Circulating tumor DNA sequencing of pediatric solid and brain tumor patients: An institutional feasibility study

Abstract

The potential of circulating tumor DNA (ctDNA) analysis to serve as a real-time “liquid biopsy” for children with central nervous system (CNS) and non-CNS solid tumors remains to be fully elucidated. We conducted a study to investigate the feasibility and potential clinical utility of ctDNA sequencing in pediatric patients enrolled on an institutional clinical genomics trial. A total of 240 patients had tumor DNA profiling performed during the study period. Plasma samples were collected at study enrollment from 217 patients and then longitudinally from a subset of patients. Successful cell-free DNA extraction and quantification occurred in 216 of 217 (99.5%) of these initial samples. Twenty-four patients were identified whose tumors harbored 30 unique variants that were potentially detectable on a commercially-available ctDNA panel. Twenty of these 30 mutations (67%) were successfully detected by next-generation sequencing in the ctDNA from at least one plasma sample. The rate of ctDNA mutation detection was higher in patients with non-CNS solid tumors (7/9, 78%) compared to those with CNS tumors (9/15, 60%). A higher ctDNA mutation detection rate was also observed in patients with metastatic disease (9/10, 90%) compared to non-metastatic disease (7/14, 50%), although tumor-specific variants were detected in a few patients in the absence of radiographic evidence of disease. This study illustrates the feasibility of incorporating longitudinal ctDNA analysis into the management of relapsed or refractory patients with childhood CNS or non-CNS solid tumors.

Introduction

For most pediatric central nervous system (CNS) and non-CNS solid tumors, standard diagnostic testing involves tissue biopsy for histopathologic evaluation along with increasing utilization of comprehensive molecular classification, when feasible. Obtaining this critical tissue via biopsy or surgical resection is invasive with potential risk for surgical morbidity. The clinical utility of such biopsies is potentially limited by intratumoral heterogeneity, with next-generation sequencing studies demonstrating vast clonal and subclonal diversity within a single tumor specimen^1-3. As such, molecular profiles obtained from diagnostic tissue samples can fail to capture the full breadth of genetic changes represented throughout the bulk of the tumor. Likewise, temporal heterogeneity due to treatment-related selection pressure may drive clonal evolution contributing to subsequent tumor progression, metastasis, or recurrence^2,3. To adequately capture the comprehensive spectrum of molecular aberrations and track clonal changes over time would require serial tissue biopsies, diminishing the feasibility of this approach.

In recent years, analysis of circulating cell-free DNA (cfDNA) has garnered significant attention as a less invasive, adjunctive source of tumor-related genomic information^4,5. When cfDNA originates from apoptotic or necrotic tumor cells released into the blood stream, it is referred to as circulating tumor DNA (ctDNA). Analysis of ctDNA isolated from patient plasma carries the potential to serve as a real-time “liquid biopsy,” capturing the emergence of resistant clonal populations, predicting treatment response and the potential for relapse, and allowing for more refined prognostic stratification^5,6. In adult solid tumor patients, numerous studies have demonstrated the highly sensitive and specific nature of ctDNA as a measure of evolving tumor molecular profiles^7-12. Data on the feasibility of applying similar methodologies in the pediatric population and demonstration of its clinical validity and utility are less well established. With the recent momentum for broadening the applications of ctDNA analysis to pediatric malignancies and its expanding inclusion into pediatric clinical trials for numerous cancer types, a better understanding of ctDNA detection techniques and strategies for their use in tracking pediatric cancer is essential.

There has been rapidly growing interest in the incorporation of ctDNA analysis into the diagnosis and longitudinal tracking of pediatric malignancies^13,14. Most early studies have focused on the detection of structural variants in serial ctDNA samples as opposed to sequencing specific point mutations¹⁵. Other recent applications of ctDNA assays in the pediatric population include copy-number change and single nucleotide variant detection in plasma and urine as a biomarker for Wilms tumor^16,17, preoperative diagnostic liquid biopsy for patients with clear cell carcinoma of the kidney to avoid unnecessary tumor spread^18,19, a prognostic marker for inferior outcomes in patients with Ewing sarcoma and osteosarcoma²⁰, an assay for detection of tumor suppressor gene hypermethylation in high-risk neuroblastoma²¹, and a diagnostic and prognostic marker in intermediate-risk rhabdomyosarcoma²².

Despite these encouraging early studies, continued optimization of ctDNA assays is necessary before this can be incorporated into routine, real-time clinical practice. There are numerous obstacles to the effective utilization of liquid biopsy in pediatric solid tumor patients. For example, the comparatively smaller body size of pediatric patients limits the acceptable plasma volumes that can be safely collected. These low sample volumes exacerbate the already poor sensitivity of tumor-specific ctDNA detection that exists based on the lower somatic tumor burden of pediatric versus adult solid tumor patients, especially in the setting of localized disease or early relapse without macroscopic disease²³.

