Serial Profiling of Circulating Tumor DNA Identifies Dynamic Evolution of Clinically Actionable Genomic Alterations in High-Risk Neuroblastoma

Neuroblastoma is an embryonal tumor that arises from the aberrant growth of neural crest progenitor cells and that disproportionately accounts for up to 10% of childhood cancer mortality (1). Overall high-risk neuroblastoma survival rates remain below 50%, and relapsed neuroblastoma is an even more significant clinical challenge (2), providing substantial motivation to better understand genomic evolution and mechanisms of therapeutic resistance to inform the development of new therapeutic strategies for this unrelenting childhood malignancy.

It has been known for decades that cells from human solid tumors such as neuroblastomas, as well as their nucleic acids, are present in the circulation of patients, but only recently have technological advances made it possible to consider the clinical application of methodologies to specifically detect this circulating tumor DNA (ctDNA) in patients (3). In turn, multiple adult cancer practitioners have begun to incorporate serial ctDNA profiling into clinical practice to aid in disease surveillance (4), characterize tumor heterogeneity (5), and define mechanisms of therapeutic resistance (6, 7). Furthermore, a recent targeted analysis of MYCN amplification and anaplastic lymphoma kinase (ALK) gene variants in neuroblastoma suggests the potential clinical value of ctDNA profiling for some of these children (8). However, how to optimally utilize comprehensive ctDNA profiling data in the pediatric oncology clinic has not been determined.

Although tumors from high-risk neuroblastoma patients at diagnosis are in part characterized by the presence of several clinically prognostic segmental chromosomal aberrations (e.g., MYCN amplification, chromosome 1p or 11q loss, or chromosome 17q gain) and several recurrent structural variants have been identified (e.g., TERT, SHANK2, and PTPRD), clonal somatic point mutations in cancer-associated genes other than ALK are rare (9–12). However, it is now clear that both primary and relapsed neuroblastomas acquire a significant number of new mutations under the selective pressure of standard neuroblastoma multimodal cytotoxic therapy (13–15), many of which occur in ALK or RAS–MAPK pathway genes (11, 13, 15). The precise timing for the acquisition of these genomic aberrations is unknown, as is how relapsed neuroblastomas continue to evolve under additional therapeutic pressures, including ALK or other targeted therapies. Furthermore, despite these emerging data, pediatric oncologists are still hesitant to recommend a tissue or bone biopsy for a child with relapsed disease given the specificity of ¹²³I-metaiodobenzylguanidine (MIBG) imaging. Direct neuroblastoma tumor biopsies also have additional challenges, including often being in technically difficult locations (e.g., paraspinal or within bone) and/or requiring procedural sedation. Although biopsies can be safely achieved via interventional radiologic procedures at some large pediatric academic centers (16), inter- and intratumoral heterogeneity remains a major issue in a cancer that is commonly widely disseminated, perhaps making it an ideal disease to utilize ctDNA profiling.

Current neuroblastoma surveillance strategies rely on cross-sectional anatomic imaging and ¹²³I-MIBG scans, along with following urinary vanillylmandelic acid (VMA) and homovanillic acid (HVA) catecholamine levels. However, these methods are clearly insufficient to detect disease progression or relapses early enough to make a difference in clinical outcomes for a significant subset of patients. Additionally, imaging evaluations not only expose young children to radiation but also usually necessitate procedural sedation adding additional clinical risk to these studies. Finally, given that neuroblastomas exist on a spectrum from undifferentiated, clinically aggressive tumors to more benign, differentiated lesions that can harbor significant amounts of ganglioneuroblastoma and/or ganglioneuroma, imaging can often be misleading in deciphering ongoing tumor activity. Thus, utilizing serial ctDNA profiling via a liquid biopsy may directly address many of these common challenges in neuroblastoma clinical care and has the potential for major diagnostic, monitoring, and therapeutic decision-making implications for children with this lethal childhood cancer.

Serial ctDNA analysis (Fig. 1A) was performed on the peripheral blood of 48 evaluable patients with the FoundationACT assay, a previous version of the FoundationOne Liquid CDx test, profiling a total of 62 genes (27 with complete exon coverage, 34 with partial exon coverage, 6 with intron coverage; Supplementary Table S1). Most of the patients profiled for ctDNA had relapsed neuroblastoma (n = 42), but children with newly diagnosed high-risk neuroblastoma (n = 4) and children who had disease progression while receiving up-front neuroblastoma therapy (n = 2) were also evaluated (Supplementary Table S2). The median age at the time of the first ctDNA sample was 89.4 months (7.5 years), and 25% of patients had tumors harboring MYCN amplification. A total of 167 ctDNA samples were sequenced (median of 3 and range, 1–10 ctDNA samples/patient) over a median time of 4.4 months (Supplementary Tables 2 and 3).

Ninety-eight percent (47/48) of evaluable patients had samples that yielded cell-free DNA (cfDNA) for sequencing (patient 40 had one ctDNA sample that failed quality control assays). ctDNA samples were sequenced to a median exon coverage depth of 9,276× (range, 3,168–18,854×). At least one known or likely pathogenic genomic alteration (mutation, amplification, or rearrangement) was found in 56% (93/167) of samples (median of 1 ctDNA variant/sample; range, 0–7 variants/sample). Moreover, 73% (35/48) of patients had at least one ctDNA sample with a pathogenic genomic alteration detected (Fig. 1B). There was a higher trend but no significant difference between the number of ctDNA variants detected in samples drawn from patients with relapsed disease or disease progression versus newly diagnosed patients with high-risk neuroblastoma (Fig. 1C). Missense pathogenic variants in ALK and TP53 were the most common alterations found in ctDNA, followed by MYCN amplification and BRAF alterations (Fig. 1D and E). The imputed mutated allele frequency (MAF) varied widely across ctDNA variants, most notably with TP53 variants having lower MAFs when compared with ALK and other genes with point mutations (Fig. 1F). More­over, 21 of the 24 TP53 variants detected occurred only in one ctDNA sample, and 14 of these 21 transient TP53 variants occurred simultaneously with each other with groups of two to four TP53 mutations in a given ctDNA sample, collection time points that were often preceded by ¹³¹I-MIBG radiotherapy (Supplementary Table S4).

