Hypermethylated RASSF1A as Circulating Tumor DNA Marker for Disease Monitoring in Neuroblastoma.

PURPOSE: Circulating tumor DNA (ctDNA) has been used for disease monitoring in several types of cancer. The aim of our study was to investigate whether ctDNA can be used for response monitoring in neuroblastoma.
METHODS: One hundred forty-nine plasma samples from 56 patients were analyzed by quantitative polymerase chain reaction (qPCR) for total cell free DNA (cfDNA; albumin and β-actin) and ctDNA (hypermethylated RASSF1A). ctDNA results were compared with mRNA-based minimal residual disease (qPCR) in bone marrow (BM) and blood and clinical patient characteristics.
RESULTS: ctDNA was detected at diagnosis in all patients with high-risk and stage M neuroblastoma and in 3 of 7 patients with localized disease. The levels of ctDNA were highest at diagnosis, decreased during induction therapy, and not detected before or after autologous stem-cell transplantation. At relapse, the amount of ctDNA was comparable to levels at diagnosis. There was an association between ctDNA and blood or BM mRNA, with concordant results when tumor burden was high or no tumor was detected. The discrepancies indicated either low-level BM infiltration (ctDNA negative/mRNA positive) or primary tumor/soft tissue lesions with no BM involvement (ctDNA positive/mRNA negative).
CONCLUSION: ctDNA can be used for monitoring disease in patients with neuroblastoma. In high-risk patients and all patients with stage M at diagnosis, ctDNA is present. Our data indicate that at low tumor load, testing of both ctDNA and mRNA increases the sensitivity of molecular disease monitoring. It is likely that ctDNA can originate from both primary tumor and metastases and may be of special interest for disease monitoring in patients who experience relapse in other organs than BM.

INTRODUCTION

Neuroblastoma is the most common extracranial solid tumor of childhood. In approximately 50%, patients present with high-risk (HR) disease and are treated with intensive multimodality treatment protocols that encompass induction therapy, primary tumor surgery, myeloablative chemotherapy with autologous stem-cell rescue, local irradiation, and anti-GD2–based immunotherapy[1] (Appendix Fig A1). Despite this intensive therapy, in approximately one half of HR patients, the tumor will relapse and result in a fatal outcome.[2] Assessment of treatment response is based on the International Neuroblastoma Response Criteria. Meta-iodobenzylguanidine (MIBG) scintigraphy, imaging (magnetic resonance imaging/positron emission tomography scans), and bone marrow (BM) examinations by histology or (immuno)cytology are combined to assess the extent of disease.[3] Because the median age at diagnosis is 18.8 months,[4] response evaluation that is based on imaging and BM testing often must be performed under general anesthesia. Therefore, alternative methods for monitoring response would potentially result in fewer risks to these patients. Reverse transcription quantitative polymerase chain reaction (RT-qPCR) is a more sensitive technique for detection and monitoring of minimal residual disease (MRD) in neuroblastoma. Several prospective studies investigating the clinical significance of this technique for mRNA-based MRD detection in HR neuroblastoma are ongoing or have been published.[5-8]

FIG A1.

Overview of the DCOG NBL2009 treatment protocol. ASCT, autologous stem-cell transplantation and myeloablative therapy (carboplatin, etoposide, and melphalan); cis-RA, cis-retinoic acid; EBRT, external beam radiation therapy; N5, vindesine, etoposide, and cisplatin; N6, vincristine, dacarbazine, ifosfamide, and doxorubicin; S, surgery (was performed after vindesine, etoposide, and cisplatin and vincristine, dacarbazine, ifosfamide, and doxorubicin courses; optimal timing of surgery was discussed).

Key Objective

Can hypermethylated RASSF1A be used as a circulating tumor DNA (ctDNA) marker for minimal residual disease detection in neuroblastoma?

Knowledge Generated

When testing cell free DNA, we were able to detect tumor-derived hypermethylated RASSF1A in all patients with stage M disease at diagnosis. ctDNA levels decreased during treatment and were high again at relapse. Comparison between ctDNA and blood- or bone marrow (BM)–derived mRNA revealed that discrepancies were found when BM infiltration was low or when there were primary tumor lesions without BM involvement.

