Molecular profiling of 888 pediatric tumors informs future precision trials and data-sharing initiatives in pediatric cancer

Childhood cancer mortality has declined over recent decades through intensified multimodal therapy and risk stratification, yet cancer remains a leading cause of disease-related death in children and adolescents in the United States, and many survivors face long-term treatment sequelae. Progress has been uneven, with limited improvement for certain diagnoses such as pediatric and AYA brain tumors and sarcomas. Precision oncology—molecular characterization of tumors paired with targeted therapies—has driven improved outcomes and reduced toxicity in some settings, but efforts in pediatrics are hindered by the rarity and heterogeneity of many solid and CNS tumor types. Since 2013, the Profile Cancer Research Study at Dana-Farber/Boston Children’s has offered clinical-grade targeted next-generation sequencing to all pediatric patients with suspected or confirmed cancer, returning results to clinicians. This study leverages that unselected, institution-wide cohort to characterize the genomic landscape across 95 diagnoses, including many rare and ultra-rare tumors, to identify targetable alterations matched to active precision oncology trials, inform trial design and real-world practice, and contribute standardized data to national data-sharing initiatives such as the NCI Childhood Cancer Data Initiative (CCDI).

Early large-scale pediatric cancer genomics efforts (e.g., TARGET) and pan-pediatric landscape analyses emphasized more common tumors, leading to underrepresentation of rare solid and CNS malignancies in foundational datasets. Basket trials such as Pediatric MATCH, NCI-MATCH, and ASCO TAPUR have established actionable mutation lists and trial arms guided by tumor profiling but rely on broad, diverse accrual to capture rare genotypes and histologies. Recent precision medicine programs (e.g., Genomes for Kids, INFORM, MAPPYACTS, Zero Childhood Cancer) vary in sequencing breadth (panel, WES/WGS, RNAseq) and inclusion criteria (newly diagnosed vs. relapsed/refractory), collectively demonstrating feasibility of integrating molecular profiling into care while highlighting persistent gaps for ultra-rare diagnoses. The need for standardized diagnosis classification (e.g., ICD-O) and robust data-sharing to aggregate genomic and clinical outcomes data is widely recognized as essential to realizing precision oncology benefits in pediatric rare cancers.

Study design and cohort: Pediatric patients (<~25 years) with intracranial (CNS) or extracranial solid tumors presenting between September 2013 and March 2019 were eligible for the Profile Cancer Research Study, which generated clinical-grade targeted NGS reports returned to treating physicians and the EMR. Patients with hematologic malignancies or benign tumors were excluded. Tumor samples were requested from pathology after standard evaluation; acquisition procedures were not altered for the study and typically used leftover FFPE tissue. If multiple samples per patient existed, one was selected (preferably at initial diagnosis and pre-treatment; otherwise, the best-quality earliest pre-treatment relapse sample). Data from a small number of additional patients sequenced under a waiver of consent and 27 patients from a similar study were included. Sequencing: Targeted DNA sequencing was performed using the OncoPanel assay (Center for Advanced Molecular Diagnostics, Brigham and Women’s Hospital; CLIA-certified), across versions 1–3 (up to 447 genes and intronic regions for select SV detection). The panel detects SNVs, indels, copy number alterations, and selected structural variants. The standard variant allele fraction (VAF) reporting threshold was ≥5%, with inclusion of variants down to 2.5% VAF if judged oncogenic by a molecular pathologist. Molecular pathology reports were returned at the time of sequencing. Variant processing and interpretation: Variant call files were generated via institutional pipelines. Additional filtering removed variants present in ClinVar or gnomAD (population variants). Tumor mutation burden was calculated as total SNVs/indels divided by bases analyzed. Variants were annotated and classified as Oncogenic/Likely Oncogenic/Predicted Oncogenic using established knowledge bases (e.g., MSK criteria). OncoPrints were generated using ComplexHeatmap. Diagnosis classification and clinical data: A multidisciplinary committee (oncology, pathology, neuropathology, sarcoma experts) reviewed pathology and uniformly classified diagnoses using ICD-O-3.2, with categorization into intracranial vs. extracranial and further disease subgroups. Clinical/demographic data (sex, race/ethnicity, age, year of diagnosis, histology, prior therapies, treatments) were abstracted from the EMR into REDCap using a curated guide with dual abstraction quality control for ~12% of cases. For selected diagnoses (osteosarcoma, Ewing sarcoma, Wilms tumor, neuroblastoma), disease-specific EMR models (PRISMS) were adapted to pediatric contexts; baseline imaging reports were used to derive Toronto Stage. Trial matching and therapy ascertainment: Genomic alterations were matched against actionable mutation lists (AML) from Pediatric MATCH, NCI-MATCH, and ASCO TAPUR using a two-step approach: (1) precise AML match accounting for resistance variants, including tumor suppressor loss-of-function; or (2) same gene and variant type (activating fusion/amplification/oncogenic SNV/indel in oncogenes; LOF mutation/deletion in tumor suppressors). For matched patients, records were reviewed to determine whether molecularly targeted therapy in the same drug class was received, and via what mechanism (clinical trial, single-patient protocol, or off-label prescription). Generalizability assessment and follow-up: Cohort representation was compared with national pediatric cancer registries (CiNA/NAACCR 1995–2018; SEER). Median follow-up was 38 months (range 0–500). Statistical considerations: No predetermined sample size, no randomization or blinding; no data excluded.