High-risk neuroblastomas often present with multifocal disease, making it likely impossible to capture all potential driver mutations in a single tumor biopsy. Furthermore, biopsy at the time of neuroblastoma relapse is not always pursued, in part due to the diagnostic specificity of ¹²³I-MIBG scintigraphy, despite recent data that has shown that relapsed neuroblastomas contain significantly more somatic mutations than diagnostic tumors that often occur in potentially clinically targetable genes (11, 15). Thus, we next looked to compare ctDNA sequencing findings with those from direct tumor biopsy first in the 28 patients who had temporally paired (within 70 days) blood and tumor sequencing performed (Supplementary Table S5). When considering genes profiled in both tumor and ctDNA assays, 39% (11/28) of patients had discordant ctDNA and tumor sequencing results. Further, 59% (17/29) of detected ctDNA nucleotide variants from these 28 patients were not present in temporally matched tumor samples, including pathogenic alterations in ALK, TERT, TP53, NF1, CDKN2A, FLT3, PIK3CA, and PTPN11 (Fig. 2A and B; Supplementary Table S5). However, there was a significantly higher MAF for those variants detected in both ctDNA and tumor sequencing versus variants detected only by ctDNA profiling (median MAF 20.4 vs. 0.7%; Fig. 2C). Conversely, there was 100% concordance between the detection of MYCN (9/9) and ALK (3/3) amplification in paired ctDNA/tissue samples (Fig. 2A and B). For the 11 patients with ctDNA unique variants detected, we next directly queried the paired tumor sequencing reads and found that in only one case was there any evidence of the ctDNA-identified variant in the direct tumor sequencing reads (Supplementary Table S5).

One illustrative example of ctDNA/tumor sequencing discordance was with patient 5, a 1-year-old boy who presented to our institution with abdominal pain and proptosis. ¹²³I-MIBG scintigraphy (Fig. 2D, left) supported a diagnosis of neuroblastoma, revealing widely metastatic, MIBG-avid lesions characteristic of this disease, and abdominal MRI (Fig. 2D, top right) and a CT scan (Fig. 2D, bottom right) showed significant tumor burden in his abdomen and orbital bones, respectively. An abdominal tumor biopsy confirmed a diagnosis of neuroblastoma, and analysis with our institutional next-generation sequencing (NGS) assay revealed amplification of MYCN and other characteristic neuroblastoma segmental chromosomal aberrations but no clonal somatic point mutations. Conversely, analysis of this patient's ctDNA at the same time revealed three ALK mutations (F1245L, MAF 18%; R1275Q, MAF 1%; F1174L, MAF <1%), revealing significant tumor heterogeneity. Subsequent reanalysis of the temporally paired–direct tumor sequencing showed one read supporting the highest abundance ALK^F1245L ctDNA variant (1/2,774 reads) but not the ALK^R1275Q or ALK^F1174L variant (Supplementary Table S5). Furthermore, the serial ctDNA sequencing data from this patient correlated closely with regular surveillance ¹²³I-MIBG imaging and urine catecholamine (VMA/HVA) levels throughout his disease course (Fig. 2E). The ALK variants decreased in frequency after treatment initiation, and the ability to detect MYCN amplification and a somatic ERBB2^G1201V variant was lost from the blood only when the patient was concurrently in clinical remission (Fig. 2E, right).

We also observed ctDNA/tumor sequencing discordance in patients profiled with relapsed neuroblastoma. One example is patient 33, who presented to our institution with disease progression after initial neuroblastoma relapse over a year earlier and had a left tibial tumor biopsy performed that did not reveal any genetic variants, yet we identified a PTPN11^E76G variant in the temporally paired ctDNA analysis (Fig. 2F, middle red box). Reanalysis of the paired–direct tumor sequencing done on the same day did not uncover any evidence of the PTPN11 variant (Supplementary Table S5). Serial ctDNA data for this patient revealed that the PTPN11^E76G variant allele frequency tracked well with the patient's urinary catecholamine (VMA/HVA) values and ¹²³I-MIBG–defined disease burden, all decreasing in parallel after ¹³¹I-MIBG radiotherapy (Fig. 2F, day 1,426). Further, the exponential tumor genomic evolution revealed in this patient's third ctDNA sample done on day 1,446, including identification of a canonical ALK^R1275Q hotspot mutation, preceded the patient's overt clinical progression noted 2 months later (¹²³I-MIBG Curie score increased from 3 to 23, significant rise in urinary VMA/HVA; Fig. 2F, day 1,510) and ultimately the patient's death.

Next, we considered the comparison of ctDNA-identified genetic variants (including for patients 28, 42, and 43, ctDNA variants also identified in testing done as part of clinical care during this same period) with all direct tumor or tumor-involved bone marrow sequencing done for each patient over the course of their disease to quantify the additional genomic data that ctDNA profiling provided throughout each patient's therapy. This analysis showed that 68% (88/129) of the ctDNA variants identified in the 48 evaluable patients were unique to ctDNA and not identified in any direct sequencing performed on tumor tissue or bone marrow involved with metastatic tumor (Fig. 3A; Supplementary Table S6). Many of these variants were known driver mutations in ALK or RAS–MAPK pathway genes such as BRAF, NF1, NRAS, PTPN11, and FGFR1 (Fig. 3B–D; Supplementary Table S6), further validating the importance of these pathways in neuroblastoma tumorigenesis (11, 15). There was also a high prevalence of presumptive loss-of-function (LOF) variants in genes that modulate response to DNA damage such as BRCA1, BRCA2, and TP53—many of them also uniquely found in the ctDNA profiling (Fig. 3B, C, E, and F; Supplementary Table S6). For TP53, while two frameshift variants were identified, most variants identified were missense mutations in the DNA binding domain (DBD), including an R248Q variant that was found in two different patients. Two of the TP53 missense variants (R135F and R158C) and one BRCA2 nonsense variant (Y1655*) were found to have MAFs near 50%, suggesting that these variants may be of germline origin. These data validated what was known for one patient (patient 14, TP53^C135F) based on prior direct tumor sequencing, but for the other two patients (patient 23, TP53^R158C; patient 45, BRCA2^Y1655*) these presumptive germline mutations had not been defined prior via tumor (or germline) sequencing done as part of their routine clinical care. In turn, for patient 23, the TP53^R158C variant was also observed in a paired tumor biopsy done on the same day as the ctDNA profiling (Supplementary Table S5) as well as a germline sample tested later in their clinical course. However, no tumor or germline sequencing was performed for patient 45 in whom the BRCA2^Y1655* variant was identified.