Relevance

ctDNA is an interesting source for monitoring disease in patients with neuroblastoma. Our data indicate that testing of both ctDNA and mRNA increases the sensitivity of molecular disease monitoring.

However, even with mRNA-based RT-qPCR, many patients with low or negative MRD results during treatment will experience recurrent disease.[5,9] As an alternative to mRNA, circulating tumor DNA (ctDNA) might be a valuable source of tumor-derived material. ctDNA comprises circulating DNA fragments (cell free DNA [cfDNA]) that carry tumor-specific alterations, which can be found in the plasma of patients with cancer.[10-12] Because of the invasiveness of tumor biopsy and the lack of repeated biopsies during follow-up, the use of liquid biopsies is being investigated. Several studies have shown the feasibility of detecting mutations or tumor-specific translocations in the ctDNA by high-depth targeted sequencing or mutation-specific PCR to monitor disease in various types of adult cancer.[13-19] However, neuroblastoma tumors, like many pediatric tumors, lack recurrent mutations and translocations. Broader analysis, such as whole-exome sequencing (WES) and shallow whole-genome sequencing (sWGS), to detect tumor-specific mutations or copy number alterations have been performed successfully using cfDNA of patients with neuroblastoma at diagnosis and relapse.[20-26] Nevertheless, these techniques are only informative when the ctDNA content is approximately ≥ 10%[27] and are, therefore, less suited for the detection of MRD, when the tumor burden is low.

In contrast to the copy number alterations and tumor-specific mutations, a methylation-specific qPCR assay could potentially be a more general and sensitive ctDNA marker. Previously, our group demonstrated that the RASSF1A gene is inactivated by hypermethylation in all stage M and MS and in 86% of localized neuroblastoma tumors. Hypermethylated RASSF1A (RASSF1Am) can be detected in BM with a similar sensitivity as mRNA and has shown added value in mRNA-negative BM.[28] In addition, RASSF1Am already has been described as a prognostic ctDNA marker at diagnosis.[29] The aim of this study was to investigate the feasibility of using ctDNA (RASSF1Am in plasma) to monitor treatment response in patients with HR/stage M neuroblastoma. We retrospectively performed qPCR for RASSF1Am on cfDNA from stored remains of previously collected plasma samples of patients with localized or metastatic neuroblastoma at diagnosis and for patients with HR/stage M neuroblastoma during treatment and at relapse. To test the additional value of ctDNA monitoring, we compared it with other techniques for disease monitoring: MIBG scans, urinary catecholamines, immunocytology, and RT-qPCR RNA-based MRD detection in BM and peripheral blood (PB).

METHODS

Between 2013 and 2016, from all consecutively diagnosed patients who were included in this study (N = 56), 149 PB samples for mRNA and cfDNA and 105 BM samples for mRNA were tested. Because stored remains were used, not all patients had samples for all time points. In this feasibility study, both HR and non-HR patients were included. Patients were treated at the Amsterdam University Medical Center (UMC), Erasmus Medical Center, or the Princess Máxima Center for Pediatric Oncology. Written informed consent from parents or guardians was obtained for all patients. The study was approved by the medical research ethics committee of Amsterdam UMC (MEC07/219#08.17.0836). Clinical data (including urinary catecholamines (homovanillic acid and vanillylmandelic acid) and imaging data (primary tumor [longest diameter], MIBG Curie score[30]) were collected from electronic patient files. Seventy-three PB samples from healthy adult male volunteers were collected as controls from Sanquin Blood Supply (Amsterdam, the Netherlands). Because RASSF1Am is also present in plasma of pregnant women, women were excluded as control donors. Pediatric PB control samples were collected from the Amsterdam UMC (Appendix Table A1).

TABLE A1.

Pediatric Control Group Data

Sample Collection, DNA Isolation, Bisulfite Conversion, and Real-Time qPCR

Methods for sample collection, DNA isolation, bisulfite conversion, and real-time qPCR for RASSF1Am[28] and mRNA markers[31] can be found in the Appendix and Appendix Table A2.