• Sequencing throughput and cohort size: Of 1,120 enrolled pediatric solid/CNS tumor patients, OncoPanel sequencing was successful for 76% (848/1,120). Including additional eligible patients, the final analytic cohort comprised 888 pediatric patients with successful somatic profiling across 95 distinct histologic diagnoses.
• Spectrum of diagnoses: 58% (512/888) extracranial solid tumors; 42% (376/888) CNS tumors. Ten common pediatric cancers accounted for 55% (451/888), while 45% (398/888) spanned 85 rare diagnoses (<25 patients per histology). Diagnoses included entities not covered in prior pan-pediatric sequencing analyses.
• Sample characteristics: Most sequenced samples were from the primary tumor site (~92%); a large proportion were obtained at initial diagnosis prior to treatment. Multiple OncoPanel versions were used (predominantly v2 and v3).
• Genomic landscape: Expected hallmark events recapitulated within diagnoses (e.g., TP53 alterations in osteosarcoma; EWSR1 rearrangements in Ewing sarcoma; ALK mutations and MYCN amplification in neuroblastoma; BRAF fusions and IDH1 mutations in gliomas). Potentially targetable alterations distributed across histologies included PIK3CA activating mutations in 18 cases (~2% overall) and ARID1A inactivating mutations in 10 cases (~1%). In tumors not represented in prior pan-cancer studies, recurrent oncogenic alterations unique to extracranial tumors included CTNNB1, DICER1, NF1; in CNS tumors, CTNNB1, NF2, KIT. Additional potentially actionable events uniquely present in rare tumors included CRLF2 (carcinomas), KIT activating mutations (CNS and non-CNS germ cell tumors), and CTNNB1 alterations (carcinomas, liver, desmoid, craniofacial tumors).
• Trial matching and targeted therapy: 33% (289/888) had at least one genomic variant matching an arm in Pediatric MATCH, NCI-MATCH, or TAPUR. Among matched patients, 14% (41/289) received a molecularly targeted therapy aligned with the identified alteration, obtained via clinical trials, single-patient protocols, or off-label use. Receipt of matched therapy varied by histology.
• Data scale and follow-up: Per-patient EMR abstraction captured extensive pathology and imaging; median follow-up was 38 months (range 0–500), with an average of 3 cancer treatment regimens (range 1–12).

Comprehensive, institution-wide clinical tumor profiling with return of results generated a large, unselected cohort spanning 95 pediatric solid and CNS tumor histologies, including many ultra-rare entities typically underrepresented in prior genomic studies. The analysis reproduced known disease-specific genomic features while uncovering potentially actionable alterations dispersed across diverse histologies, emphasizing the need for basket trial designs and cross-disease therapeutic strategies in pediatrics. One-third of patients harbored alterations matching active precision oncology trial arms, and a subset received matched targeted therapy, demonstrating real-world translation of profiling into treatment across diagnoses. Because many pediatric solid tumors are rare, aggregating data through standardized diagnosis classification (ICD-O-3.2) and robust data-sharing is essential to refine alteration frequencies (e.g., PIK3CA), identify biomarkers of outcome, and better predict therapeutic response or resistance. The study contributes data to national repositories (CCDI/dbGaP; GENIE) and underscores the feasibility and importance of harmonized clinical and genomic data capture to inform current and future precision oncology trials for children and AYAs.

This study profiles 888 pediatric solid and CNS tumors across 95 ICD-O-classified histologies using a clinical targeted sequencing panel, identifies a broad spectrum of oncogenic alterations—including actionable variants spanning both common and ultra-rare tumors—and systematically matches findings to contemporary basket trials. Approximately one-third of patients had trial-matching alterations, with a subset receiving targeted therapy, illustrating the clinical utility of institutional profiling programs. The work advocates for standardized diagnosis ontologies within EMRs and expanded data-sharing to enable sufficiently powered analyses of rare pediatric cancers and to optimize precision trial design. Future directions include harmonizing clinical outcomes with genomic data at scale, refining actionability frameworks for pediatric contexts, and expanding multi-omic profiling to improve detection of targetable drivers and resistance mechanisms.

• Single-institution, hospital-based cohort may limit generalizability despite comparison with national registries.
• Targeted panel sequencing (multiple versions over time) may miss alterations detectable by broader platforms (WES/WGS/RNA-seq) and introduces version-dependent coverage variability.
• Only 76% of enrolled patients had successful sequencing; tissue availability and quality may introduce selection bias.
• Observational design without randomization or blinding; no predefined sample size.
• Treatment matching was contingent on trial availability, clinician choice, and patient eligibility; only a fraction of matched patients received targeted therapy, limiting outcome assessment.
• Tumor-only sequencing with filtering may affect interpretation of variant pathogenicity compared to paired germline analyses.
• Heterogeneity and small numbers within many rare histologies limit precision of alteration frequency estimates.