Finally, to evaluate the potential therapeutic utility of serial ctDNA analysis, we estimated the number of these ctDNA unique variants that could potentially be clinically targeted by available drugs (Supplementary Table S6). Of the 88 unique ctDNA variants identified across the 48 evaluable patients in this study, we estimate approximately 49% (43/88) of these ctDNA unique variants are potentially clinically targetable, primarily with small molecules that inhibit ALK, CDK4/6, or RAS–MAPK pathways, all of which have shown efficacy in neuroblastoma preclinical models and are being tested clinically (Supplementary Table S6; refs. 17–21). Taken together, serial ctDNA profiling significantly enhances the detection of potentially clinically actionable driver mutations in cancer-associated genes throughout neuroblastoma therapy.

The use of ALK inhibition (ALKi) strategies for children with neuroblastoma harboring kinase activating missense mutations has become increasingly common with the growing evidence of this mutated receptor being a bona fide driver in somatically acquired tumors (9) and the major cause of familial neuroblastoma (22), as well as the increasing availability of multiple generations of inhibitors of aberrant ALK activation (18, 19, 23). Although ctDNA has been found to be a critical tool in adult cancers in noninvasively defining resistance to ALKi (6, 7), much less is known about how neuroblastomas evolve under ALKi therapeutic pressure (19, 24–26). Fifteen patients enrolled in this study received an ALKi, including both the first-generation ALKi crizotinib and also several next-generation drugs (Supplementary Table S2), 14 of which had a documented ALK mutation. Nine of these 14 patients with ALK-mutant tumors had 3+ serial ctDNA blood draws while receiving ALKi therapy. For six of these nine patients (patients 10, 28, 32, 35, 42, and 43; Fig. 4A), serial ctDNA analysis revealed clear genomic evolution of their disease with the acquisition of not only additional mutations in the ALK tyrosine kinase domain (Fig. 4B) but also variants in RAS–MAPK pathway genes predicted to activate potential ALK signaling bypass pathways (Fig. 4A). Several of these potential ALK bypass pathway mutations were found in the ctDNA from multiple patients, including BRAF^V600E (patients 32, 35, and 43), NRAS^Q61R (patients 32, 42, and 43), NRAS^Q61K (patients 32 and 43), and FGFR1^N546K (patients 32 and 43) variants. Furthermore, in these cases, the evolution of ctDNA variant complexity under ALKi therapeutic pressure either immediately preceded or was concurrent with disease progression documented by serial imaging, a rise in urine catecholamine levels, and/or a significant increase in clinical symptoms, including death (Fig. 4C and D; Supplementary Fig. S1A–S1C). Thus, serial ctDNA profiling can define clinically relevant genomic evolution in response to ALKi therapy in patients with neuroblastoma.

We next sought to characterize whether serial ctDNA sequencing data were comparable with or enhanced standard neuroblastoma clinical surveillance methods. We focused on patients who had at least three ctDNA blood draws generally timed with other standard disease evaluations with at least one recurrent ctDNA-identified variant. In a majority (15 of 16) of these patients, following serial ctDNA variant MAFs correlated well with clinical disease surveillance data (¹²³I-MIBG Curie score, PET, MRI, or CT scan, and/or urine HVA/VMA levels), some of which have been mentioned earlier due to ctDNA and tumor profiling discordance (patients 5 and 33; Fig. 2D–F) or ALKi-induced genomic evolution (patients 28, 32, 42, and 43; Fig. 4A–D; Supplementary Fig. S1A–S1C).

Additionally, for three other patients (7, 13, and 21) with prolonged treatment courses and multiple ctDNA samples that revealed genomic evolution of their disease over time, we compared their ctDNA profiling with other disease surveillance data (Fig. 5A–F). The rising MAFs of several presumptive driver mutations (NRAS^Q61K for patient 7, CDKN2A p14^{ARF Q57*} and PTPN11^G503E for patient 13, and BRCA1^C1225* for patient 21) in these patients’ ctDNA and overall increased ctDNA variant complexity were both temporally associated with times of disease progression (e.g., increasing ¹²³I-MIBG Curie scores and/or an increase in urine VMA/HVA levels; Fig. 5A–F).

For patient 25, a 7-year-old girl with a direct tumor seq­uencing identified ALK^F1245L mutation and chemotherapy-refractory disease, ctDNA profiling also provided a useful adjunct to her neuroblastoma surveillance (Fig. 6A). After a partial disease response to 4 cycles of chemoimmunotherapy (irinotecan/temozolomide/dinutuximab), she was started on ALKi. She initially had a partial response to ALKi followed by stable disease for over 16 months (a stable ¹²³I-MIBG Curie score of 7 and normal urine VMA/HVA levels; Fig. 6A), with 10 unrevealing ctDNA analyses during this almost 500-day period. However, an ALK^F1245L mutation was detected in her 11th ctDNA blood draw on day 875, which correlated temporally with a new MIBG-avid lesion being noted in her left pelvic bone on day 876. Biopsy of this pelvic lesion revealed an ALK^F1245L-mutated neuroblastoma.

Next, for patient 44, a 9-year-old boy with relapsed neuroblastoma who had a KIT^D816V variant and MYCN amplification found concurrently in his tumor and ctDNA at the time of relapse, serial ctDNA data revealed KIT^D816V MAF and MYCN amplification trends that correlated well with an initial response to chemoimmunotherapy and ultimately with disease progression after 14 cycles of this therapy (Supplementary Fig. S2A). Further, for patient 36, a 2-year-old boy with a BRAF^G469A-mutated, relapsed neuroblastoma, loss of detection of this BRAF variant in his ctDNA correlated well with a clinical disease response (Supplementary Fig. S2B). This case also highlights that for some neuroblastomas with bulky ganglioneuroblastoma/ganglioneuroma components where imaging is not always an accurate measurement of disease activity (e.g., stable MIBG Curie score), serial ctDNA monitoring may provide an important adjunct to disease surveillance. Finally, for patient 10, whose tumors harbored both an ALK amplification and an ALK^F1174L variant, serial ctDNA analyses revealed both correlation with serial surveillance PET scans and also possible spatial tumoral heterogeneity of these genetic alterations, with the ALK amplification likely arising from the neck tumor and the ALK^F1174L variant arising from a left-sided abdominal tumor (Supplementary Fig. S1C).