TABLE A2.

Primer and Probe Sequences

Data Analysis

Total cfDNA was quantified by qPCR for albumin (ALB) or β-actin (ACTB). A maximum of a 3.3-Ct difference between preconverted ALB and postconverted ACTB was accepted to ensure decent conversion. Samples with a ΔCT between ALB and ACTB > 3.3 are not included in the analysis for RASSF1Am. RASSF1Am was scored positive not quantifiable (PNQ) if not all wells of the triplicate were positive or one of the replicates had a Ct value > 1.5 than the mean Ct of the replicates. Quantification of RASSF1Am was performed relative to the neuroblastoma cell line IMR-32. For quantification with mRNA markers, relative values were calculated using the equation 2ΔΔCT (ΔCT sample – ΔCT IMR-32) × 100%. The median relative expression of 5 markers was used for the analysis. cfDNA and ctDNA levels were not normally distributed and are presented as median (interquartile range). Kruskal-Wallis tests were used for comparison of cfDNA or ctDNA levels. McNemar’s test was used for concordance between ctDNA and PB and BM mRNA MRD levels. All statistical analyses were performed with SPSS version 23 (IBM Corporation, Chicago, IL) or GraphPad Prism 8 (GraphPad Software, La Jolla, CA) software.

RESULTS

Patients and Samples

From 48 patients with HR and/or stage M and 8 patients with non-HR neuroblastoma, 149 samples were tested in this study (Fig 1). From the 8 patients with non-HR neuroblastoma, only diagnostic samples were tested. Patient characteristics are listed in Table 1. Six of the 149 patient samples and 12 of 73 healthy control samples were not included for RASSF1Am qPCR because too much DNA had been lost during bisulfite conversion (Fig 1). In 2 of 61 adult control samples, RASSF1Am amplification was observed (Ct value, 40.1 and 37.1), but this occurred in only 1 of 3 replicates. In the 20 pediatric control samples, no amplification of RASSF1Am was found.

FIG 1.

Flowchart of samples tested for total cell free DNA (cfDNA), number of samples excluded after too much DNA had been lost after bisulfite conversion, and number of samples tested for circulating tumor DNA (ctDNA) by hypermethylated RASSF1A (RASSF1Am). ACTB, β-actin; ALB, albumin.

TABLE 1.

Patient Characteristics

Amount of cfDNA

The amount of cfDNA per milliliter of plasma was determined by ALB or ACTB qPCR in 73 adult and 20 pediatric control samples and compared with 31 samples from patients with neuroblastoma (diagnosis or relapse). Compared with adult or pediatric control samples, samples from patients with neuroblastoma at diagnosis or relapse (all stages) had significantly more cfDNA (median, 1.5 ng/mL [interquartile range, 0.4-4.2 ng/mL], 3.1 ng/mL [interquartile range, 1.4-6.7 ng/mL], and 22.07 ng/mL [interquartile range, 5.7-98.90 ng/mL]; P < .0001 and P = .0045, respectively). Patients with stage M disease at diagnosis had the highest cfDNA levels (median, 73.1 ng/mL; interquartile range, 5.2-285.5 ng/mL; Fig 2A). There was no significant difference in total cfDNA levels during treatment and follow-up (Fig 2B). In the 28 samples where ctDNA was detected and quantified, the cfDNA levels were higher compared with the 86 patient samples where no ctDNA was detected (median, 34.2 ng/mL [range, 9.2-98.7 ng/mL] v 7.9 ng/mL [range, 3.5-25.8 ng/mL]; P = .044). Twenty-nine samples with detectable but not quantifiable ctDNA had significantly higher cfDNA levels (median, 12.9 ng/mL; range, 4.4-39.7 ng/mL) compared with adult control samples (median, 1.5 ng/mL; range, 0.4-4.2 ng/mL; P < .001). When no ctDNA was detected, cfDNA levels were still higher compared with adult control samples (median, 7.9 ng/mL; range, 3.5-25.8 ng/mL; P < .001; Fig 2C). Compared with pediatric control donors, only patients with quantifiable ctDNA levels had significantly higher levels of cfDNA (P = .0007).