We therefore conducted a study to investigate the feasibility and clinical utility of ctDNA extraction, quantification, and next-generation sequencing in a cohort of pediatric CNS and non-CNS solid tumor patients enrolled on an institutional clinical genomics study at six Texas medical institutions. The primary study objective was to determine the feasibility of detecting tumor-specific mutations in patient plasma samples. Secondary objectives included identifying factors associated with increased likelihood for plasma ctDNA detection and investigating whether ctDNA detection mirrors patient clinical course.

Materials and Methods

Patient identification and eligibility

Patients were enrolled in the Texas KidsCanSeq study (KCS), a National Human Genome Research Institute Clinical Sequencing Evidence-Generating Research (CSER) program study^24,25 seeking to evaluate the utility and implementation of clinical genomic analysis of tumor and blood samples in pediatric patients. Children less than 18 years of age with newly diagnosed or recurrent solid tumors (CNS and non-CNS), lymphomas, and rare histiocytic disorders being treated at any of six collaborating study sites were eligible. Children diagnosed with benign, non-CNS solid tumors (e.g. neurofibromas) or other hematologic malignancies were excluded. The KCS study was approved by the Baylor College of Medicine institutional review board and corresponding review boards for each collaborating sites. Patients were enrolled after parental consent and age-appropriate patient assent. Enrolled patients were given the option of consenting to longitudinal plasma collection as part of this ctDNA feasibility study.

Tumor panel sequencing

DNA was extracted from tumor samples collected by standard tissue biopsy or therapeutic surgical resection. Clinical DNA panel sequencing was performed in the Texas Children’s Hospital Cancer Genomics Laboratory as part of the KCS study, as previously described²⁶. Results were released to the patient’s medical record and returned to the treating oncologists.

Plasma collection

Patients from all study sites had initial blood samples collected at the time of KCS study enrollment (median 4 mL, range 1 to 7 mL). Additional longitudinal blood samples were collected from a subset of Texas Children’s Hospital patients (median 8 mL, range 3 to 12 mL). These longitudinal samples were collected at approximately 3-month intervals (dictated by patient clinic follow-up schedules), at times of significant change in clinical status (e.g. relapse or progression), or change in treatment modality or approach (e.g. surgical resection, radiation therapy, bone marrow transplant). Blood was collected in Streck Tiger Top Cell-Free DNA blood collection tubes²⁷ and was maintained at room temperature until cfDNA extraction was performed on average 3 days after sample collection.

Cell-free DNA extraction and quantification

Whole blood samples underwent centrifugation at 2,800 rpm for 10 minutes at room temperature. Cell-free DNA was extracted with the QIamp MinElute ccfDNA kit (Qiagen, Germantown, MD) according to manufacturer protocol. The cfDNA concentration was measured by Qubit fluorometric assay (Invitrogen, Waltham, MA) with dsDNA HS (High Sensitivity) Assay Kit. The total cfDNA concentration per mL of plasma (ng cfDNA/mL plasma) was then calculated. For a subset of patients, the quality of cfDNA extraction was confirmed by Bioanalyzer Agilent 2100 (Agilent, Santa Clara, CA) analysis using the High Sensitivity DNA chip with cfDNA quality expressed as the 180-bp fragment fraction. Samples with fragment sizes restricted to less than 300 base pairs were deemed free of genomic DNA contamination during cfDNA purification. The final elution was re-suspended in ultrapure water and stored at −20°C in cryovials until further use.

Next-generation sequencing of ctDNA

Tumor panel profiling results from patients enrolled between August 2018 and March 2020 were compared to a list of variants detectable using the commercially available Archer LiquidPlex ctDNA 28 NGS kit (ArcherDX Inc., Boulder, CO). Patients with tumor variants identified that could in theory be detected by this ctDNA panel were selected for downstream sequencing studies. ctDNA samples from these patients underwent targeted next-generation sequencing (NGS) to attempt to detect these tumor-specific mutations. ctDNA libraries were constructed with 1-30 ng of cfDNA input based on collected sample volume using Anchored Multiplex PCR according to manufacturer’s protocol to generate target enriched, molecular-barcoded libraries. Libraries were quantified using a KAPA Illumina Library Quantification Kit (Roche Diagnostics) and underwent paired-end sequencing in 4-plex reactions on an Illumina MiSeq using V3 600 bp chemistry. The samples achieved 185,810 average unique total DNA reads per sample (range: 39,343-323,870) with an average variant coverage of 2,871 (range: 230-11,060). Resultant FASTQ files were analyzed using the Archer Analysis v5.1.1.1 (ArcherDX, Boulder, CO), utilizing digital error correction for detection of SNVs and small insertions/deletions. Subsequent variant call format (VCF) files were annotated by variant effect prediction web interface²⁸. Variants were retained based on the following variant filters: protein altering variants within NCBI RefSeq or MANE Select transcripts, splice-site variants within 2bp from coding sequence, maximum allele frequency within gnomAD²⁹, 1000 genomes³⁰, and the Exome Sequencing Project (Seattle, WA; http://evs.gs.washington.edu/EVS/) from any subpopulation <0.01, total depth at variant position >200, allele count ≥3, and variant allele fraction (VAF) ≥ 0.01. Specific alterations previously detected on patient tumor sequencing were excluded from the above variant filters and retained through subsequent VCF filtering even if associated with suboptimal coverage or variant call quality. These variants were determined to be detected solely by the presence of ≥1 mutant allele detected. Sensitivity and specificity of the pipeline was determined after variant filtration and limited to the regions included in the targeted BED file (Figure 1).

