Towards a Network-Based Molecular Taxonomy of Newly Diagnosed Multiple Myeloma

Recent advances in computational biology have led to the development of novel and sophisticated methods to model large datasets measured from complex organisms based on integrative network biology. Networks can provide valuable insight into key biological processes and allow for a deeper understanding of the complexity of cellular systems and disease mechanisms.

We developed and applied a network biology approach to infer an improved molecular model and understanding of newly diagnosed multiple myeloma (MM). We constructed the first co-expression network of MM based on RNA-seq data from the current release (IA4) of the Multiple Myeloma Research Foundation (MMRF) CoMMpass Study dataset. The data set consists of 92 samples from newly diagnosed MM patients. Whole Exome Sequencing (WES) data available for 77 out of 92 samples allowed the integration of somatic mutations into the network. Our analysis organized 23,033 genes into 50 co-expression modules. We then evaluated the molecular activity of co-expression modules for concordance with molecular traits. We performed module enrichment analysis against Gene Ontology terms, pathways, chromosome locations, protein-protein interaction networks, MM-associated gene sets and drug-target databases.

Analysis of the newly diagnosed multiple myeloma network model (MMNet) revealed known and novel molecular features of multiple myeloma. The integration of MMNet with somatic mutations data unveiled a significant association between mutation burden and the activation of several modules. Fundamental biological processes such as DNA repair, cell cycle, signal transduction, NK-kappaB cascade and MAPK signaling characterized such modules. Interestingly, a number of mutated genes demonstrated pluripotent associations with co-expression module activity. For example, FGFR3 was correlated with expression of several modules, including one enriched for RNA processing and translation-related processes and included the known MM-associated genes FRZB and CCND3. Similarly, the frequently mutated gene DIS3 was significantly associated to five different modules, including the translation-related module and a module enriched for the 1q locus. Our results have identified novel key driver genes that may inform therapy prioritization.

The MMNet topology revealed a far greater molecular heterogeneity in primary MM underscoring opportunities to improve the molecular taxonomy of this disease. We identified several modules associating with previously described MM classes, including a module enriched for genes up regulated in the UAMS MS class characterized by spiked expression of WHSC1 and FGFR3. Module connectivity confirmed the central role of both genes, WHSC1 being the top hub gene, i.e. the most connected gene in the module, and FGFR3 being among the top 10 hubs. Consistent with previous findings, this module was characterized by negative correlation with aneuploidy. We found other modules enriched for genes dysregulated in other UAMS classes, such as MF, CD1 and CD2. We also identified several modules associating with relevant biological processes such as apoptosis, cell communication, Wnt and Toll-like receptor signaling. Correlation of modules expression with clinical traits identified insights into genetic subgroups of MM that are not previously described. For examples, we found a module positively correlated to the African American ethnicity. This module was also characterized by enrichment for genes in the fragile regions 5q31 and 6q21. These findings may provide important and exciting insights into the biology of MM among African Americans as they are at increased risk for MM.

Our integrative network analysis of the CoMMpass dataset uncovers novel and complex patterns of genomic perturbation, key drivers and associations between clinical traits and genetic markers in newly diagnosed MM patients.