FIG 2.

Amount of cell free DNA (cfDNA) and circulating tumor DNA (ctDNA). (A) Comparison of amount cfDNA (measured by albumin [ALB] or β-actin [ACTB]) between patients with neuroblastoma and healthy control donors. (B) Amount of cfDNA (measured by ALB or ACTB) at different time points during treatment. (C) Amount of cfDNA in samples with ctDNA compared with samples from healthy control donors and samples without ctDNA detected. (D) Amount of ctDNA (measured by hypermethylated RASSF1A [RASSF1Am]) at different time points during treatment. Begin induction indicates until 2 courses of induction therapy; mid-induction indicates after 3-5 courses of induction chemotherapy, unless additional courses were given after 6 courses, and samples before last course were also included at this time point; end induction indicates at the end of induction therapy; surveillance indicates during follow-up or at relapse suspicion; and event indicates relapse or progression. HR, high risk; PNQ, positive not quantifiable; PQ, positive and quantified. (*) P < .05, (**) P < .01, (***) P < .001, (****) P < .0001.

Level of ctDNA

In all diagnostic samples from patients with stage M neuroblastoma, RASSF1Am was detected. In 3 of 7 diagnostic samples from patients with localized disease and in 1 sample from a patient with stage MS disease, ctDNA was detected, although not in the quantitative range. During induction chemotherapy (patients with HR/stage M disease only), in 14 (38%) of 37 patients, ctDNA was detected (median Ct value, 30.6; min-max range, 24.7-33.8). At surveillance, 3 samples were positive, and these patients eventually experienced recurrent disease. In 8 of 9 samples from patients with relapse at the time of sampling, ctDNA was detected. Results are listed in Table 2 and in more detail in Appendix Table A3. The levels of ctDNA were highest at diagnosis, decreased during induction therapy, and undetectable at the end of induction chemotherapy. At relapse, ctDNA levels were comparable to levels at diagnosis (Fig 2D). The percentage of ctDNA of total cfDNA, calculated with the equation [RASSF1Am / (RASSF1Am + unmethylated RASSF1A) × 100], was 94% (range, 82%-98%) in the 14 diagnostic samples from patients with stage M disease. In the 28 samples where RASSF1Am could be quantified, the median percentage of ctDNA was 87% (range, 0.7%-99.9%); 29 additional samples were positive for RASSF1Am but could not be quantified.

TABLE 2.

Sample Characteristics: Detection of RASSF1Am

TABLE A3.

Clinical and Sample Data

Comparison of ctDNA and the Detection of Neuroblastoma mRNA in PB and BM

To study whether ctDNA, measured as RASSF1Am, can be used as an MRD marker in patients with HR/stage M disease, we compared it with our panel of mRNA markers.[31] In 128 matched PB samples, ctDNA could be compared with neuroblastoma mRNA, which demonstrated 79% concordant results (Fig 3A). Compared with the individual mRNA markers, RASSF1Am was more often positive, but the combined mRNA markers identified the same positive samples as RASSF1Am (Appendix Table A4).

FIG 3.

(A) Association between mRNA in peripheral blood (PB) samples and circulating tumor (ctDNA). (B) Association between mRNA in bone marrow (BM) samples and ctDNA. PNQ, positive not quantifiable; RASSF1Am, hypermethylated RASSF1A.

TABLE A4.

RASSF1Am Results Versus mRNA Panel and Individual mRNA Markers

In 93 matched BM mRNA and ctDNA (PB) samples, double-negative or double-positive results were found in 77% (Fig 3B). In contrast to PB, the BM mRNA panel identified more positive samples than ctDNA, and the individual markers PHOX2B and TH correlated best with RASSF1Am (Appendix Table A4).