Fig 1.

1. Patients presented to medical attention with an array of clinical symptoms specific to the site of their primary and/or metastatic disease. 2. Based on these symptoms, they underwent standard diagnostic work-up typically consisting of a combination of imaging studies, blood work, and/or biopsy for histologic confirmation. 3. Depending on the diagnosis, the majority of patients then proceeded to definitive therapy of either surgical resection, chemotherapy, radiotherapy, or a combination of these modalities. 4. Patients meeting KCS study inclusion criteria were then approached for consent. 5. Upon enrolling in the study, available tumor tissue was sent for further testing and initial blood samples were collected for germline testing. 6. cfDNA was then extracted and processed on initial and longitudinal blood samples, as described. Upon completion of clinical tumor sequencing, findings were disclosed to the primary oncologist who then reviewed results with the patient and family. When somatic mutations included on the aforementioned Archer LiquidPlex ctDNA 28 kit were detected, blood samples from these patients were set aside for possible future next-generation sequencing analysis.

Results

Patient consent and enrollment

From August 2018 through March 2020, 240 patients enrolled on the KCS study across all study sites. One hundred seventy-two of 240 (72%) patients and families consented to longitudinal plasma sample collection; 13/240 (5%) patients declined and 55/240 (23%) were not approached about longitudinal collection. The most common reasons provided for declining longitudinal sample collection were patient or parental anxiety followed by concern for pain associated with additional blood draws or required volume of blood draws (particularly for young children).

cfDNA Extraction and Quantification

We completed the cfDNA extraction process from 217 of the 240 (90%) patients at the time of study enrollment (Figure 2). These patients had a wide variety of CNS and non-CNS solid tumor diagnoses with the most common CNS tumors being pilocytic astrocytoma (n=14), glioblastoma (n=9), and medulloblastoma (n=8) and the most common non-CNS tumors were neuroblastoma (n=20), rhabdomyosarcoma (n=14), and lymphoma (n=11) (Supplemental Table 1). Successful cfDNA quantification was completed for 216 of the 217 (99.5%) initial plasma samples, with one sample having insufficient extracted cfDNA (Figure 2). Eighty-three patients had a second plasma sample collected and extracted with third (n=33), fourth (n=14), and fifth (n=3) longitudinal samples processed for a subset of patients. (Supplemental Figure 1). The cfDNA samples exhibited a mean DNA fragment size of 128bp (range: 114-168bp). The mean yield for cfDNA extractions performed on all initial plasma samples was 19.6 ng cfDNA/mL plasma (median 10.5, range 1.7 to 290.8). For all longitudinal cfDNA extractions (n=133), the mean yield was 21.1 ng cfDNA/mL plasma (median 7.0, range 0.57 to 1,440; p=0.87). Non-CNS solid tumors exhibited a mean cfDNA yield of 23.0 ng cfDNA/mL plasma (n=135, median 10.8, range 1.8 to 290.8)) compared to CNS tumors with a mean yield of 13.9 ng cfDNA/mL plasma (n=81, median 10.2, range 1.7 to 82.8; p=0.039) (Supplemental Figure 2).

Fig 2.

CONSORT diagram of all study participants.

Patient selection for sequencing study

Twenty-four patients with a total of 30 tumor variants (24 unique variants) were selected for further ctDNA sequencing as their tumors were found to have DNA point mutations that were theoretically detectable by the Archer 28-gene ctDNA panel. The most common genes with relevant variants in this cohort were TP53 (n=15), PIK3CA (n=5), BRAF (n=5), and KRAS (n=2) (Supplemental Table 2). The majority of patients (15/24, 62.5%) had CNS tumor diagnoses (high grade glioma, n=11; low grade glioma, n=4) while the remainder (9/24, 37.5%) had non-CNS diagnoses (colorectal adenocarcinoma, n=3; rhabdomyosarcoma, n=3; lymphoma, n=2; mucinous cystic tumor, n=1). At the time of KCS study enrollment, seven patients (29%) had relapsed or refractory disease and 11 (46%) showed evidence of metastatic disease (Table 1). Longitudinal plasma samples were available for 11 of these 24 patients (range 2 to 4 samples).

