The clinical utility of integrative genomics in childhood cancer extends beyond targetable mutations

Abstract

We conducted integrative somatic-germline analyses by deeply sequencing 864 cancer-associated genes, complete genomes and transcriptomes for 300 mostly previously treated children and adolescents/young adults with cancer of poor prognosis or with rare tumors enrolled in the SickKids Cancer Sequencing (KiCS) program. Clinically actionable variants were identified in 56% of patients. Improved diagnostic accuracy led to modified management in a subset. Therapeutically targetable variants (54% of patients) were of unanticipated timing and type, with over 20% derived from the germline. Corroborating mutational signatures (SBS3/BRCAness) in patients with germline homologous recombination defects demonstrates the potential utility of PARP inhibitors. Mutational burden was significantly elevated in 9% of patients. Sequential sampling identified changes in therapeutically targetable drivers in over one-third of patients, suggesting benefit from rebiopsy for genomic analysis at the time of relapse. Comprehensive cancer genomic profiling is useful at multiple points in the care trajectory for children and adolescents/young adults with cancer, supporting its integration into early clinical management.

Fig. 1. Overview of the KiCS study cohort and enrollment outcomes of referred patients.

Entry Point 2: Suspicion for cancer predisposition syndrome. VUS, variants of uncertain significance.

Fig. 2. The KiCS study cohort: tumor and sample characteristics and summary of actionable findings.

a, Each row, labeled by study ID and diagnosis (based on pathology report), corresponds to a study participant. The first four columns describe the tumor samples for each patient. Samples are arranged chronologically from left to right. Color indicates the disease state: green, initial diagnosis; blue, relapse; orange, progressive disease. Squares correspond to samples from the primary tumor site, and circles represent samples of metastatic sites. A star indicates a sample collected after the patient had received cancer-directed therapy. EP2 patients with multiple prior malignancies were classified according to their most recent tumor diagnosis. EP2 participants with no cancer diagnosis are coded as ‘EP2’ and are denoted with a gray dash if there was no accompanying tumor sample. The last column indicates participants with at least one actionable finding (red checkmark). Please see Supplementary Table 1 for a full list of tumor types and acronyms, categories of actionable findings and additional demographic information. n = 300 participants. Three participants (KiCS 32, 220 and 334) each had two primary tumors and are each represented twice. b, Frequency of actionable findings, by class of clinical utility. The height of each histogram is the percentage of patients with at least one actionable finding in that category. Patients with an actionable variant having more than one aspect of utility are recorded in each relevant category but are only counted once in the ‘any’ clinical utility class. The colors within each histogram represent the proportion of variants in that class detected by each NGS technology. For the A, B and C categories, percentages were calculated on the basis of a denominator of n = 264 (participants with somatic analysis). For the D and ‘any’ categories, percentages were calculated on the basis of a denominator of n = 300 total study participants. Source data

Fig. 3. Oncoprint visualization of category C (therapeutic) clinically actionable findings and BRCAness in pediatric and AYA cancers in the KiCS cohort.

a, Oncoprint visualization of the distribution of therapeutically actionable findings (category C). Findings are arranged in rows and grouped by the therapeutic agent indicated by each finding. Patients (n = 143; two patients with two primary tumors each represented twice) are arranged in columns. The top bar plot indicates the number of relevant mutations in each patient (that is, the number of variants that constitute therapeutic biomarkers in each patient). Some variants contribute together as a single actionable finding, for example, PTEN SNV and PTEN loss in KiCS 366 and MSH2 germline and somatic SNVs, POLE SNV and ultra-hypermutation in KiCS 284. Variant details are depicted in Supplementary Table 5. The right-side bar plot depicts the number of patients harboring a finding. Red square, amplification; blue square, loss; pink vertical rectangle, fusion; yellow triangle, germline SNV/indel; green triangle, somatic SNV/indel; black border, homozygous mutation; brown square, hypermutation. Please see Supplementary Table 1 for a full list of tumor types and acronyms. b, The proportion of COSMIC single-substitution signature 3 (BRCAness mutational signature) in the PCAWG dataset compared to KiCS cohort patients with absence or presence of either somatic or germline variants affecting the HR pathway. The KiCS cohort is divided into three subsets based on the absence or presence of germline or somatic HRD. The proportion of samples with the SBS3 signature in each KiCS subset as well as the PCAWG dataset is shown by the height of the bars. Sample sizes (that is, the number of biologically independent samples) are shown on the x axis. Only statistically significant P values obtained by Fisher’s exact test (two sided) comparing each pair of datasets are shown. Source data

Fig. 4. Comprehensive sequential tumor analysis provides important insights into the relationship between multiple neoplasms of the same histology presenting in the same individual (category Dii).

Results are shown for n = 4 patients. HSCT, hematopoietic stem cell transplantation; IGH-r, IGH rearrangement.

