Multimodal analysis of cell-free DNA whole-genome sequencing for pediatric cancers with low mutational burden

Abstract

Sequencing of cell-free DNA in the blood of cancer patients (liquid biopsy) provides attractive opportunities for early diagnosis, assessment of treatment response, and minimally invasive disease monitoring. To unlock liquid biopsy analysis for pediatric tumors with few genetic aberrations, we introduce an integrated genetic/epigenetic analysis method and demonstrate its utility on 241 deep whole-genome sequencing profiles of 95 patients with Ewing sarcoma and 31 patients with other pediatric sarcomas. Our method achieves sensitive detection and classification of circulating tumor DNA in peripheral blood independent of any genetic alterations. Moreover, we benchmark different metrics for cell-free DNA fragmentation analysis, and we introduce the LIQUORICE algorithm for detecting circulating tumor DNA based on cancer-specific chromatin signatures. Finally, we combine several fragmentation-based metrics into an integrated machine learning classifier for liquid biopsy analysis that exploits widespread epigenetic deregulation and is tailored to cancers with low mutation rates. Clinical associations highlight the potential value of cfDNA fragmentation patterns as prognostic biomarkers in Ewing sarcoma. In summary, our study provides a comprehensive analysis of circulating tumor DNA beyond recurrent genetic aberrations, and it renders the benefits of liquid biopsy more readily accessible for childhood cancers.

Trial registration: ClinicalTrials.gov NCT02613962.

Conflict of interest statement

The authors declare no competing interests.

Figures

Fig. 1. Whole-genome sequencing of cfDNA enables fragment-based liquid biopsy analysis in Ewing sarcoma.

The top row of the figure describes the fragmentation and fragment-based analysis of cfDNA in cancer patients. The center row introduces the study cohort (center left) and illustrates the quantification of tumor-derived DNA based on genetic evidence, which is used as a reference in this study (center right). The bottom part of the figure outlines three complementary approaches to fragment-based cfDNA analyses: global fragment-size distribution; regional fragment-size distribution along the genome; and fragment coverage at regions-of-interest (bottom left). CNA profiles were used for comparing cfDNA to matched tumors biopsies and for time-resolved monitoring of tumor evolution. Fragment-based cfDNA metrics were combined for machine learning-based tumor detection and classification (bottom right). The main figures describing each of the analyses are indicated in brackets.

Fig. 2. Global fragment-size analysis detects highly fragmented EwS tumor DNA.

a Histogram (top) showing the cfDNA fragment-size distribution for three representative samples with high (95%), low (2%), and undetectable (0%) tumor-derived DNA (ctDNA) content. The range of cfDNA fragment sizes in 22 healthy controls is shown in gray. Heatmap (bottom) showing the relative fragment-size distribution of 235 cfDNA samples subjected to whole-genome sequencing, each normalized against the median of 22 healthy controls. EwS samples are grouped by genetically inferred tumor-derived DNA content. The three samples shown in the histogram are marked by arrows. b Proportion of short cfDNA fragments (20-150 bp) for pediatric sarcomas and healthy controls (data from this study) and for adult cancers (published data). Boxes correspond to interquartile ranges (IQR), black lines indicate the median, and the whiskers extend to the lowest or highest data points that are still within 1.5 IQR of the bottom or top quartile, respectively. Significance versus the 22 healthy controls was assessed using two-sided Mann–Whitney U tests.

Fig. 3. CNA profiles in liquid biopsies reflect tumor aberrations and allow monitoring of tumor evolution.

a Comparison of CNAs detected in cfDNA versus matched tumor biopsies. Only sample pairs with tumor-derived DNA detected in cfDNA based on ichorCNA are shown (n = 29 sample pairs); four copy-number neutral sample pairs were omitted from the plot. Patients are grouped according to CNA state in cfDNA relative to the matched tumor biopsy. Gray represents CNAs detected in both cfDNA and matched tumor biopsy, orange indicates CNAs detected only in cfDNA, and turquoise represents CNAs detected only in the tumor biopsy. The CNAs detected in cfDNAs versus matched tumor biopsies are summarized in a bar plot (bottom). b CNA plot (ichorCNA) of an EwS cfDNA sample (EwS_90_1) before (middle) and after (bottom) in silico size selection to the range of 90–150 bp. A subclonal CNA on chromosome 16 (indicated by black arrows) that was clearly visible in the tumor biopsy (top) became detectable in the matched cfDNA sample only after in silico size selection. c CNA profiles (ichorCNA) of longitudinal cfDNA samples derived from the same patient (EwS_5) support the monitoring of somatic clonal evolution for individual patients. The filtered CNA profiles of samples collected at diagnosis and two subsequent relapses are shown. The day of sample collection relative to the day of diagnosis is indicated (left). Inferred chromosomal gains are shown in red, inferred deletions are shown in green, and CNA-neutral regions are shown in blue.