Table 1.

Patient characteristics of the 24 patients whose tumors contained variants covered by the ctDNA panel and whose plasma samples were selected for further next-generation sequencing for tumor variant detection.

Characteristic	Number	%
Sex
Female	14	58
Male	10	42
Age, years
0-5	5	21
5-10	8	33
>10	11	46
Race
Black or African American	9	37.5
White	7	29
Declined to report	4	17
American Indian or Alaska Native	3	12.5
Asian	1	4
Ethnicity
Non-Hispanic	16	67
Hispanic	7	29
Declined to report	1	4
Tumor Type
CNS	15	62.5
Non-CNS	9	37.5
Diagnosis
Glioblastoma	7	29
Colorectal adenocarcinoma	3	13
Diffuse intrinsic pontine glioma	3	13
Rhabdomyosarcoma	3	13
Rosette forming glioneuronal tumor	1	4
Pilocytic astrocytoma	1	4
Diffuse astrocytoma	1	4
Pleomorphic xanthoastrocytoma	1	4
Mucinous cystic tumor	1	4
High grade glioma	1	4
Anaplastic large cell lymphoma	1	4
Burkitt lymphoma	1	4
Metastatic disease
No	13	54
Yes	11	46
Relapsed/refractory disease
No	17	71
Yes	7	29

ctDNA variant detection

A total of 43 plasma samples (including initial samples and longitudinal samples) from 24 patients underwent targeted sequencing for ctDNA variant detection. Twenty of the 30 tumor-specific mutations (67%) were detected in the ctDNA samples from at least one initial or longitudinal sample of 16 patients (Table 2). Successful ctDNA variant detection occurred in 7 of 9 (78%) patients with non-CNS tumors, including all 3 patients with colon cancer, both patients with lymphoma, and 2 of 3 patients with rhabdomyosarcoma. The two patients in whom ctDNA analysis could not detect variants included gross totally resected embryonal rhabdomyosarcoma and mucinous cystic tumor of the ovary. Tumor variants were detected from ctDNA in 9 of 15 (60%) primary CNS tumor patients. The 6 CNS tumor patients without variant detection all had radiographic evidence of active disease at time of plasma collection. Furthermore, tumor-specific variants were successfully detected in 9 of 10 (90%) patients with metastatic disease at time of sample collection compared to only 7 of 14 (50%) of non-metastatic patients (Table 2).

Table 2.

Longitudinally collected plasma samples from study patients were sequenced to determine the ability to detect tumor-specific variants (ctDNA) at key time points throughout the clinical course.