In at least three cases, serial ctDNA monitoring enriched disease surveillance beyond standard clinical evaluations. One of these cases (patient 33) was previously shown in Fig. 2F. Another example was with patient 35, a 2-year-old girl with MYCN-amplified, high-risk neuroblastoma (Fig. 6B). For this child, an ALK^R1275Q variant was noted on diagnostic tumor sequencing, and she was enrolled on this serial ctDNA profiling study after a second autologous stem cell transplant (ASCT), from which she suffered extensive clinical morbidity. Although the patient was noted to be in clinical remission on a comprehensive set of disease evaluations (no tumor noted on ¹²³I-MIBG or MRI imaging, normal urine HVA/VMA levels) at study enrollment, a low-level of the ALK^R1275Q variant was noted in her first ctDNA sample (MAF of 0.17%, day 357). Serial ctDNA evaluations during the following 5 months showed a rising ALK^R1275Q MAF (5 to 41%) along with MYCN amplification and other somatic variants, again concurrent with otherwise negative clinical disease evaluations (red box in Fig. 6B). Here, the initial ctDNA evidence of recurrent neuroblastoma indicated by a rising ALK^R1275Q MAF preceded findings of tumor relapse using standard methods by greater than 100 days. Finally, for patient 6, a 6-year-old boy with an ALK^R1275Q-mutated, relapsed neuroblastoma, rapidly rising ALK^R1275Q MAFs were also predictive of his overt clinical progression and ultimately his death (Fig. 6C).

In only one of these 16 cases did ctDNA and standard clinical surveillance data not correlate (Supplementary Fig. S2C). For patient 41, a 10-year-old boy with relapsed neuroblastoma and multiple TP53 variants (V157fs*13, G334V, and C277Y) identified across serial ctDNA samples, there was little correlation between the TP53 MAFs and other clinical disease evaluations (¹²³I-MIBG Curie score or urine HVA/VMA levels). Here, we hypothesize that the variants identified in cfDNA in this patient may not have been derived from tumor and may be more reflective of clonal hematopoiesis (CH; ref. 27).

Finally, in an unusual case that falls outside of the above filtering parameters (patient 39), ctDNA analysis was supportive of a neuroblastoma relapse approximately 3 months before a confirmative tumor biopsy (Fig. 6D). This 8-year-old girl was in a prolonged disease remission after initial treatment for an MIBG-avid, MYCN-amplified neuroblastoma when a new right paraspinal lesion was noted by routine PET/CT imaging. Initial ctDNA analysis revealed the presence of MYCN amplification; however, a subsequent biopsy of the paraspinal tumor showed nodular fasciitis harboring a diagnostic PKM–USP6 fusion. Two months later though, additional PET/CT imaging identified a new left-sided perirenal lesion that correlated with the continued detection of MYCN-amplified ctDNA in her peripheral blood. This new mass was biopsied and found to be a MYCN-amplified neuroblastoma, prompting the initiation of salvage therapy for relapsed disease.

Lastly, we looked to understand the functional impact of a subset of ctDNA-defined variants in neuroblastoma cells. For these studies, we focused on two ERRFI1 frameshift putative LOF variants (identified in patients 15 and 21) and two ERRFI1 missense variants of unknown significance (VUS; identified in patients 15 and 46; Fig. 7A; Supplementary Tables 3 and 6). The ERRFI1 gene is encoded on chromosome 1p, a commonly deleted chromosomal arm in neuroblastoma tumors that is associated with a poor prognosis (28). Furthermore, ERRFI1 has been well documented as a tumor suppressor in other cancers (29–31), in large part by inhibiting EGFR signaling, but little is known about its functional role in neuroblastoma (32). Notably, the two LOF ERRFI1 variants we identified in ctDNA are predicted to result in proteins lacking the EGFR/ErbB binding domain (31), and the ERRFI1^R247K variant is in a region of the ERRFI1 protein critical to 14-3-3 protein binding and regulation of EGF signaling (Fig. 7A; refs. 33, 34). Thus, we first sought to better understand ERRFI1's function in neuroblastoma cells and then how these ctDNA variants may be perturbing these functions. First, we found that higher ERRFI1 expression was associated with a better prognosis in several larger neuroblastoma tumor datasets, consistent with ERRFI1 being a putative neuroblastoma tumor suppressor gene (Fig. 7B; Supplementary Fig. S3A and S3B). Further, significantly lower ERRFI1 expression was found in neuroblastomas with more clinically aggressive features (Fig. 7C; Supplementary Fig. S3C and S3D), and neuroblastoma cell lines had very low ERRFI1 expression [n = 39; median ERRFI1 fragments per kilobase of transcript per million mapped reads (FPKM) = 1.3; range, 0.3–6.3 FPKM; ref. 35] and no measurable ERRFI1 protein (Supplementary Fig. S3E). To directly investigate how these ctDNA-identified ERRFI1 mutations may be altering ERRFI1's normal function in neuroblastoma cells, we developed isogenic ERRFI1 wild-type (WT) and ERRFI1-mutant SK-N-AS, SH-SY5Y, and NLF neuroblastoma cell lines. These ERRFI1 isogenic cells showed comparable ERRFI1 expression, and ERRFI1 protein was identified in cell lines transduced with one of the LOF variants, along with both missense variants (Fig. 7D and E). Remarkably though, exogenous expression of WT ERRFI1 was rapidly lost in isogenic cells generated from lentiviral vectors with several unique exogenous promoters, an effect that could be reversed with 5-azacytidine, suggesting hypermethylation as an additional mechanism of ERRFI1 silencing in neuroblastoma cells (Supplementary Fig. S3F–S3H). Nonetheless, in transient assays, WT ERRFI1 displayed characteristics of a tumor suppressor, inhibiting EGF-induced neuroblastoma cell growth and migration, EGFR phosphorylation, and PI3K/AKT and MAPK/ERK pathway activation in isogenic SH-SY5Y, SK-N-AS, and NLF neuroblastoma cells (Fig. 7F–H; Supplementary Fig. S3I and S3J). However, these ERRFI1 protein inhibitory functions were lost most notably with the ERRFI1 putative LOF variants (E108fs*7 and R247fs*16), and with more modest inhibition of EGF-induced effects seen with the ERRFI1^R247K variant (Fig. 7F–H; Supplementary Fig. S3I and S3J). Further, although WT ERRFI1 was able to induce EGFR cellular internalization upon EGF stimulation, the mutant ERRFI1-truncated proteins also lost this critical tumor-suppressive function (Fig. 7I; ref. 36). Although the chromosome 1p deletion status was not available for the tumors from these patients, we hypothesize these acquired LOF ERRFI1 mutations are contributing to the biallelic loss of the ERRFI1 gene. In fact, for patient 15, the R247K ERRFI1 VUS was found in all six ctDNA samples collected at a MAF of ∼50%, suggesting a germline origin, and the R247fs*16 variant was identified in the last sample collected, suggesting a mechanism for the development of biallelic ERRFI1 loss in this patient's tumor. Thus, for genes such as ERRFI1, serial ctDNA analysis can also highlight likely subclonal events in understudied genes also critically involved in neuroblastoma tumorigenesis.