Discrepant Findings Between ctDNA and PB or BM mRNA MRD

Discrepant results between ctDNA and mRNA were detected in 27 PB and 21 BM samples, respectively, and listed in Table 3 and Appendix Table A4. Total cfDNA levels in the ctDNA-positive/mRNA-negative samples were relatively high, with a median of 38.92 and 11.09 ng/mL for the BM mRNA-negative and PB mRNA-negative samples, respectively. From 3 of 5 BM mRNA-negative/ctDNA-positive samples, cryopreserved BM cells were available and tested all negative for RASSF1Am. In some patients (N850, N865, N732), the high levels of ctDNA probably correlated with the large primary or local relapse tumors, and these patients had no or very little BM infiltration.

TABLE 3.

Assessment of Discrepant Samples

In the ctDNA-negative samples, in general, the cfDNA levels were lower, with a median of 6.1 ng/mL for the BM mRNA-positive/ctDNA-negative and 1.52 ng/mL for the PB mRNA-positive/ctDNA-negative samples. In this group, the mRNA levels (in both BM and PB) were very low, mostly < 0.1%. From 15 of the 16 BM mRNA-positive/ctDNA-negative samples, cryopreserved cells were available and tested for RASSF1Am; of the 5 positive samples, 4 were not in the quantitative range, which indicated low levels of BM infiltration. In the samples from patients N777, N798, N2011, N2014, and N802, no ctDNA was detected. Apart from the very-low mRNA levels (only used for research purposes), N777 and N798 were considered to be in complete remission at that time. Subsequent samples (if available) showed negative MRD results. At the time of sampling for patients N2011 and N2014, the MIBG score was very low. Therefore, it is likely that the (biologically active) tumor load in these patients was very low. Patient N802 was treated for an isolated CNS relapse.

In the samples from patients N2012, N2013, N2016, N2024, N2029, and N2031, no ctDNA was detected, while low amounts of mRNA were detected in the BM. In the case of restricted, minimal BM disease, mRNA detection was more sensitive than ctDNA (Appendix Table A4). However, in some patients, a primary tumor was still present (median, 50 mm) while ctDNA was negative (Table 3).

DISCUSSION

ctDNA in plasma is a powerful source for the detection of tumor-derived aberrations in a minimally invasive setting. Many ctDNA studies in adults for the detection of MRD are based on detection of tumor-specific mutations by targeted sequencing or digital droplet PCR (ddPCR).[12,16,17] Because recurrent mutations are not common in neuroblastoma,[32] tumor-specific aberrations need to be characterized before they can be used as an MRD marker. However, temporospatial heterogeneity has been reported in neuroblastoma by several studies,[20,26] which raises the question of whether we should only use the small part of the tumor that is derived from the biopsy to design tumor-specific MRD markers. In the current study, we show that RASSF1Am is a universal marker for detecting ctDNA in patients with neuroblastoma. The use of RASSF1Am as an MRD marker has several potential benefits. First, it is a sensitive marker, with a sensitivity of 1 tumor cell in 105 mononuclear cells.[28] Second, RASSF1Am qPCR can be used in all patients with stage M neuroblastoma because it has been shown that RASSF1A is hypermethylated in all previously tested stage M neuroblastoma tumors.[28] Third, detection of RASSF1Am is less costly compared with WES and even sWGS (approximately 40- and 10-fold less expensive, respectively). Finally, we show in this report that RASSF1Am in plasma is tumor specific. Hypermethylation of RASSF1A has been described in several types of cancer and in physiologic circumstances in placental cells.[33] RASSF1A is not methylated in normal hematologic cells.[28,33,34] However, in 2 of 61 samples from healthy individuals, we detected very low, nonquantifiable levels of RASSF1Am. In addition, in other studies, infrequent detection of RASSF1Am has been observed in plasma samples from healthy control participants.[35,36] Therefore, when detecting very-low levels of RASSF1Am in patients with neuroblastoma (indicated as PNQ range), results should be analyzed with caution.