Patient Number	Age at Diagnosis (years)	Diagnosis	Metastatic Disease at Diagnosis? (Y/N)	Relapsed/Refractory Disease at Diagnosis? (Y/N)	Days From Initial Resection/Biopsy	Evidence of Disease on Date of Specimen Collection? (Y/N)	Tumor Variant of Interest	Variant Allele Frequency in Tumor	Variant Detected in Plasma? (Y/N)	Variant Allele Frequency in Plasma
1	15.2	Colorectal adenocarcinoma	Y	N
					3	N	TP53 p.Pro250Leu	42.9	Y	0.185
					95	N	TP53 p.Pro250Leu	42.9	N	ND
					255	N	TP53 p.Pro250Leu	42.9	Y	0.253
					350	N	TP53 p.Pro250Leu	42.9	Y	0.14
2	10	Gliomatosis cerebri	Y	N
					14	Y	TP53 p.Pro47ArgfsTer76	45.8	Y	0.171
					14	Y	TP53 p.Arg273Cys	45.1	Y	0.676
					144	Y	TP53 p.Pro47ArgfsTer76	45.8	N	ND
					144	Y	TP53 p.Arg273Cys	45.1	Y	0.398
					242	Y	TP53 p.Pro47ArgfsTer76	45.8	N	ND
					242	y	TP53 p.Arg273Cys	45.1	Y	0.545
3	14.3	Colorectal adenocarcinoma	Y	N
					52	Y	TP53 p.Arg273Cys	72.8	Y	18.571
4	6.8	Diffuse intrinsic pontine glioma	N	N
					23	Y	PIK3CA p.Glu545Gly	15	Y	0.265
					23	Y	TP53 p.Ser241Cys	93.4	N	ND
5	4	Diffuse intrinsic pontine glioma	N	N
					50	Y	TP53 p.Gln17Ter	23.6	Y	0.12
6	10.3	Pleomorphic xanthoastrocytoma	N/A	Y
					84	Y	BRAF p.Val600Glu	38.5	N	ND
7	14	Mucinous cystic tumor	N	Y
					216	N	KRAS p.Gly12Ala	9.7	N	ND
8	3.2	Glioblastoma	N	N
					59	N	BRAF p.Val600Glu	37.8	N	ND
					128	N	BRAF p.Val600Glu	37.8	Y	0.1
					226	Y	BRAF p.Val600Glu	37.8	Y	0.06
9	12.3	Epithelioid glioblastoma	Y	Y
					824	Y	BRAF p.Val600Glu	47.1	N	ND
					1107	Y	BRAF p.Val600Glu	47.1	N	ND
					1188	Y	BRAF p.Val600Glu	47.1	N	ND
10	8.8	Diffuse astrocytoma	Y	Y
					335	Y	BRAF p.Val600Glu	39.1	Y	0.17
					487	Y	BRAF p.Val600Glu	39.1	Y	0.28
11	3.5	Embryonal rhabdomyosarcoma	N	N				83.7
					91	N	NRAS p.Gln61Lys		N	ND
12	6.9	Diffuse intrinsic pontine glioma	N	N
					108	Y	TP53 p.Arg175His	84.5	Y	0.435
					200	Y	TP53 p.Arg175His	84.5	N	ND
13	16.1	Colorectal adenocarcinoma	Y	Y				5.2
					247	Y	TP53 p.Arg324Ter		Y	5.259
14	7.4	Pilocytic astrocytoma	N	N
					131	Y	KRAS p.Ser65_Ala66delinsLeuAspGlnTyr	21.4	N	ND
					313	Y	KRAS p.Ser65_Ala66delinsLeuAspGlnTyr	21.4	N	ND
15	14.9	Burkitt lymphoma	N	Y
					−75*	N	TP53 p.Arg273His	65.4	Y	0.599
					36	N	TP53 p.Arg273His	65.4	Y	0.102
					135	N	TP53 p.Arg273His	65.4	Y	0.228
					221	N	TP53 p.Arg273His	65.4	Y	0.16
16	1.7	Embryonal rhabdomyosarcoma	Y	Y
					−45*	Y	PIK3CA p.Gly1049Arg	31.9	Y	0.575
					−45*	Y	CTNNB1 p.Gly48Val	26	Y	0.16
					36	N	PIK3CA p.Gly1049Arg	31.9	N	ND
					36	N	CTNNB1 p.Gly48Val	26	N	ND
					172	Y	PIK3Ca p.Gly1049Arg	31.9	Y	0.27
					172	Y	CTNNB1 p.Gly48Val	26	Y	0.21
17	5.7	Alveolar rhabdomyosarcoma	Y	N
					120	Y	PIK3CA p.Glu545Ala	30.3	Y	0.144
18	16.9	Rosette forming glioneuronal tumor	N	N
					77	Y	PIK3CA p.His1047Arg	18.1	N	ND
					163	Y	PIK3CA p.His1047Arg	18.1	N	ND
19	6.2	High grade glioma	N	N
					1	Y	TP53 p.Phe109LeufsTer39	82.2	Y	31.33
20	14.3	Glioblastoma	N	N
					181	Y	TP53 p.Arg306Ter	88.2	Y	0.03
					286	Y	TP53 p.Arg306Ter	88.2	Y	0.37
21	7.9	Glioblastoma	Y	N
					264	Y	BRAF p.Val600Glu	45.6	Y	0.15
					264	Y	TP53 p.Val216Met	74.8	Y	0.16
22	17.1	Glioblastoma	N	N
					247	Y	EGFR p.Asp770_Pro772dup	16.4	N	ND
					247	Y	PIK3CA p.Glu545Lys	51.2	N	ND
					247	Y	TP53 p.Arg196Pro	8.5	N	ND
23	14	Anaplastic large cell lymphoma	Y	N
					106	N	TP53 p.Arg248Gln	33.2	Y	0.06
24	3.3	Glioblastoma	N	N
					27	Y	TP53 p.Arg248Gln	87.3	N	ND

NOTE. Abbreviation: ND, not detected

^*Negative values signify the number of days initial blood sample collection occurred prior to tumor resection or biopsy

The median VAF of all detected variants in the plasma was 0.21% (mean 2.01, min 0, max 31.33). There were 3 ctDNA specimens with alterations detected at a VAF of greater than 5%. Two of these originated from patients with metastatic colorectal adenocarcinoma – one with diffuse peritoneal carcinomatosis (VAF 18.6%) and the other with numerous pulmonary nodules and rising carcinoembryonic antigen levels (VAF 5.3%). The third sample was from a patient with a thalamic high-grade glioma and diffusely infiltrative disease throughout the frontal and temporal lobes (VAF 31.3%). Excluding these three “outliers”, the mean VAF of the remaining samples was 0.25%.

