既往治疗如何塑造儿童肿瘤的突变谱

Fig. 1 | Overview of the KZM cohort. a, Composition of the KZM cohort, comprising samples from the KiCS, ZERO and MSK programmes. Bar charts show sample counts by disease state (primary versus advanced) and treatment status (pre-therapy versus post-therapy) for each dataset. b, Therapy exposure across tumour types in the KZM cohort. Stacked bars show the post-therapy (red) and pre-therapy (blue) samples by tumour type, ordered by total sample count. Post-therapy counts are shown above each bar. Tumour types with n < 2 or no treated samples are grouped under ‘other’. ACC, adrenal cortical carcinoma; AML, acute monocytic leukaemia; ARMS, alveolar rhabdomyosarcoma; ASPS, alveolar soft part sarcoma; ATRT, atypical teratoid rhabdoid tumour; BALL, B cell lymphoblastic leukaemia; CD, chordoma; CPC, choroid plexus carcinoma; DMG, diffuse midline glioma; DSRCT, desmoplastic small round cell tumour; EMBCNS, embryonal central nervous system tumour; EPD, ependymoma; ERMS, embryonal rhabdomyosarcoma; ES, epithelioid sarcoma; EWS, Ewing sarcoma; HB, hepatoblastoma; HCC, hepatocellular carcinoma; HGG, high-grade glioma; LGG, low-grade glioma; MBL, medulloblastoma; MFT, myofibroblastic tumour; MN, meningioma; MPN, myeloproliferative neoplasm; MPNST, malignant peripheral nerve sheath tumour; MRT, malignant rhabdoid tumour; NBL, neuroblastoma; OST, osteosarcoma; PTC, papillary carcinoma of thyroid; SPINDLE, spindle cell tumour, unclear diagnosis; SS, synovial sarcoma; TALL, T cell lymphoblastic leukaemia; US, undifferentiated sarcoma; WT, nephroblastoma (Wilms tumour). c, Empirical cumulative distribution function showing the proportion of patients who received up to a given number of distinct therapeutic agents. Each point on the curve corresponds to a unique therapy burden value and the cumulative percentage of patients who received that number or fewer drugs. Numbers above each step indicate the cumulative percentage up to each drug count. d, Therapy class exposure by tumour type. Heat map (top) showing the number of tumours per major tumour type (rows, same order as in b) that were exposed to each therapy class (columns). Aggregated totals across all tumour types are shown along the bottom. e, Drug- level exposure across the cohort. Each tile represents a therapeutic agent, with colour intensity scaled to the number of patients who received that drug, normalized to the most frequently used agent. Only drugs administered to at least two patients are shown individually; the remaining are aggregated into ‘others (<2)’. Patient counts are shown in each tile. GM-CSF, granulocyte– macrophage colony-stimulating factor; MIBG, metaiodobenzylguanidine.

Fig. 2 | Mutational burdens across treatment status and cancer types. a, Somatic mutation burden (total number of variants divided by the length of the reference genome) are shown for major types of genomic variants. In addition to the overall burden in the cohort, we also show the burdens for treatment-naive (primary naive), and treated primary (primary treated) and advanced (advanced naive and advanced treated), tumours. Each point represents the burden for a single sample, and the red horizontal lines indicate the median burden. The number of samples in each category is shown below the horizontal axis. The dependent variable (mutation burden) was first rank- transformed, and the effects of covariates (age, sex and tumour purity) were regressed out via ordinary least squares linear regression. The resulting residual ranks were then compared across groups using a one-way ANOVA (F-test; two- sided). For post hoc comparisons, pairwise Wilcoxon rank-sum tests (two- sided) were performed, with P values adjusted for multiple comparisons using the Benjamini–Hochberg false discovery rate procedure at α = 0.05. b, Somatic SBS burden for major tumour types (those with a minimum of five primary naive and five advanced treated samples). All figure elements are the same as in a. P values are calculated as in a and indicate the significant increase in SBS burden post-therapy. As only two groups were compared (pre-therapy versus post-therapy for each tumour type), no post hoc pairwise test or adjustment for multiple comparisons was performed.

