A Methodologically Updated De-Novo Extraction of Copy Number Signatures in Multiple Myeloma: Clinical Significance and Putative Aetiologies

BACKGROUND: The tumor genomes of most Multiple Myeloma (MM) patients are heavily burdened with highly heterogenous copy number (CN) alterations, as detected by multiple molecular methods including whole genome sequencing (WGS), and have been shown to have a strong prognostic and predictive significance for patients' survival. Cutting-edge computational developments have made it possible to analyze complex genomic CN changes by identifying CN alterations' recurrent patterns among large cohorts of patients, named CN signatures (CNS), that are the result of cumulative chromosomic instability (CIN) processes throughout the course of cancer cells evolution. In fact, a robust methodological framework for CNS computation, along with a large compendium of CNS aetiologies have been recently published for many cancer types, notably not including MM (Drews 2022). To our knowledge, the only study that analyzes CNS in MM focused on the role of CNS in predicting chromothripsis events, but remarkably it did not include neither CNS aetiologies nor complete CNS characterization for all the discovered CNS (Maclachlan 2021). AIM: To define a novel methodological framework, based on previous studies, aimed at the identification of CNS in MM, in order to assess both the aetiologies and the biological significance of MM CNS and to evaluate their prognostic impact on MM clinical outcome. The newly extracted CNS defined at diagnosis will be also used as novel disease biomarkers, to develop an improved, aetiology-based MM patients' stratification in different molecular subtypes. METHODS: We calculated the MM genomic distributions of the six essential CN features (segment length, breakpoint per 10 Mb, breakpoint per chromosome arm, segment CN change, CN value, length of oscillating CN states) that have been previously shown to encode patterns of CN alterations, underlying the observed tumor CIN. Starting from 886 WGS generated CN profiles, included in the CoMMpass study, the above mentioned features were computed. Features were then categorized into components, by using mixture models' decomposition and CNS were finally extracted from the components, by using both a Hierarchical Dirichlet Process (HDP) and a Non-Negative Matrix Factorization (NNMF) approach. RESULTS: The main novel characteristics of the developed methodological framework aimed at CNS assessment were 1) the use of a "continuous CN value" feature, which enabled the evaluation of sub-clonal events and 2) the use of a logarithmic scale in "segment length" feature, which favored a higher resolution for categorizing focal and/or gene level CN events, that are very common in MM. Thanks to these implementations, 33 Gaussian mixture components were identified (as compared to 28 detected in Maclachlan 2021). After deriving a Sample x Component - Sum of Posteriors Matrix, the signatures were extracted by applying two parallel state-of-art approaches, namely HDP and NNMF. This allowed the extraction of 9 CNS that were characterized by their component's composition. The signature's exposure levels were correlated to well-known MM biomarkers (e.g. TP53 mut and/or del, 1q CN gain, 13q CN loss, t-IgH, hyperdiploidy), showing that all signatures correlated to at least one of the well known MM biomarkers; in particular, CN.SIG5 exposures was found to correlate to high-risk MM biomarkers (TP53 p<0.001, 1q CN gain p<0.001, t(4;14) p<0.001), thus suggesting its possible involvement in the aetiology of this peculiar genomic configuration. Finally, a survival analysis was performed in patients characterized by high exposure (4th quartile) to the CN.SIG5 (75 patients), as compared to the others (811 patients), showing a significant negative impact of this CNS on both overall (OS p<0.001)) and progression free survivals (PFS p<0.001). Cox-analysis revealed an OS HR= 1.37 p<0.001, PFS HR= 1.16, p<0.001, per 5% increase in exposure. CONCLUSION: By employing a novel bio-informatic approach, based on the use of continuous CN data for CNS extraction, 33 feature's components were identified. We observed that CN.SIG5 significantly affected patients carrying well-known high-risk genomic features, and patients highly exposed to this CNS had decreased PFS and OS. Additional characterizations are needed to unveil the biological meaning of CNS exposure; however, MM CNS, while informing on disease outcome, might be considered as new comprehensive biomarkers in this disease.