Mutational topography reflects clinical neuroblastoma heterogeneity

Abstract

Neuroblastoma is a pediatric solid tumor characterized by strong clinical heterogeneity. Although clinical risk-defining genomic alterations exist in neuroblastomas, the mutational processes involved in their generation remain largely unclear. By examining the topography and mutational signatures derived from all variant classes, we identified co-occurring mutational footprints, which we termed mutational scenarios. We demonstrate that clinical neuroblastoma heterogeneity is associated with differences in the mutational processes driving these scenarios, linking risk-defining pathognomonic variants to distinct molecular processes. Whereas high-risk MYCN-amplified neuroblastomas were characterized by signs of replication slippage and stress, homologous recombination-associated signatures defined high-risk non-MYCN-amplified patients. Non-high-risk neuroblastomas were marked by footprints of chromosome mis-segregation and TOP1 mutational activity. Furthermore, analysis of subclonal mutations uncovered differential activity of these processes through neuroblastoma evolution. Thus, clinical heterogeneity of neuroblastoma patients can be linked to differences in the mutational processes that are active in their tumors.

Keywords: cancer genomics; clinical heterogeneity; complex rearrangements; ecDNA; mutational processes; mutational signatures; neuroblastoma; structural variation; tumor evolution.

Graphical abstract

Figure 1

Cohort characteristics and analysis strategy Top: description of the distribution (in percentage) of the 150 neuroblastoma samples from the discovery (n = 114; see also Figure S1) and validation (n = 36) cohorts, within sex groups (female, male, and unknown), risk groups (HR MNA, HR non-MNA, and non-HR: low-risk; low-risk stage 4S; intermediate-risk), and age at diagnosis groups (<1 year old, 1–5 years old, and >5 years old). Middle: summary of the sequencing datasets available for our analysis. WGS matched tumor-normal pairs. Bottom: description of the main steps carried out in our study, starting with the variant discovery and the mutational signatures analysis, followed by validation of the signatures and subclonal SNV-based signatures extraction, characterization of the complex rearrangements present in our samples, unsupervised clustering, and definition of the three clinically relevant neuroblastoma mutational scenarios presented in this work.

Figure 2

Distribution and correlation of SNV-, indel-, SV-, and CNA-associated signatures in neuroblastoma clinical risk groups (A) Exposure (in percentage) of the four SNV-associated signatures (SBSs) identified in our neuroblastoma discovery cohort by clinical risk group (n = 114). Each color displays a different signature: SBS3, SBS5, SBS18, and SBS40 (see also Figures S2A–S2C). Columns are ordered by neuroblastoma clinical risk classification. (B) Exposure (in percentage) of the six indel signatures (ID; insertions and deletions <50 bp) identified in our cohort by clinical risk group. Each color displays a different signature: ID1, ID2, ID4, ID6, ID8, and ID9 (see also Figures S2D–S2F). Columns are ordered by neuroblastoma clinical risk classification. (C) Exposure (in percentage) of the eight CNA signatures (CX; gains, losses, amplifications, and homozygous deletions) extracted in our cohort by clinical risk group. Each color displays a different signature: CX1, CX2, CX3, CX5, CX7, CX11, CX14, and CX15 (see also Figures S2G and S2H). Columns are ordered by neuroblastoma clinical risk classification. (D) Exposure (in percentage) of the six SV signatures (SV; deletions, duplications, translocations, and inversions) identified in our cohort by clinical risk group. Each color displays a different signature: SV1–SV6. SV2, and SV3 correspond to reference signatures R6b and R6a, respectively (see also Figures S3A–S3D). Columns are ordered by neuroblastoma clinical risk classification. (E) Heatmap depicting the positive (red) and negative (blue) correlations between the signatures associated with different types of mutations (SNV, indel, SV, and CNA). Colors display the Spearman correlation coefficient. Only significant correlations are included (p < 0.05, false discovery rate [FDR] correction). (F) Activity trajectories of the four SNV-associated signatures (SBSs) per cancer cell fraction by clinical risk group in the validation cohort (n = 36). Thick lines correspond to average exposure for all samples. Thin lines correspond to per-sample exposure (see also Figure S4).