The technology now exists to quantify and sequence ctDNA from the plasma of patients. However, how to successfully incorporate comprehensive ctDNA profiling into the clinical care of children with cancer has not yet been realized. Here we provide a broad assessment of the potential clinical utility of serial ctDNA sequencing in children with the commonly lethal pediatric malignancy high-risk neuroblastoma, along with validating the functional relevance of ctDNA-defined and novel ERRFI1 variants in neuroblastoma cellular models.

These data support findings from previous studies focused on interrogating ALK (8, 37, 38) or MYCN amplification or other characteristic neuroblastoma segmental chromosomal aberrations (8, 38–42), further validating the ability to commonly detect ctDNA from the blood of children with neuroblastoma. Additionally, these data build upon prior ctDNA gene panel profiling of neuroblastomas from 11 patients at the time of diagnosis (43) and a whole-exome sequencing (WES) ctDNA study that focused on validating clonal evolution with serially sequenced samples in this disease (44). Here we significantly expand upon this work and study the clinical utility of neuroblastoma-focused ctDNA sequencing by using the readily available Foundation Medicine platform to serially profile ctDNA across a panel of cancer driver genes in a diverse cohort of patients with high-risk neuroblastoma, both under the therapeutic pressure of conventional neuroblastoma chemoradioimmunotherapy and more molecularly targeted therapies at the time of relapse. Importantly, the Foundation Medicine platform utilizes NGS technology that enables sensitive detection of variants even at allele frequencies <1% of the total cfDNA (45), which is more sensitive than WES approaches previously used in this disease (44). We focus on not only defining the evolution of ctDNA-defined variants in cancer-associated genes, but also discovering how these data complement or enhance real-world clinical surveillance imaging and biochemical assays and paired tumor sequencing data. For example, this ctDNA sequencing compared with tumor profiling done throughout each patient's disease course exemplified the likely extensive intra- and intertumoral heterogeneity of neuroblastomas, as over 50% of variants identified were unique to ctDNA and not found in paired tumor profiling. Although many of the ctDNA genetic variants identified were known driver mutations in ALK or RAS–MAPK pathway genes, supporting the critical importance of these signaling cascades in neuroblastoma and validating early studies in neuroblastoma focused on ALK variant detection in temporally paired blood and tumor specimens (8, 37), some identified variants have not been as commonly associated with neuroblastoma tumor­igenesis. The high prevalence of ctDNA variants in DNA damage response genes such as BRCA1, BRCA2, and TP53 was unanticipated, as were variants in the ERRFI1 gene. In fact, ERRFI1 mutations have not been identified in several prior large neuroblastoma tumor sequencing studies (9, 46), highlighting their potential subclonality and the perhaps transient nature of these mutations specifically during disease progression. We hypothesize that these lesions may be acquired “second hits” to the ERRFI1 gene in the setting of a somatic chromosomal 1p deletion that were specifically acquired at the time of disease progression. We modeled these ERRFI1 mutations in neuroblastoma cells and indeed found that they induced profound ERRFI1 LOF, consistent with this hypothesis and directly supporting a role for ctDNA-identified variants in disease pathogenesis.

Likewise, although TP53 mutations are enriched in relapsed neuroblastomas (11), the number of transient and paired ctDNA TP53 variants with low MAFs identified here was notable. These data, taken together with the increased appreciation for the prevalence of CH (27) that often involves TP53 mutations (47) that can be enriched after radiation or chemotherapy (48), raise the question that some of the TP53 ctDNA mutations identified here may be signs of treatment-induced CH rather than being tumor-derived. A majority (14/21) of the transient TP53 mutations observed in this study were preceded by genotoxic and bone marrow–suppressive ¹³¹I-MIBG therapy, suggesting that this therapy may have either induced these TP53 mutations or promoted the transient expansion of hematopoietic cells already harboring these TP53 mutations. Most strikingly, for patient 41 with two recurrent TP53 mutations identified serially in the patient's ctDNA analyses (V157fs* and G334V), the ctDNA MAFs had no correlation with other disease surveillance data, directly supporting a potential nontumor origin to these specific TP53 variants in this patient. However, although contamination by CH variants should be contemplated in all ctDNA studies, especially when using ctDNA data to make treatment decisions, most cfDNA identified variants here (and overall in children) are likely tumor-associated. This is particularly true for the ALK or RAS–MAPK pathway variants identified and other variants with a robust connection to cancer tumorigenesis that are not commonly found in CH clones (27, 47). Furthermore, specifically considering TP53, even in older patient populations with smoking histories where CH is significantly more prevalent, only a small subset of TP53 variants identified in cfDNA have been proven to be from CH and not tumor-derived (49). Nonetheless, to directly address this issue in future pediatric ctDNA studies, paired sequencing of peripheral blood cells should be routinely implemented to define the origin of cfDNA-identified variants—again a distinction of utmost importance when contemplating utilizing ctDNA data to guide therapeutic decisions.

