Copy number signatures predict chromothripsis and clinical outcomes in newly diagnosed multiple myeloma

Abstract

Chromothripsis is detectable in 20-30% of newly diagnosed multiple myeloma (NDMM) patients and is emerging as a new independent adverse prognostic factor. In this study we interrogate 752 NDMM patients using whole genome sequencing (WGS) to investigate the relationship of copy number (CN) signatures to chromothripsis and show they are highly associated. CN signatures are highly predictive of the presence of chromothripsis (AUC = 0.90) and can be used identify its adverse prognostic impact. The ability of CN signatures to predict the presence of chromothripsis is confirmed in a validation series of WGS comprised of 235 hematological cancers (AUC = 0.97) and an independent series of 34 NDMM (AUC = 0.87). We show that CN signatures can also be derived from whole exome data (WES) and using 677 cases from the same series of NDMM, we are able to predict both the presence of chromothripsis (AUC = 0.82) and its adverse prognostic impact. CN signatures constitute a flexible tool to identify the presence of chromothripsis and is applicable to WES and WGS data.

Fig. 1. A schema demonstrating the definition of copy-number (CN) features from multiple myeloma whole-genome sequencing data.

a Input genome-wide copy number gain (red) and loss (blue) data from 752 newly diagnosed multiple myeloma whole genomes. b Measure copy number as classified by 6 key features. c Define the optimum number of categories for each copy number feature by a mixed-effects model (mclust). d Tally the number of CN variation for each of 28 CN categories to produce a matrix of key CN features. This comprises the input matrix for the hierarchical Dirichlet process (hdp) for de novo extraction of CN signatures.

Fig. 2. De novo extraction from whole-genome sequencing data produces 5 copy-number (CN) signatures in 752 newly diagnosed multiple myeloma.

The 5 CN signatures extracted comprise varying contribution across the 28-CN-feature matrix. The 2 chromothripsis-associated signatures are CN-SIG4 and CN-SIG5. (CN-SIG: copy-number signature, n = 752 samples, data are presented as median values ± SD).

Fig. 3. Clinical data demonstrates the correlation of copy number (CN) signatures with high-risk multiple myeloma prognostic features and complex genomic change.

a A heatmap of MM mutational and structural features demonstrates that contribution from CN-SIG4 and CN-SIG5 cluster with features of high-risk MM. Presence of biallelic TP53 inactivation and chromosome 1q21 amplification (i.e., >3 copies) are annotated in dark red; presence of chromothripsis in purple; all the other genomic features are in bright red when present. b–g There is a significantly higher median contribution from CN-SIG4 and/or CN-SIG5 on the samples having translocations involving b MAF/MAFB (p = 0.0005), c increased APOBEC mutational activity (p = 9.9e⁻⁵), d biallelic TP53 inactivation (p = 1.3e⁻⁶), e gain/amplification of chromosome 1q21 (p = 1.3e⁻⁸), f chromoplexy (p = 1.7e⁻⁷) and g chromothripsis (p = 2.2e⁻¹⁶). Boxplots show median and interquartile range (IQR), with whiskers extending to 1.5 * IQR, n = 752, p-values indicate significance by a 2-sided Wilcoxon rank-sum test. (CN-SIG: copy number signature, neg: lacking the feature, pos: containing the feature, WT: wild type).

Fig. 4. De novo extraction from whole-genome sequencing data produces 10 structural variant (SV) signatures in 752 newly diagnosed multiple myeloma samples.

The 10 SV signatures extracted comprise varying contribution across the 32-SV-feature matrix. SV-SIG1, SV- SIG2, and SV-SIG3 contain clustered SV features and are associated with chromothripsis. The remainder of the signatures consist of non-clustered events. (SV-SIG: structural variant signature).

Fig. 5. Copy number (CN) signatures in newly diagnosed multiple myeloma are strongly predictive of chromothripsis.

Receiver operating curve (ROC) for the prediction of chromothripsis from CoMMpass whole-genome sequencing (WGS) data (n = 752) from a structural variant (SV) and CN signature analysis, b ShatterSeek features, c SV signatures alone and d CN signatures alone. e ROC for the prediction of chromothripsis from the validation set of other hematological cancers (n = 235). f ROC for the prediction of chromothripsis from the newly diagnosed multiple myeloma subset of the validation WGS (n = 34). Blue lines represent individual ROC (from 10-fold cross validation in (a–e) and 5-fold validation in (f), red lines represent the mean of individual ROC, AUC: mean area-under-the-curve.

Fig. 6. Copy number (CN) signatures in newly diagnosed multiple myeloma are independently predictive of clinical outcomes.

a Progression-free survival (PFS) probability in the CoMMpass dataset according to high (blue) or low (red) CN-prediction score for chromothripsis (CN_pred). b Overall survival (OS) probability in the CoMMpass dataset according to high (blue) or low (red) CN_pred. c Multivariate analysis of the effect of CN_pred on PFS after correction for International Staging Score (ISS), age, Eastern Cooperative Oncology Group (ECOG) score, and APOBEC mutational activity. d Multivariate analysis of the effect of CN_pred on OS after correction for the same factors. All p-values for Kaplan–Meier curves were generated according to a 2-sided log-rank test. Multivariate analysis was performed by the Cox proportional hazards model with p-values according to a 2-sided Wald test. Data are presented as median values ± 95% confidence interval.

Fig. 7. Extracted copy number (CN) feature profiles from whole-genome sequencing (WGS) and whole-exome sequencing (WES) are highly analogous.

a An example of chromothripsis from the CoMMpass dataset (MMRF_1646_1_BM; chr: chromosome). The horizontal black line indicates total copy number; the dashed orange line minor copy number. Vertical lines represent structural variant breakpoints for deletion (red), inversion (blue), tandem-duplication (green), and translocations (black), involving chromosomes 6 and 9. The extracted CN category profile from the same example patient (MMRF_1646_1_BM) from b WGS and c WES.

Fig. 8. Copy number (CN) signatures extracted from whole-exome sequencing (WES) in newly diagnosed multiple myeloma are highly predictive of clinical outcomes.

a Progression-free survival (PFS) probability in the CoMMpass dataset according to high (blue) or low (red) exome CN-prediction score (eCN_pred) for chromothripsis. b Overall survival (OS) probability in the CoMMpass dataset according to high (blue) or low (red) eCN_pred. c Multivariate analysis of the effect of eCN_pred on PFS after correction for International Staging Score (ISS), age and APOBEC mutational activity. d Multivariate analysis of the effect of eCN_pred on OS after correction for the same factors. All p-values for Kaplan–Meier curves were generated according to a 2-sided log-rank test. Multivariate analysis was performed by the Cox proportional hazards model with p-values according to a 2-sided Wald test. Data are presented as median values ± 95% confidence interval.