Figure 3

Co-occurrence of mutational signatures, complex rearrangements, and cancer-related gene alterations in neuroblastoma (A) Heatmap showing the correlations between mutated neuroblastoma driver genes and DNA-damage-repair genes and all the mutational signatures and complex rearrangement types identified in our cohort (n = 114). Below are rows showing the correlation between HRD probability score, mutations in HRR genes, and the mutational signature exposures and complex rearrangements identified in our cohort. In both heatmaps, colors display the Spearman correlation coefficient. Only significant correlations are included (p < 0.05, FDR correction). (B) Heatmap depicting the correlations between the signatures associated with different variant types (SNV, indel, SV, and CNA) and the nine complex rearrangement classes identified in our cohort (n = 114). (C) Box plot comparing the distribution of HRD probability scores across neuroblastoma risk groups (HR MNA, HR non-MNA, and non-HR). Each dot represents a patient. To assess whether there are differences between risk groups, we used the non-parametric Kruskal-Wallis test (p value in the upper-left corner). (D) Frequency of patients with mutated HRR genes across the three neuroblastoma risk groups. (E) Box plot comparing the distribution of HRR mutated genes across neuroblastoma risk groups (HR MNA, HR non-MNA, and non-HR). Each dot represents a patient. To assess whether there are differences between risk groups, we used the non-parametric Kruskal-Wallis test (p value in the upper-left corner). Pairwise comparisons were done using the non-parametric Wilcoxon rank-sum test. Significance: ∗p < 0.1, ∗∗p < 0.05, ∗∗∗p < 0.01. All analyses were performed in the discovery cohort (n = 114).

Figure 4

Co-occurrence and distribution of complex rearrangements in neuroblastoma (A) Top: frequency of SVs involved in each complex rearrangement type, in percentage. Bottom: pie chart showing the percentage of SVs involved in complex and simple events in our cohort. (B) Upset plot depicting the co-occurrence of the different types of complex SV patterns within patients. The number of patients with each combination of rearrangements is shown in the top histogram (colors display the risk group for each patient). (C) Pie chart showing the frequency in percentage of each complex rearrangement type in the whole neuroblastoma cohort. (D) Relative frequency in percentage of the nine different complex rearrangement types identified in our cohort by clinical risk group (HR MNA, HR non-MNA, and non-HR). All analyses were performed in the discovery cohort (n = 114). See also Figure S5.

Figure 5

Genomic distribution of topographically defined complex rearrangement patterns in neuroblastoma (A) Density plot showing the number of regions affected by complex rearrangements per chromosome across the whole human genome. Each row/color corresponds to a different type of rearrangement. (B) Frequency of patients with complex rearrangements affecting neuroblastoma driver genes or DNA-damage-repair genes. Each column and color corresponds to a different type of rearrangement. All analyses were performed in the discovery cohort (n = 114). See also Figure S5.

Figure 6

Characterization and definition of mutational scenarios linked to clinical heterogeneity (A) Unsupervised clustering analysis (hkmeans method; k = 3) obtaining three mutational scenarios from the scaled mutational signatures exposure and complex rearrangements. All mutational signatures from different variant types and complex rearrangements detected in our cohort are included in the analysis. Color scale indicates frequency. (B) Summary of the defining features for each of the three mutational scenarios/clusters. Color gradation corresponds to clustering distance. (C) Summary of the three mutational scenarios described in our neuroblastoma cohort, including different characteristic features such as mutational signatures, complex rearrangements, mutational processes, and risk classification correspondence associated with each of them. CRs, complex rearrangements. (D) Univariate Cox proportional hazards model. Forest plot shows the proportional risk of the three mutational scenarios and the three neuroblastoma risk groups. (E) Kaplan-Meier survival curves showing the clinical impact of the three mutational scenarios and the three neuroblastoma risk groups. (p = 0.00087 and p < 0.0001, respectively by log-rank test). Colors on Kaplan-Meier plot display each condition. All analyses were performed in the discovery cohort (n = 114).