Fig. 4. Regional fragment-size analysis detects an EwS tumor-specific epigenetic signature in cfDNA samples.

a Schematic illustration of the regional fragment-size analysis, measuring the ratio of short (S) versus long (L) cfDNA fragments in 100 kb bins along the genome. Genomic regions that overlap with CNAs are excluded in order to focus the analysis on epigenetic signatures reflected in the cfDNA fragmentation patterns. b Heatmap comparing the genome-wide fragmentation profiles of cfDNA samples from patients with pediatric sarcoma to those of healthy controls. In each 100 kb bin (n = 20,706 bins), the log2(S/L ratio) of each sarcoma sample was compared to the distribution of log2(S/L ratios) of healthy controls via z-scores. Both CNA-affected and CNA-neutral bins are shown. EwS samples are grouped by genetically inferred tumor-derived DNA content. c Regional cfDNA fragmentation in patients with pediatric sarcoma compared to healthy controls. Only chromosome arms that are recurrently affected by CNAs in EwS tumors are shown. Box plots illustrate z-scores for EwS samples with genetic tumor evidence and without detected CNAs on the chromosomal arm (red), non-EwS sarcomas with genetic tumor evidence and without detected CNAs on the chromosomal arm (black), EwS samples without genetic tumor evidence (yellow), and healthy controls (green). The significance of the first group versus each of the other three groups was assessed using the two-sided Mann–Whitney U test; Bonferroni-corrected p-values are shown. Boxes correspond to interquartile ranges (IQR), thick black lines indicate the median, and the whiskers extend to the lowest or highest data point that are still within 1.5 IQR of the bottom or top quartile, respectively. d Functional enrichment analysis for regions with significantly shorter/longer cfDNA fragment size compared to healthy controls based on the LOLA software. A selection of enriched terms is shown, while the full list is provided in Supplementary Data 6. e EwS tumor-specific epigenome profiles for selected regions with significantly shorter cfDNA fragment size compared to healthy controls. The genome browser profiles show open chromatin-associated histone H3K27 acetylation (for regions with shorter fragments) based on ChIP-seq data for primary EwS tumors. EwS-specific DHSs along the selected genomic region are also indicated.

Fig. 5. Fragment analysis for EwS-specific genomic regions quantifies tumor-derived cfDNA in EwS patients.

a Conceptual outline of the LIQUORICE method and software for fragment analysis of cfDNA based on tumor-specific epigenetic alterations. b Aggregated, bias-corrected, and normalized coverage signals at selected genomic region sets shown for healthy controls, for non-EwS sarcomas, and for EwS cfDNA samples. EwS samples are grouped by genetically inferred tumor-derived DNA content and clinical tumor evidence. cfDNA samples with coverage signals significantly different (three standard deviations) from healthy controls are displayed in red; the total number of those samples and the direction of the deviation (arrow) are indicated. Total dip depth was used as the metric of choice for the sharp dips at hematopoietic-specific and universal DHSs; area over the curve (AOC) was used for the other region sets. c cfDNA-based coverage signal at EwS-specific DHSs (bottom, nEwS = 38) reflects the aggregate DNA methylation profiles at these regions in matched tumor biopsies (top, nEwS = 38). d Scatterplots showing the correlation of the coverage signal at EwS-specific DHSs with the genetically inferred tumor-derived DNA content of the cfDNA samples. Pearson correlation coefficients (r) and linear trend lines are shown. The x-axes are shown in a log scale from 1% onwards. e Same as d but showing the coverage signal at hematopoietic-specific DHSs. Blue arrows indicate samples with significant liver signature. f Aggregated, bias-corrected, and normalized coverage signal (AOC) at alveolar rhabdomyosarcoma (ARMS)-specific DHSs for cfDNA samples from healthy controls and patients with EwS, RMS, and other pediatric sarcomas (left; p-values were calculated using two-sided Mann–Whitney U tests without correction for multiple testing). For ARMS patients with at least 9% ctDNA (genetic-based evidence), a striking reduction of fragment coverage was observed (right). A cfDNA sample from a patient with embryonal rhabdomyosarcoma (ERMS) did not show any reduction of fragment coverage at ARMS-specific DHSs (bottom right).

Fig. 6. Fragment-based analysis of cfDNA enables accurate tumor detection and classification.

Prediction performance of machine learning classifiers trained to distinguish patients with EwS from healthy controls (a) and from patients with other pediatric sarcomas (b), based on the following sets of input features: global fragment-size distribution (blue); fragment coverage at EwS-specific DHSs (orange); read depth in 5 Mb bins (green); and regional fragmentation patterns (red). Results are also shown for a meta-learner combining the predictions of all individual classifiers into a weighted consensus prediction (purple). The performance of each model was evaluated by and averaged over 100 iterations of bootstrapping, separately for the different sequencing coverage levels (median of 12×, 1×, and 0.1×). CI is the 95% confidence interval obtained by bootstrapping. a ROC curves show, for each feature set, the performance for distinguishing between cfDNA samples from patients with clinical evidence for EwS (nsamples = 103) and healthy controls from three independent sets (22 controls sequenced in this study; 22 controls from Cristiano et al.; and 24 controls from Ulz et al.). Machine learning models were trained separately for each of the 3 control sets; the mean results over the 3*100 bootstrap iterations are shown. b ROC curves show the performance of each feature set for distinguishing between cfDNA samples from patients with EwS (nsamples = 98) and from patients with other pediatric sarcomas (nsamples = 18). For both sets of samples, we ensured the presence of tumor-derived cfDNA in the blood based on genetic evidence.