The putative activation of potential bypass pathways upon ALKi therapy described here, such as RAS–MAPK signaling, is consistent with prior reports in patients with neuroblastoma and lung cancer and in model cellular systems (24, 26, 50). However, this study was not powered to define precise mechanisms of resistance to specific ALK inhibitors, but rather to show the potential clinical utility of ctDNA profiling in this disease under the therapeutic pressure of both targeted and conventional neuroblastoma therapies. The experience in lung cancer has shown that mechanisms of ALKi resistance can vary across drugs (51), which may also be the case in neuroblastoma and thus requires investigation in prospective ALKi-focused trials. Importantly, the serial profiling of ctDNA here also revealed genomic evolution under the pressure of other neuroblastoma therapies, which not only correlated with disease progression but also nominated other potential therapies for these patients.

Prior studies have shown the potential utility of serial ctDNA profiling in enhancing neuroblastoma surveillance and providing clinically relevant data about ongoing tumor activity (8, 44), which is further supported by the data presented here and especially critical in a disease where tumor size does not always correlate directly with tumor activity. Here serial analysis of ctDNA enabled the detection of low levels of neuroblastoma that was below the sensitivity of current clinical surveillance methods, enabling earlier detection of disease progression. Though FoundationACT was not designed for disease surveillance, the data generated here provide insight into the future utility of liquid biopsies in disease monitoring. At the time this study was initiated, the FoundationACT assay was available for the NGS of ctDNA using a limited panel of 62 cancer-associated genes. In August of 2021, the FDA approved the FoundationOne Liquid CDx test that included a 324-gene panel that was analytically and clinically validated across multiple tumor types (52). The availability of this updated liquid biopsy test could allow for the detection of an increased number of potentially targetable gene alterations in pediatric patient samples. However, this and most other NGS panels are focused on those cancer driver genes more commonly aberrant in adult cancers rather than pediatric cancers; therefore, it will also be critical to do follow-up studies with pediatric cancer– and/or neuroblastoma-specific gene panels to fully realize the extent of enhanced genetic data acquired from serial ctDNA profiling in this disease.

Finally, within this rapidly expanding field, different approaches to profiling ctDNA have been developed, including droplet digital PCR (ddPCR; refs. 8, 37, 38), next-generation sequencing (43), WES (44), and low-passage whole-genome sequencing (WGS; refs. 39, 41, 42), each with distinct advantages and disadvantages. For example, although ddPCR provides a scalable, cost-effective, and exquisitely sensitive method to detect a single or a few predefined ctDNA variants without the need for complex bioinformatic pipelines, it does not allow for the complex tumor genotyping that can be achieved with NGS-based approaches. Conversely, although WES or NGS approaches have the advantage of being able to detect a large number of ctDNA variants in parallel providing information on tumor evolution and treatment resistance (44), they can also detect often difficult-to-interpret VUS and thus require robust bioinformatic pipelines. Finally, the estimation of ctDNA fraction in total cfDNA may differ dramatically using different sequencing platforms. For example, targeted NGS approaches such as that used here can sensitively detect variants at or below a 1% MAF in cancer-associated genes, which is likely identifying subclonal intra- or intertumoral heterogeneity. However, other sequencing methods (e.g., WES) can only identify variants with higher MAFs that are often likely somatic passenger mutations. Thus, these differences should be considered when computing total ctDNA fraction using variant MAFs, and it is likely that the field will rapidly develop more accurate algorithms for quantification of the total ctDNA fraction of cfDNA. Taken together, orthogonal ctDNA detection methods may have complementary roles in the oncology clinic, such as utilizing highly sensitive ddPCR to serially quantify a few predefined ctDNA variants in a panel of patients harboring tumors with those variants or alternatively using low-passage WGS focused on detecting tumor histotype–specific copy-number aberrations, allowing for the sensitive detection of tumor progression or minimal residual disease identification. Alternatively, a WES or an NGS gene panel approach may be utilized at larger academic centers with available analytic pipelines for a smaller number of patients receiving targeted inhibitors to identify clinically actionable variants or relevant treatment-resistant clones that could facilitate the development of novel therapeutic strategies (3, 44).

Based on the observations here, we suggest that serial ctDNA profiling should be integrated into clinical practice for children with high-risk neuroblastoma to provide real-time data on genomic evolution and the development of therapeutic resistance, to identify potentially clinically targetable mutations, and to better enable less invasive and more sensitive tumor surveillance and detection of tumor heterogeneity. In time, ctDNA profiling may allow for decreased use of traditional disease surveillance methods, many of which require procedural sedation along with significant radiation exposure for children already at high risk for secondary malignancies.

Forty-nine patients with high-risk neuroblastoma treated at CHOP were enrolled on this study after signing written informed consent between September 2016 and February 2020. The study was approved by the Children's Hospital of Philadelphia (CHOP) Institutional Review Board (#15-012553) and was conducted in accordance with recognized ethical guidelines. Patient 29 was removed from the study when the patient was ultimately determined to have intermediate-risk neuroblastoma. Clinical data were collated from CHOP clinical databases, including ctDNA data from the identical Foundation Medicine ctDNA assay. In a few cases, therapy data were generalized (e.g., using “ALKi” vs. naming specific drug) or only used in an aggregate fashion to maintain the integrity of ongoing clinical trials.

Plasma from patients with neuroblastoma was evaluated using the FoundationACT hybrid capture–based NGS assay (Foundation Medicine), for which the full validation and methodology have been previously described (45). cfDNA (≥20 ng) was extracted from 5 mL of plasma and subjected to comprehensive genomic profiling using a 62-gene panel with >30,000× raw coverage, 5,000× unique coverage, and approximately 3,000× redundant (e.g., error-corrected) coverage. A set of 12 fragment-level indexed adapters with variable 6-bp DNA barcodes were used to align read pairs and exclude any sequencing errors. The bait set for this assay targets 0.14 Mb of the human genome including all exons of 27 genes, selected exons of 33 genes, selected introns of six genes, and the TERT promoter region that is recurrently mutated in cancer (Supplementary Table S1). Samples were assessed for base substitutions, short insertions/deletions, rearrangements/fusions, and copy-number variations as previously described (45). FoundationACT has previously been validated to have >99% overall sensitivity [95% confidence interval (CI), 99.1%–99.4%] for short variants (base substitutions and short insertions/deletions) at MAF >0.5%, >95% sensitivity (95% CI, 94.2–95.7%) at MAF 0.25% to 0.5%, and 70% sensitivity (95% CI, 68.2%–71.5%) at MAF 0.125% to 0.25%.

