A High-Risk Multiple Myeloma Group Identified By Integrative Multi-Omics Segmentation of Newly Diagnosed Patients

There is currently no standard approach to classify high-risk newly diagnosed Multiple Myeloma (ndMM). Various efforts have yielded approaches based on single molecular data types, including gene expression (GE), mutation (SNV), copy number alteration (CNA), and structural variation (SV) profiling. A comprehensive classification integrating heterogeneous molecular information may improve prognosis and treatment of high-risk ndMM patients.

Clinical data (ISS, Age, Sex), genomics datasets, and clinical outcomes were assembled. 18 common structural variants were derived from whole genome and whole exome data. CNAs were aggregated into cytobands and SNVs into pathways.

Two multi-omics integrative clustering methods (Cluster of Clusters and iClusterPlus) were applied to the integrated dataset to identify patient subgroups. Each algorithm was run 1000 times, using patient and feature resampling and optimized over a range of pre-defined clusters (K=[2,20]). Selection of clusters was based on Bayesian Information Criterion and Normalized Mutual Information. Patient subgroups defined using coherence across the outputs.

Biological interpretation of the high-risk group was performed using differential expression analysis (voom-limma), gene set enrichment of canonical pathways, master regulator (MR) inference to identify putative drivers, protein-protein interaction network (PPIn) analysis for identification of downstream effects, and Fisher's test for SNV, SV and CNA association. A relevance network topology-based method (DART) was used to compare results against previous high-risk GE signatures. Hypergeometric tests were applied to identify DNA signature enrichment across patient subgroups.

Multi-dimensional unsupervised analysis identified 12 stable patient subgroups (≥~5% of samples) including patient tumors enriched in t(4;14) and t(11;14), various patient subgroups stratified by CNAs including Del1p, Del13q, Del14q, Amp1q and amplifications of odd chromosomes, and distinct GE patterns.

Association of patient subgroups with PFS identified one high-risk group (median PFS < 505 days, referred to here as C8) which contains 11% of samples and is defined by a combination of genomic features including Del1p/13q/14q/17p and Amp1q, combination of t(4;14) and t(11;14) translocations, down-regulation of related gene expression profiles, and up-regulation of cell cycle related genes. An MR analysis of C8 tumors versus all others identified 10 potential driver genes including E2F2 and CKS1B.

C8 displays significant enrichment in GE signatures including EMC92, UAMS70, UAMS PR (proliferative) and M9. Published high-risk GE signatures overlap substantially with genes downstream of the MRs identified as driving C8, suggesting regulatory association with previously defined signatures (based on supervised analysis). C8 also showed a significant enrichment on t(4;14) with low FGFR3 expression versus the rest of t(4;14) patients, suggesting that not all t(4;14) patients are high-risk. Finally, the amp1q+ISS3 group and one third of bi-TP53 is significantly associated to C8, indicating that the bi-TP53 factor is a high-risk marker but not a driver of a patient sub-group.

C8 showed significant association to previously described high-risk signatures; importantly however, not all patients with high-risk markers (eg, t(4;14) and Double Hit patients) were present in this subgroup. The separation of t(4;14) and Double Hit patients into multiple subgroups with differing clinical outcome suggests a separation between high-risk biology and key poor prognosis markers. While these high-risk prognostic biomarkers may impact clinical outcome, our unsupervised analysis suggests that there is an interplay between these biomarkers and biological disease drivers that determine the poor prognosis of these patients.

Analysis of SVs, CNAs and GE differences identified 12 patient subgroups driven by SVs, CNAs, and GE profiles. In contrast, SNVs contributed less to the variance observed across and between MM subgroups. These analyses revealed potential biological drivers underlying a high-risk patient sub-group. The novel molecularly-defined subgroups identified will be further validated. This analysis provides an opportunity to identify biological drivers within molecularly-defined patient subgroups that can lead to development of novel therapies in MM.