Some tumors contained multiple somatic mutations of interest and multiple longitudinal samples were available for analysis in a subset of patients. Of the 53 total opportunities for ctDNA variant detection, 38 occurred in patients with radiographic evidence of active disease at time of sample collection with 22 (58%) successful variant detections. However, active disease was not a prerequisite for successful ctDNA detection. Nine ctDNA variants were successfully detected out of 15 (60%) opportunities for ctDNA detection (from 4 unique patients) in the absence of radiographic evidence of disease. This included patients with anaplastic large cell lymphoma, Burkitt lymphoma, glioblastoma, and colorectal adenocarcinoma (Table 2).

Case studies: ctDNA analysis of longitudinal plasma samples

Case 1

A 20-month-old female with neurofibromatosis type 1 presented with fever, emesis, and abdominal distention. Imaging revealed a large, heterogeneous 8 x 5 x 9 cm mass localized to the right lower abdomen and pelvis. Surgical resection was complicated by tumor spillage and histology was consistent with fusion negative embryonal rhabdomyosarcoma. Following upfront chemotherapy with vincristine, dactinomycin, and cyclophosphamide, the patient presented to her 3-month off-therapy visit with abdominal distention and was found to have a recurrent 16 x 14 x 9 cm mass at the site of her original tumor (Relapse #1, Figure 3A).

Fig 3.

A) Axial (i) Maximum intensity projection (MIP)(ii) images from a PET/CT at time of first recurrence three months off-therapy demonstrates extensive hypermetabolic soft tissue throughout the abdomen and pelvis. A few separate hypermetabolic implants are noted adjacent to the liver and spleen.

B) Axial (i) and MIP (ii) images from a PET/CT six months later following salvage chemotherapy and whole abdomen radiation demonstrates resolution of the abnormal hypermetoblic activity consistent with a complete metabolic response.

C) Axial (i) and MIP (ii) images from a PET/CT two months later after completion of salvage therapy demonstrates recurrence of extensive metabolically active disease throughout the right hemiabdomen and pelvis consistent with recurrent disease with metastatic implants noted adjacent to the liver.

D) Variant allele frequencies of the patient-specific somatic mutations of interest detected at key clinical time points pertaining to the radiologic images above each data point.

The patient was enrolled on the KCS study and tumor tissue from the relapsed specimen was sent for molecular testing, revealing NF1 p.Arg304Ter (VAF 45%, germline), PIK3CA p.Gly1049Arg (VAF 32%) and CTNNB1 p. Gly48Val (31%) missense variants. Analysis of a plasma sample collected at this time for ctDNA successfully detected the PIK3CA and CTNNB1 somatic variants (at VAF of 0.57% and 0.16%, respectively). She underwent re-resection of her tumor and salvage therapy with ifosfamide and doxorubicin followed by whole abdomen radiation (36 Gy). No measurable disease was detected by imaging at the end of salvage therapy; ctDNA analysis of a plasma sample collected at this time (81 days after initial sample collection) did not detect the PIK3CA or CTNNB1 variants (Figure 3B).

Two months after completion of salvage therapy the patient again presented with abdominal distention and radiographic evidence of a second relapse with multiple abdominal masses involving the liver, omentum, abdominal wall, adrenal glands, and lungs (Relapse #2, Figure 3C). Both known tumor variants again became detectable by ctDNA analysis (PIK3CA VAF 0.27%, CTNNB1 VAF 0.21%) (Figure 3D). The patient passed away approximately one month following her second relapse.

Case 2

A 9-year-old female presented with new onset focal seizures of the bilateral upper extremities. Initial imaging demonstrated a large, T1 hypointense and T2 hyperintense left temporoparietal, infiltrative mass with poorly defined borders and associated midline shift (Figure 4A). She underwent near total tumor resection (Figure 4B) and histology was consistent with a high-grade glioma, with immunohistochemistry for p53 consistent with presence of a TP53 mutation (Supplemental Figure 3). She enrolled on the KCS study at this time and tumor DNA panel sequencing revealed multiple variants, including two covered on the Archer ctDNA sequencing panel: TP53 p.Pro47ArgfsTer76 (VAF 45.8%), and TP53 p.Arg273Cys (VAF 45.1%).

Fig 4.

Ai) Initial presentation axial T1-weighted postcontrast image demonstrates subtle areas of solid enhancement in the left inferior parietal lobe and a confluent adjacent area of low T1 signal intensity.

Aii) Initial presentation axial T2 weighted image demonstrates a confluent region of T2 hyperintense signal in the inferior left parietal lobe subcortical white matter, left superior temporal lobe subcortical white matter, and extension into the posterior periventricular white matter.