Tumor sequencing was done as part of routine clinical care, data were collated from the patient medical record, and variants were compared with ctDNA sequencing. Paired tumor sequencing was done either via Foundation Medicine's FoundationOne CDx assay or using CHOP's Comprehensive NGS Solid Tumor Panel, both of which have been comprehensively validated and described previously (53–55), and are further described in the supplementary methods. Briefly, the CHOP Comprehensive NGS Solid Tumor Panel interrogates 238 cancer genes for sequence and copy-number alterations and 110 cancer genes for fusion genes (53). For the FoundationOne CDx assay, ≥50 ng of DNA was extracted from 40 μm of formalin-fixed, paraffin-embedded tumor sections and subjected to whole-genome shotgun library construction and hybridization-based capture. Sequencing was performed on 324 cancer-related genes, including all coding exons from 309 cancer-related genes, one promoter region, one noncoding RNA, and select intronic regions from 34 commonly rearranged genes. Base substitutions, insertions and deletions, rearrangements, and copy-number changes were assessed as described previously (https://www.fda.gov/medical-devices/recently-approved-devices/foundationone-cdx-p170019s029; refs. 54, 55). Bioinformatic pipelines used for variant calling and filtering for FoundationOne CDx and FoundationACT are similar but not identical as previously described (45, 55), a limitation of these analyses when comparing tumor and ctDNA data.

Human-derived neuroblastoma cell lines (SK-N-AS, RRID:CVCL_1700; SH-SY5Y, RRID:CVCL_0019; and NLF, RRID:CVCL_E217) were obtained from the CHOP cell line bank and cultured in RPMI containing 10% FBS, 2 mmol/L L-glutamine, and 1% streptomycin/penicillin (RPMI Complete) at 37°C under 5% CO₂. Cell lines were genotyped (GenePrint 24 System; Promega) and tested for Mycoplasma, both routinely before cells were placed into the cell bank. Cells were typically utilized for the described experiments <5 passages from thaw, genotyping, and Mycoplasma testing.

The ERRFI1 pLenti CMV puro vector (WT ERRFI1) was constructed from an ERRFI1 pDONR221 vector (DNASU Plasmid Repository, HsCD00044200) cloned into a pLenti CMV puro DEST (w118-1) plasmid, which was a gift from Eric Campeau (56) and was purchased via Addgene (RRID:Addgene_17452), with a Gateway LR clonase enzyme (Invitrogen) via the manufacturer's instructions to make an ERRFI1 pLenti CMV puro vector (WT ERRFI1). To achieve the ERRFI1 mutants, the ERRFI1 pLenti CMV puro vector (WT ERRFI1) was mutated utilizing a QuickChange II XL Site-Directed Mutagenesis Kit (Agilent) according to the manufacturer's instructions. All sequences were verified with Sanger sequencing.

Total RNA was isolated utilizing Qiashredder and RNeasy Mini Kits (Qiagen), and mRNAs were converted to cDNA using a SuperScript III First-Strand Synthesis System for RT-PCR (Thermo Fisher Scientific). TaqMan gene expression assays were used to quantitate ERRFI1 (Hs01086257_g1) and HPRT1 (Hs02800695_m1). qPCR analysis was performed on an Applied Biosystems 7900HT Sequence Detection System using standard cycling conditions, and relative expression values were calculated using the delta-delta Ct method.

ERRFI1 (WT, E108fs*7, R247fs*16, R247K, and A70T) pLenti CMV puro vectors were transfected, along with pMD2.G (RRID:Addgene_12259), pMDLg/pRRE (RRID:Addgene_12251), and pRSV-Rev (RRID:Addgene_12253) vectors, into HEK293T cells (RRID:CVCL_0063) utilizing Lipofectamine 2000 (Thermo Fisher Scientific). The virus-containing supernatant was collected after 48 and 72 hours, filtered with 0.45 μm nitrocellulose membranes, and concentrated by ultracentrifugation. The virus was added to cells in the presence of 8 μg/mL polybrene (Sigma). Media were changed the next day, and puromycin (Sigma) was added for selection 48 hours prior to further experiments.

ERRFI1 isogenic neuroblastoma cells were incubated in RPMI Complete overnight, serum-starved in 1% FBS RPMI overnight, and then incubated with EGF (50 ng/mL; STEMCELL Technologies, #78006.1) for 5 minutes at 37°C. Cells were then harvested for protein lysate preparation. For cell proliferation assays, ERRFI1 isogenic cells were plated in at least triplicate in a 96-well plate in RPMI Complete overnight, serum-starved in 1% FBS RPMI overnight, and then incubated with EGF (50 ng/mL) for 72 hours at 37°C. Luminescence was measured with a CellTiter-Glo Luminescent Cell Viability Assay (Promega) according to the manufacturer's instructions in a GloMax plate reader (Promega), and luminescence was quantified relative to vehicle-treated cells.

Transwell chambers with an 8.0-μm pore polycarbonate membrane insert (Costar) were used to evaluate EGF-induced neuroblastoma cell migration. ERRFI1 isogenic SK-N-AS cells were seeded in the upper chamber in 1% FBS RPMI Complete, and 500 μL of 1% FBS RPMI + EGF (50 ng/mL) was added to the lower chamber. After 24 hours, cells in the lower chamber were fixed with 4% paraformaldehyde, stained with crystal violet, imaged, and counted.

ERRFI1 isogenic cells were pretreated with cycloheximide (10 μg/mL) for 1 hour before being treated with EGF (50 ng/mL) for 15 to 30 minutes. Cells were then washed, fixed for 15 minutes with 4% paraformaldehyde, washed, stained with an anti-EGFR antibody (BioLegend #352904; RRID:AB_10896794) for 30 minutes at 4°C, and subjected to flow-cytometric analysis using a CytoFLEX S cytometer.