Fig. 5. Impact of therapy on TMB in pediatric and AYA cancers.

a, Bar charts: KiCS tumor samples ordered by SNVs per Mb, displaying SNVs per Mb and structural variant (SV) count on the y axis. Red bars indicate samples that had a high mutational load for at least one of the two mutation types. Pie charts: combined red slices indicate the proportion of samples with high mutational load for respective mutation type. The dark red slices indicate the proportion of samples with a high mutational load for only one of the mutation types (that is, not high for the other mutation type). For a, n = 326 individual tumor samples from 249 patients. b,c, Box plots showing SNVs per Mb (b) and structural variant count (c) for samples obtained before treatment versus after treatment (chemotherapy and/or radiation). Wilcoxon two-tailed P values are shown. Box plots show quartiles with the center line representing the median and whiskers representing 1.5 times the interquartile range. For b, n = 326 individual tumor samples from 249 patients; test statistic (z score) = 2.81; effect size = −0.156. For c, n = 217 individual tumor samples from 180 patients; test statistic (z score) = 2.65; effect size = −0.180. d, Example of a tumor sampled before and after treatment. Light blue indicates mutations seen before treatment and dark blue indicates new mutations present at relapse (after treatment). Source data

Fig. 6. Evolution of childhood cancers across time.

a, Each row corresponds to a patient with a single tumor diagnosis (n = 38 patients). Pie charts represent samples analyzed by cancer panel for each tumor where SNVs were detected at a VAF of greater than 0.10. Pie charts are colored by the proportion of mutations identified at each time point that were private to that sample (red) or shared with at least one other sample for that tumor, at a VAF of greater than 0.05 (black). Open circles represent no SNVs detected above the threshold. The center panel depicts samples (gray squares) in sequential order with time on the x axis, showing the number of days since the initial sample was obtained. A star represents the emergence or loss of a targetable driver, leading to a potential change in clinical action. Note that samples obtained at the same time point (days since diagnosis) correspond to anatomically distinct lesions (for example, local relapse versus lung metastasis). b, Proportion of mutations shared by each primary tumor with its paired relapse (n = 25 individual patients with 38 initial tumor–relapse pairs). Using WGS data, initial tumor samples were compared to relapse samples (with a one-to-one comparison comprising a ‘pair’). The proportion of SNVs from the initial sample shared with the paired relapse sample is characterized as parallel or linear on the basis of a 75% threshold. Source data

Extended Data Fig. 1. KiCS cohort characteristics.

(a) Distribution of tumor types (n = 303; 3 participants had two independent primary tumors. N = 186 for solid tumors, n = 67 for CNS tumors, n = 50 for leukemia/lymphomas). OTHER (Solid Tumors): ACC, AF, DCIS, DSRCT, EMC, EMS, ES, GCT, HCC, IDC, LB, LS, MAO, MFT, MM, MPNST, MRT, N, OF, PFHT, PGL, PT, RB, RCC, SBC, SCST, SN. OTHER (CNS Tumors): AM, CPC, CSRC, GG, GS, MN, MT, N, NE, SW. OTHER (Leukemia/Lymphoma): BPDCN, CML, LPD, MPAL. Please see Supplementary Table S1 for a full list of tumor types and acronyms. (b, c) Representative example of prior therapeutic exposures of sequenced samples (N = 232 tumor samples from 176 patients). (d) Age distribution (N = 300 participants). (e) Ethnicity distribution (based on self-reported ethnicity provided by 41% of study participants, N = 123 participants (59% missing data)). Source data

Extended Data Fig. 2. Sample sequencing and purity.

(a) Venn diagram of sequencing technologies used. The proportion of patients whose tumors were analyzed using each combination of cancer-panel (CP -pink), whole-genome (WGS – green) and whole-transcriptome (WTS –violet) sequencing, and the proportion of patients whose germline DNA was analyzed using each combination of CP and WGS. (b) KiCS cohort purity distribution by tumor class. N = 292 tumor samples. Greater than 75% of samples have a purity of over 25%. Source data

Extended Data Fig. 3. Therapeutic biomarkers in the KiCS cohort.