Bi) First post-surgical axial T1-weighted postcontrast images following surgical resection demonstrate T1 hyperintense signal within the resection cavity related to blood products and subtle focus of enhancement along the medial resection cavity margin.

Bii) First post-surgical axial T2 weighted image demonstrates a left parietal lobe resection cavity with moderate surrounding T2 hyperintense signal in the adjacent subcortical white matter extending into the posterior aspects of left subinsular white matter, left posterior limb of the internal capsule, and posterior left periventricular white matter.

Ci) Axial T1-weighted postcontrast images 5 months after surgical resection demonstrate a heterogeneously enhancing mass at the medial margin of the parietal resection cavity.

Cii) Axial T2 weighted image 5 months after surgical resection demonstrates a rounded mass-like area of T2 hyperintense signal along the medial resection cavity margin and surrounding confluent region of mild T2 hyperintensity in the adjacent subcortical white matter, and corona radiata.

Di) Axial T1-weighted post contrast images 1 year after surgical resection demonstrate peripheral enhancement as in the left parietal lobe which has increased in size compared to the prior MRI, and new areas of homogeneous contrast enhancement within the right thalamus and within the left parietal lobe anterior to the resection cavity.

Dii) Axial T2 FLAIR image 1 year after surgical resection demonstrates a large confluent region of T2 FLAIR hyperintensity involving the left frontal, temporal, and parietal lobe white matter with extension into the corona radiata, centrum semiovale and periventricular white matter.

E) Variant allele frequencies of the patient-specific somatic mutations of interest detected at key clinical time points pertaining to the radiologic images above each data point.

The patient received six weeks of focal cranial radiotherapy followed by four months of everolimus and ribociclib on a clinical trial. The first plasma sample was collected prior to the start of radiotherapy, at which time both the TP53 p.Pro47ArgfsTer76 and TP53 p.Arg273Cys mutations were detected at VAF of 0.17% and 0.68%, respectively; after two months of everolimus and ribociclib only the p.Arg273Cys variant was detectable (0.4% VAF). Approximately one month after completion of treatment, she developed status epilepticus with a repeat MRI showing interval increase in size of the temporoparietal mass and worsening infiltration into the central gray matter and brainstem (Figure 4C). Pembrolizumab therapy was initiated, but the patient continued to have evidence of disease progression (Figure 4D). Analysis of a repeat plasma sample obtained following the first dose of pembrolizumab again revealed only the TP53 p.Arg273Cys variant (VAF 0.55%) (Figure 4E). The patient eventually developed status epilepticus and near complete right-sided hemiparesis. She passed away with hospice care following 3 doses of pembrolizumab, approximately 10 months from her initial diagnosis.

Discussion

Compared to the robust ctDNA literature in adult cancers such as breast cancer^31-33, non-small cell lung cancer^34-36, and colon cancer^37,38, there is a relative paucity of studies demonstrating the feasibility of detecting and quantifying ctDNA in pediatric solid tumors. Early studies exploring the incorporation of ctDNA analysis into pediatric malignancies have been encouraging^15,18-20,39. This feasibility study demonstrates the successful integration of longitudinal plasma collection and ctDNA analysis into the care of patients with pediatric solid and CNS tumors and provides a model for future utilization of ctDNA assessment in pediatric clinical genomic trials. The vast majority of patients and families approached about longitudinal sample collection as part of the study were amenable, with only 5% electing to opt out.

Cell-free DNA was successfully extracted from initial patient plasma samples in nearly all patients enrolled on the KCS study despite the young age of many enrolled patients and small volume of available plasma samples. A statistically significant higher cfDNA yield was extracted from initial samples in patients with non-CNS primary tumors compared to those with CNS tumors, implicating the presence of the blood-brain barrier as a hurdle complicating successful CNS tumor-derived cfDNA isolation. Similar findings have been demonstrated in other recent publications attempting to optimize the diagnostic approach and real-time disease-tracking capabilities for patients with CNS tumors^40-43. This highlights the need to identify alternative non-plasma biospecimens such as the cerebrospinal fluid to avoid the confounding impact of the blood-brain barrier. Preliminary studies in adult patients with CNS tumors have confirmed the superiority of CSF as a source for ctDNA⁴⁴. Extending these efforts into the pediatric population is critical. Inclusion of low-grade CNS tumors (e.g pilocytic astrocytoma, rosette forming glioneuronal tumor) in the study cohort with inherently less aggressive phenotypes and presumably lower propensity for blood-brain barrier invasion and cfDNA dissemination compared to their high-grade counterparts may also have contributed to the lower cfDNA yield from CNS tumors.

