing therapy response, secondary drug resistance, or minimal residual disease. Here we demonstrate that ddPCR-based ctDNA surveillance in liquid biopsies is applicable in the routine clinical setting and suitable for molecular profiling, early relapse detection, and actionable target identification in pediatric patients with high-risk neuroblastoma. Validation studies based on standard operating procedures for liquid biopsy sampling are warranted to test preanalytic feasibility in the multicenter setting and define the potential of ctDNA diagnostics for future interventional trials.

M. Lodrini reports grants from DKTK during the conduct of the study. J.H. Schulte reports grants from Berlin Institute of Health and German Cancer Aid during the conduct of the study. A. Eggert reports grants from Berlin Institute of Health and German Cancer Aid during the conduct of the study. H.E. Deubzer reports grants from Berlin Institute of Health (BIH), German Cancer Aid, European Union/Federal Ministry of Education and Research, German Cancer Consortium (DKTK), and Charité - University Medicine Berlin/Berlin Institute of Health, as well as non-financial support from Federal Ministry of Education and Research (BMBF) during the conduct of the study. No disclosures were reported by the other authors.

M. Lodrini: Conceptualization, data curation, formal analysis, supervision, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. J. Graef: Formal analysis, methodology. T.M. Thole-Kliesch: Formal analysis, investigation, methodology. K. Astrahantseff: Visualization, writing–original draft. A. Sprüssel: Methodology. M. Grimaldi: Data curation, investigation. C. Peitz: Formal analysis, investigation, methodology. R.B. Linke: Investigation. J.F. Hollander: Investigation. E. Lankes: Investigation. A. Künkele: Investigation. L. Oevermann: Investigation. G. Schwabe: Investigation. J. Fuchs: Investigation. A. Szymansky: Formal analysis, investigation. J.H. Schulte: Investigation. P. Hundsdörfer: Resources, investigation. C. Eckert: Resources, formal analysis, investigation. H. Amthauer: Formal analysis, supervision. A. Eggert: Conceptualization, supervision, writing–original draft. H.E. Deubzer: Conceptualization, resources, formal analysis, supervision, funding acquisition, validation, investigation, visualization, writing–original draft, project administration, writing–review and editing.

The authors thank the patients and their parents, who agreed to take part in this study, and Daniela Tiburtius, Jasmin Wünschel, Jutta Proba, Constanze Passenheim, and Nadine Sachs for excellent technical assistance.

This work was supported by the Berlin Institute of Health (BIH) through the collaborative research consortium CRG-04 TERMINATE-NB, to H.E. Deubzer, A. Künkele, J.H. Schulte, and A. Eggert; through the Translational Oncology program of the German Cancer Aid within the consortium ENABLE (70112951) to H.E. Deubzer, A. Künkele, J.H. Schulte, and A. Eggert; by the European Union and the Federal Ministry of Education and Research through the TRANSCAN-2 consortium LIQUIDHOPE (01KT1902) to H.E. Deubzer; and by the German Cancer Consortium (DKTK) partner site Berlin to M. Lodrini, A. Eggert, and H.E. Deubzer. A. Künkele and H.E. Deubzer are supported by the Advanced Clinician Scientist Program funded by the Charité – University Medicine Berlin and the BIH. A. Eggert and H.E. Deubzer are members of the MSTARS consortium (161L0220A) financed through the Federal Ministry of Education and Research (BMBF) as part of the National Research Cores for Mass Spectrometry in Systems Medicine (MSCoreSys).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.