Fig. 3 | Mutational activity and therapy associations of COSMIC and novel mutational signatures in paediatric cancers. a, Somatic mutations attributable to COSMIC or novel mutational signatures. The proportions of mutations found in low-burden tumours (n = 577) in the cohort irrespective of past therapy exposure are shown. These are divided into mutations associated with any therapy (including platinum) versus those specifically associated with platinum therapies, and then further divided by mutation type (SBS and DBS). b, Mutation burden (somatic mutations per Mb) of therapy-associated SBS signatures (COSMIC and novel) detected across all tumour genomes (n = 611). Each dot represents one sample; red lines indicate median values and dashed lines show reference thresholds. The number of samples carrying each signature is shown below the horizontal axis. c, Association between mutational signatures and therapeutic classes. Each circle represents a significant association (one-tailed empirical P value < 0.05, AUROC ≥ 0.7 and regression coefficient ≥ 0.7), with size proportional to the regression coefficient (log of odds ratio; shown in circles) and colour indicating the observed AUROC scores from the logistic regression model. Novel and COSMIC signature names are shown in green and black, respectively. d, Same as c, but for individual therapeutic agents. Beyond the primary therapy associations detailed in the main text, we identified secondary associations that either failed robustness testing, arose from treatment co-occurrence rather than direct causation (for example, DBS5 with antimetabolites) or reflected well-known endogenous processes (for example, SBS13 with anthracyclines). These are further discussed in Supplementary Note 1.7.

Fig. 4 | Mutational signature analysis of platinum exposure. a, All the platinum-treated KZM samples with available detailed therapy information are shown. See Supplementary Table 1 for the full tumour names. The top right quadrant depicts where we expect the tumours to manifest the imprints of platinum therapy (samples that pass both burden and time thresholds). A summary plot of the number of samples from this quadrant with (Sig+) and without (Sig−) platinum signatures for each cancer type is provided on the right-hand side. AF, aggressive fibromatosis; COST, chondroblastic osteosarcoma; CRC, colorectal carcinoma; GCT, germ cell tumour; HOG, Hodgkin lymphoma; MAO, mucinous adenocarcinoma (ovarian); NE, neuroepithelioma; Neo, malignant neoplasm; PBL, pineoblastoma. b, Left, comparison of platinum therapy against all therapies combined (including platinum therapies), showing the percentage of KZM participants who exhibit therapy-associated signatures over time. Key milestones, including signature prevalence at 12 months and time to 50% prevalence, are marked in red. Right, the same analysis for two other major therapeutic classes: anthracyclines and antimetabolites. Platinum-specific signatures were used for the platinum curve, and the composite set of all therapy signatures (Fig. 3b) was used for other therapy curves owing to the absence of known anthracycline-specific mutational signatures and low abundance of antimetabolite signatures in our cohort. Mo, months.

Fig. 5 | Platinum mutational signatures and clinical outcome. a, Enrichment of two key platinum resistance genes (GSTA1 and ABCC2) in signature-positive versus signature-negative tumours. Signature+ > thresholds, n = 62; signature− > thresholds, n = 50, and signature− < thresholds, n = 24. The first two columns are signature-positive and signature-negative samples from the top right quadrant of Fig. 4a, that is, samples that passed both time and burden thresholds. The third column are samples that did not pass the thresholds and did not show platinum signatures. Differential expression was assessed using a two-sided Wald test; P values were adjusted using the Benjamini–Hochberg false discovery rate procedure. Box plots show the median (centre line), interquartile range (box bounds: 25th–75th percentile) and 1.5× interquartile range from the box bounds (whiskers). Individual data points are overlaid as dots. Violin plots show the kernel density estimation of the data distribution. Padj, adjusted P value. b, Worsening of clinical response in signature-positive tumours. All paediatric patients with progressive disease (PD) sampled more than 91 days post-therapy showed strong platinum signatures. Six additional patients who were signature-positive who were initially recorded as partial response (PR), complete response (CR) or stable disease (SD) later developed recurrence or metastasis. No platinum signature was detected in five complete response or partial response tumours, suggesting either treatment sensitivity or the presence of sub-detection-threshold resistant clones. WGS, whole- genome sequencing. c, In a cohort of 166 patients treated with platinum (Hartwig Medical Foundation) who developed metastases and were re-exposed to platinum therapies, the presence of platinum signatures in the primary tumour was consistently associated with worsening clinical response in the metastasis. At the time of last clinical evaluation, patients with platinum signature (compared to those without) had 2.24-fold higher odds of progressive disease versus stable disease and 2.88-fold higher odds of progressive disease versus partial response or complete response. The association between platinum signatures and progression remains significant after adjusting for primary histology (Cochran–Mantel–Haenszel P value = 0.0097; odds ratio (OR) = 3.41), metastatic site (Cochran–Mantel-Haenszel P value = 0.0176; OR = 3.20), and platinum treatment frequency (multivariate logistic regression P value = 0.018; OR = 2.98). All odds ratios are reported with 95% confidence intervals (CIs); P values are two-sided. NS, not significant.