Through comprehensive tumor molecular profiling as part of an institutional clinical genomics trial, we were able to identify patients from our cohort with specific tumor variants and then attempt to detect these same variants within the plasma ctDNA using standard testing platforms. We demonstrated the feasibility of detecting tumor-specific mutations in the plasma ctDNA of patients with non-CNS solid tumors and lymphoma (7/9, 78%) as well as primary CNS tumors (9/15, 60%) down to extremely low VAFs. In total, 20 of 30 (67%) tumor mutations represented in our cohort were detected in at least one longitudinally-collected plasma sample. We provide early evidence that metastatic disease correlates with superior ctDNA mutation detection.

The impact of active disease on successful plasma ctDNA variant detection has been well-documented^45-47. It is important to note, we had successful ctDNA detection in 4 patients in whom the preceding imaging studies failed to demonstrate any evidence of active disease. This illustrates one of the principal applications for which ctDNA analysis has garnered such excitement – the ability to detect evidence of residual or relapsed disease prior to sufficient growth necessary to render a tumor visible on surveillance imaging. Two of these 4 patients carries primary diagnoses of lymphoma and thus their hematogenous spread lends itself to ctDNA detection despite lack of radiographic evidence of disease. The remaining 2 patients with detectable ctDNA and no active disease were a patient with glioblastoma and another with metastatic colorectal adenocarcinoma. The highly aggressive nature and associated increased risk for relapse make these diagnoses ideal candidates for a longitudinal, prospective ctDNA assay attempting to detect early signs of relapse. As illustrated by the two clinical cases presented herein, ctDNA analysis provides a real-time snapshot of tumor activity. With continued refinement of ctDNA extraction methodology and development of more precise detection capabilities, the potential implementation of ctDNA analysis into routine pediatric cancer follow-up offers great promise.

Precise isolation of tumor-specific ctDNA without prior knowledge of specific tumor mutations can be extremely difficult. Thus, this study provides a blueprint for the incorporation of ctDNA analysis into the routine clinical care of pediatric oncology patients by utilizing a priori knowledge of tumor molecular findings obtained from clinical tumor testing to attempt to identify and isolate clinically meaningful ctDNA from the peripheral blood at clinically relevant time periods. With the prior knowledge of tumor mutations afforded by the comprehensive genomic analysis performed as part of the KCS study, the ctDNA analytic pipeline could be enforced to retain any hotspot variant that had previously been seen on tumor testing. This was critical to successful ctDNA detection due to the extremely low VAF at which most of the ctDNA variants were detected. Without this information, many of these variants would have been eliminated through standard variant filters. For example, on average 4,271 (range: 19 – 9,116) variants were called per sample before applying the variant filters and only 13 (range: 0 – 46) variants retained per sample after applying the filter. Ten of 20 (50%) mutations were salvaged by the force-call mechanism based on prior knowledge of tumor mutations. Making the variant caller filters any more permissive would introduce too many false positive calls, further illustrating the importance of prior knowledge of tumor-specific mutations in the context of working with ctDNA.

There are some inherent limitations to this feasibility study in addition to the small sample size. Plasma samples were not collected from patients until they had been officially enrolled on the KCS study and consented to additional blood sample collection. In several instances, the initial plasma collection did not occur until after surgical resection, initiation of systemic chemotherapy, and/or local control with radiation therapy, potentially confounding the ability to detect tumor variants. Another limitation is the relatively small ctDNA gene mutation panel used in the study with only 28 genes commonly involved in pediatric solid tumors (CNS and non-CNS). Although suitable for a feasibility study, a more comprehensive panel containing additional genes, as well as the ability to analyze other alterations such as RNA fusions and copy number changes, will be essential to expand the clinical applications of this testing. Greater ctDNA input combined with ultra-deep sequencing at a higher average depth, together with further improvements in ctDNA workflows and analytic pipelines might also be necessary to improve detection of low-level variants.

Additional studies are required to further delineate the role of liquid biopsy in the diagnosis, prognostic stratification, clinical management, and routine surveillance of pediatric oncology patients. Including additional longitudinal samples from myriad histologies, tumor locations, and clinical presentations will be essential to better understand factors that contribute to successful ctDNA detection. Continued refinement of molecular strategies and technologies are also necessary to improve the sensitivity and specificity of such testing. Furthermore, expanding the application of ctDNA assays to include detection of gene fusions and copy number alterations in addition to point mutations broadens the impact this technology can have on the field of precision oncology. The ability to non-invasively track molecular changes in response to treatment remains an appealing concept with exciting future implications, including investigation of CSF as a potential superior source of ctDNA detection in patients with primary CNS tumors^42,48-54. Efforts are ongoing to better understand ctDNA biology and its application to pediatric oncology patients. Our study also provides an example of an effective workflow from patient enrollment, sample collection and processing, cfDNA extraction and quantification, and ultimately targeted NGS for ctDNA detection.