(a) Level of evidence for category C variants (therapeutic biomarkers) and proportion targeted with a therapeutic agent (N = 222 variants). Details about levels of evidence are available in Extended Data Fig. 4. (b) Frequency of targets in various therapeutic agent classes. Targets in each category (not inclusive list): PARP inhibitors: BRCA1, BRCA2, PTEN, CHEK2, PALB2, ARID1A, RAD51C, RAD51D, BARD1, BRIP1, ATM, ATRX variants. MEK/ERK inhibitors: NF1/2, N/K/HRAS, MAP2K1, PTPN11 variants, BRAF fusions. Immune checkpoint inhibitors: Tumor mutational burden >5 mutations/Mb, or biallelic loss of a MMR gene. CDK4/6 inhibitor: CDK4/6 amplification, CCND2 or CCND3 amplification, CDKN2A/B homozygous deletion. Growth TKI: KIT, FGFR, PDGFR, RET, EGFR, VEGFR, FLT3 variants/amplification and ABL fusions. ALK inhibitors: ALK hotspots, fusions, MET and ROS1 fusions. PI3K/AKT/mTOR inhibitors: PIK3CA, PIK3R1, AKT1, FBXW7, PTEN, TSC1/2 variants, ARID1A variants (in addition to genes targetable by MEK/ERK inhibitors above). EZH2 inhibitor: EZH2, SMARCB1, SMARCA4, ARID1A variants. BRAF inhibitor: BRAF variants. NTRK inhibitors: NTRK fusions. Other: IDH1 inhibitor, WEE1 inhibitor, Wnt pathway inhibitor, Hif inhibitor, BRD and BET inhibitor, SHH inhibitor. Source data

Extended Data Fig. 4. Classification of actionable variants.

Classification is adapted from AMP/ASCO/CAP, OncoKB guidelines and NCI/COG Pediatric Match levels of evidence, and with reference to ACMG/ClinGen criteria^,,,.

Extended Data Fig. 5. Examples of clonal evolution in the KiCS cohort.

(a) Clonal evolution of an embryonal rhabdomyosarcoma in an individual with constitutional Neurofibromatosis-1 (KiCS 15). Upper panel: Ancestral ‘drivers’ are detected early and maintained through tumor evolution regardless of spatial evolution (FGFR4 SNV, chr17p-). New clones emerge at progression with additional SNV/CN changes. Tumor progression within the same spatial region show significant SNV/CN commonalities (D and E), while others despite temporal/spatial changes show important similarities with ancestral clones (C and F). Lower panel: The phylogenetic tree annotated with SNV drivers. Each node represents a subpopulation whereby arrows extend from ancestors to descendants. This is supportive of the findings that although ancestral driver events may diminish or be displaced during tumor progression, they may also be maintained as subclones that drive relapse. (b) Parallel and linear evolution in multi-sample KiCS cases (N = 2 individual cases; one case shown on each side of the vertical line). Phylogenetic trees from WGS annotated with SNV drivers, number of SNVs comprising each tree node, and the number of SNVs unique or shared between paired initial and subsequent tumor samples. Left panel: KiCS 319 represents linear evolution, with 79.9% of initial SNVs shared with the relapse. In the KiCS cohort, linear evolution is uncommon; only 4 patients shared >75% SNVs between initial and relapse tumor samples. Right panel: Chronic myelogenous leukemia (CML) was initially diagnosed in January 2014 for KiCS 35, followed by development of precursor B-ALL sampled in April 2014 and a relapse of B-ALL in July 2018. The initial CML shares 35.4% and 38.8% of SNVs respectively with each subsequent sample. KiCS 35 displays extreme branching and parallel evolution.

Extended Data Fig. 6. Categories of clinical actionability used to broadly classify sequencing variants.

Four classes of actionability were considered to classify the clinical relevance of sequencing variants.

Extended Data Fig. 7. KiCS program internal website and database for automated variant prioritization.

(a) Summary and Quality Metrics tab on website. (b) Germline website – Variants of Interest tab. (c) Somatic website – Variants of Interest tab.

Extended Data Fig. 8. Somatic variant interpretation pipeline.

Somatic variant interpretation scheme (Class I-V) is further detailed in Extended Data Fig. 4. VUS – clinical actionability of this variant is currently unknown; however, the variant may be biologically relevant in the patient’s tumor type. Unknown – no known association of this variant with cancer.

Extended Data Fig. 9. Germline variant interpretation pipeline.

Abbreviations: VOI – variant of interest; VUS – variant of uncertain significance; ACMG – American College of Medical Genetics; HGMD – Human Genetic Mutation Database; UTR – untranslated region.

Extended Data Fig. 10. Agreement between whole genome and cancer panel tumor mutation burden.

(a) Scatter plot of variant allele fraction from variants discovered on cancer panel target regions from 242 tumors from 193 patients detected on both cancer panel sequencing and whole genome sequencing. Axis plots are count histograms of variants at certain variant allele fraction bins. Variants discovered from one platform, but not the other, have the NULL value replaced with zero to allow plotting of the point, but these variants were not used within the correlation calculation. (b) Correlation of cancer panel tumor mutation burden to whole genome sequencing tumor mutation burden across 201 tumor samples. Only samples where cancer panel tumor mutation burden was greater than zero are plotted due to log conversion. Line indicates linear regression fit of these samples, Pearson (two sided) correlation R = 0.810, p = 4.888E⁻⁴⁸. The Pearson (two sided) correlation and significance of non-log transformed data, including cancer panel samples with TMB of zero (303 tumor samples) was R = 0.956, p = 6.281E⁻¹⁴⁹. Source data