It has been shown that neuroblastoma tumors shed high amounts of ctDNA in the plasma.[25,26,37] In the current study, we found a median cfDNA concentration of 73.1 ng/mL at diagnosis for patients with stage M disease. This study confirms that cfDNA levels of patients with neuroblastoma are significantly higher than that of healthy donors, with patients with stage M disease having the highest levels. However, the levels we found are lower compared with previously published studies.[25,26,37] This inconsistency may be due to differences in isolation of cfDNA because we did not use a circulating nucleic acid kit or to differences in quantification methods. We used qPCR, whereas Chicard and colleagues[25,26] used the Qubit fluorometric assay (Thermo Fisher Scientific, Waltham, MA). We found the majority of the cfDNA (94% at diagnosis) to be tumor derived in patients with stage M or HR disease, which is also supported by previous research.[26,37]

We tested 143 samples from 54 patients with neuroblastoma and detected ctDNA in 57 samples. ctDNA was detected at diagnosis in all 14 patients with stage M and 4 of 8 patients with localized and stage MS neuroblastoma. Misawa et al[29] described detection of RASSF1Am at diagnosis in the serum of 17 of 68 patients (all stages) and in 11 of 18 patients with stage M disease. There are two likely causes for the increased ctDNA detection in our study. First, we used plasma, whereas Misawa et al tested serum, which is known to be more contaminated by genomic DNA originated from leukocytes during ex vivo clotting.[38] Second, Misawa et al used conventional PCR, which is less sensitive than qPCR. In the current study, ctDNA levels decreased during induction chemotherapy and were high again at relapse. This suggests that with increasing tumor burden, ctDNA levels also increase. Our group has previously described that hypermethylation of RASSF1A is variable in tumors of patients with stage MS (median, 65%) and localized (median, 30%) disease[28]; therefore, the level of ctDNA can be slightly underestimated in these patients when using RASSF1Am as marker.

We compared the performance of ctDNA with PB and BM mRNA in 128 and 93 samples, respectively. There was a strong correlation between ctDNA and BM mRNA when tumor burden was high or no tumor was detected. However, in some samples, discrepancies were observed for which additional clinical data about tumor response status were retrieved. Most patients in whom we detected relatively high levels of ctDNA compared with PB or BM mRNAs still had considerable tumor volumes or negative or low MIBG scores (data not shown); therefore, it is likely that the ctDNA in these patients originated from the primary tumor. No ctDNA was detected in 17 samples with very low PB or BM mRNA levels (< 1%). Two of these patients were in complete remission but in the other 15 patients, considerable tumor volumes were detected on imaging or urine catecholamines were still positive, which indicate the need to optimize pre-analytic sample handling and prospective study of cfDNA kinetics in well-characterized patient cohorts with available paired (nuclear) imaging and BM assessment.

While the detection of ctDNA is very promising for future MRD studies, the current study has some limitations. Stored remains were used, which resulted in missing samples and paired clinical data. Prospective collaborative studies on the use of ctDNA in the new SIOPEN HR-2 (ClinicalTrials.gov identifier: NCT04221035) patient cohort are being initiated within the SIOPEN liquid biopsy group. For detection and quantification of low levels of ctDNA in the plasma, DNA extraction methods can be optimized with an isolation method specific for cfDNA, and ddPCR[39] may be a more suited technique compared with qPCR.[40] Moreover, large amounts of cfDNA (up to 96%) could be destroyed during bisulfite conversion[41]; therefore, we are investigating alternative methylation-specific ddPCR methods. Finally, Stutterheim et al[28] showed that the percentage of RASSF1Am can be variable in neuroblastoma tumors, especially in tumors of patients with localized disease. Previous studies showed that RASSF1A was the most frequent hypermethylated tumor suppressor gene in neuroblastoma as well as identified other hypermethylated tumor suppressor genes, and inclusion of these genes as MRD markers might increase the sensitivity.[42,43]

In this study, we used RASSF1Am as a ctDNA marker. We analyzed 135 sequential samples at diagnosis, during treatment, and at follow-up for 46 patients with HR/stage M neuroblastoma. In conclusion, ctDNA can be used for monitoring disease in patients with neuroblastoma. In HR patients and all patients with stage M at diagnosis, ctDNA is present. Our data indicate that at low tumor load, the testing of both ctDNA and mRNA increases the sensitivity of molecular disease monitoring. It is likely that ctDNA can originate from both primary tumor and metastases and may be of special interest for disease monitoring in patients who experience relapse in other organs than the BM.