ERRFI1 isogenic cells were incubated with 5-aza-2′-deoxycytidine (5-AZA; 2.5–5 μmol/L) for 48 hours at 37°C. The 5-AZA medium was prepared and replaced daily. Cells were then harvested for whole-cell protein lysate preparation.

Whole-cell protein lysates were prepared with cell lysis buffer [Cell Signaling Technology (CST), #9803], PMSF (Sigma-Aldrich), and phosphatase inhibitor cocktail 2 (Sigma-Aldrich, P5726) and 3 (Sigma-Aldrich, P0044), briefly sonicated, rotated for 15 minutes at 4°C, and centrifuged for 10 minutes, and then the supernatant was removed and protein concentration was quantified by Bradford assays. Lysates were separated on 4% to 12% Bis-Tris gels (Life Technologies), transferred to a PVDF membrane, blocked in 5% nonfat milk in Tris-buffered saline and Tween-20 (TBS-T), and blotted using standard protocols. Membranes were typically incubated at 4°C overnight in primary antibody, washed 3× in TBS-T, incubated in 1:2,000 diluted HRP-labeled secondary antibody at room temperature for 1 hour, washed an additional 3× with TBS-T, and then developed with a chemiluminescent reagent (SuperSignal West Femto, Thermo Fisher Scientific). The following primary antibodies were used: β-Actin (1:5,000; CST #4967; RRID:AB_330288), MIG6/ERRFI1 monoclonal antibody (1:1,000; Santa Cruz Biotechnology #sc-137154; RRID:AB_2101524), phospho-p44/42 MAPK (Erk1/2; T202/Y204; D13.14.4E) XP (R) Rabbit mAb (1:1,000; CST #4370; RRID:AB_2315112), p44/42 MAPK (Erk1/2; 137F5) Rabbit mAb (1:1,000; CST #4695; RRID:AB_390779), EGF Receptor Rabbit Ab (1:1,000; CST #2232; RRID:AB_331707), Phospho-EGF Receptor (Tyr1173; 53A5) Rabbit mAb (1:1,000; CST #4407; RRID:AB_331795), phospho-AKT (ser473) Rabbit Ab (1:1,000; CST #4060; RRID:AB_2315049), and AKT Rabbit Ab (1:1,000; CST #9272; RRID:AB_329827).

The ctDNA data generated in this study are available within the article and its supplementary data files. The tumor sequencing and clinical covariate data collated in this study are not publicly available due to the risk of compromising patient privacy but are included in the article and supplementary data files in a coded and aggregate manner. Any additional data are available upon reasonable request from the corresponding author.

K.R. Bosse reports grants from the Damon Runyon Cancer Research Foundation, Alex's Lemonade Stand Foundation, the NCI, The EVAN Foundation, the Department of Defense, and Laney's Legacy of Hope during the conduct of the study; personal fees and other support from Tmunity Therapeutics outside the submitted work; and a patent for glypican 2 as a cancer marker and therapeutic target licensed and with royalties paid from Tmunity Therapeutics, a patent for chimeric antigen receptors (CAR) containing glypican 2 binding domains licensed and with royalties paid from Tmunity Therapeutics, a patent for CARs targeting glypican 2 issued, and a patent for dual targeting of pediatric malignancies through CAR T cells secreting bispecific innate immune cell engagers (BICEs) pending. G. Pascual-Pasto reports grants from Alex's Lemonade Stand Foundation during the conduct of the study, as well as a patent for dual targeting of pediatric malignancies through CAR T cells secreting BICEs pending. M.L. Margolis reports personal fees from Foundation Medicine, Inc. during the conduct of the study, as well as other support from Grail, LLC outside the submitted work. M.M. Li reports personal fees from Bayer HealthCare Pharmaceuticals outside the submitted work. M. Brisson Tierno reports personal fees from Foundation Medicine, Inc. during the conduct of the study. G.R. Oxnard reports personal fees from Foundation Medicine, Inc. and Roche during the conduct of the study. Y.P. Mossé reports other support from Foundation Medicine, Inc. during the conduct of the study, as well as other support from Pfizer outside the submitted work. No disclosures were reported by the other authors.

K.R. Bosse: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. A.M. Giudice: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, project administration, writing–review and editing. M.V. Lane: Methodology, project administration. B. McIntyre: Investigation. P.M. Schürch: Data curation, formal analysis, investigation, methodology. G. Pascual-Pasto: Data curation, formal analysis, investigation, methodology. S.N. Buongervino: Methodology, project administration. S. Suresh: Data curation. A. Fitzsimmons: Project administration. A. Hyman: Project administration. M. Gemino-Borromeo: Project administration. J. Saggio: Project administration. E.R. Berko: Data curation. A.A. Daniels: Data curation. J. Stundon: Data curation, formal analysis, writing–review and editing. M. Friedrichsen: Data curation, software, investigation, methodology, writing–review and editing. X. Liu: Data curation, software, investigation, methodology. M.L. Margolis: Data curation, software, investigation, methodology. M.M. Li: Data curation, software, formal analysis. M. Brisson Tierno: Data curation, supervision, project administration, writing–review and editing. G.R. Oxnard: Resources, supervision, funding acquisition, methodology, project administration, writing–review and editing. J.M. Maris: Conceptualization, resources, formal analysis, supervision, funding acquisition, investigation, methodology, project administration, writing–review and editing. Y.P. Mossé: Resources, formal analysis, supervision, funding acquisition, investigation, methodology, writing–review and editing.

K.R. Bosse is a Damon Runyon Physician-Scientist supported (in part) by the Damon Runyon Cancer Research Foundation (PST-07-16). This work was also supported by NCI K08 CA230223 (K.R. Bosse), the Department of Defense PRCRP TTSA CA190126P2 (K.R. Bosse), NCI R35 CA220500 (J.M. Maris), NCI P01 CA217959 (J.M. Maris and Y.P. Mossé), the Giulio D'Angio Endowed Chair (J.M. Maris), the Patricia Brophy Endowed Chair (Y.P. Mossé), Alex's Lemonade Stand Foundation (K.R. Bosse, G. Pascual-Pasto, and Y.P. Mossé), the Devaney Family Fund (J.M. Maris), the Band of Parents Foundation (Y.P. Mossé), Laney's Legacy of Hope (K.R. Bosse), and The EVAN Foundation (K.R. Bosse).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.