Abstract: Multiple myeloma (MM) is a malignancy of post-germinal centre B-cells whose pathogenesis is only partially understood. Chromosomal hyperdiploidy and recurrent immunoglobulin gene locus rearrangements are frequent, but are insufficient for malignant transformation, which is associated with additional events such as somatic mutations, epigenomic aberrations, and chromosomal copy-number changes. To investigate genomic event underlying MM pathogenesis and evolution, we used whole exome sequencing, copy number profiling and cytogenetics in 67 patients and 84 samples. For 15 patients, 2 or 3 serial samples (median 299 days apart) were available. Exome reads were used to call substitutions and indels. We used the Genome-Wide SNP Array 6.0 or exome reads to estimate the allele-specific copy number of the tumor. To cluster variants and estimate the clonal architecture of each sample and its evolution over time, we used the mutation burden, corrected for copy number and normal cell contamination. Analysis of the clonal structure of the tumors showed at least one subclone in 94% of patients at diagnosis, suggesting that myeloma is a heterogeneous disease at presentation. Interestingly, many mutations of known MM driver genes (KRAS, NRAS, BRAF, TP53, FAM46C) were subclonal at diagnosis. In 5/67 patients, BRAF and KRAS/NRAS mutations co-existed in the same sample, raising therapeutic implications given the paradoxical ERK-activating effect of BRAF inhibitors in RAS-mutated cells. Furthermore, only 3/10 BRAF variants were V600E, the current target of most inhibitors. Altogether, only the 5 previously known genes were significantly enriched in our cohort, highlighting marked heterogeneity of the spectrum of candidate driver gene mutations across MM patients. Nevertheless, we identified several new recurrent gene lesions: inactivating mutations of SP140 (7.5%), a gene previously linked to germline susceptibility to CLL, and in ROBO1 (7.5%), a gene recently implicated in pancreatic cancer; clustered missense substitutions in EGR1 (6%), a gene previously implicated in plasma cell apoptosis; clustered truncating mutations in LTB (4.5%), a TNF-family protein implicated in lymphoid development. The subclonal structure of the sample changed over time in 72% paired samples, highlighting genomic evolution at relapse. We described 4 different scenarios with striking concordance between mutations and chromosomal copy number changes: no change, linear evolution (a new clone appears in the later sample), differential clonal response (the relative proportions of the subclones change over time), and branching evolution (new clones emerge, while others decline in frequency or disappear). All subclonal variants in known driver myeloma genes increased their clonal fraction at the later time-point, consistent with the expected positive selection for the subclones harboring them. To investigate mutational processes responsible for the generation of the mutational repertoire in MM, we extracted the variant context and analyzed the mutational signatures. We found two signatures in our samples. The most represented one is enriched for spontaneous deamination of methylated cytosines, a common process in cancer and aged cells. The second signature was more represented in samples showing extremely high numbers of variants, sometimes clustered in small regions of ∼200 bp (kataegis). We hypothesize that it results from aberrant activity of the APOBEC family of cytosine deaminases, recently described in breast cancer. Interestingly, cases of extramedulary relapse were always associated with branching evolution and showed increased contribution from this APOBEC signature. In conclusion, in our cohort of MM samples we show: 1) evidence of tumor heterogeneity at the time of diagnosis; 2) discernable genetic changes and shifts in the clonal structure of disease at the time of progression; 3) different mutational processes responsible for an heterogeneous mutational repertoire across patients, and over time in the same patient; 4) a comprehensive list of recurrent variants, many of which are previously unreported. Our study provides new insights into the genomic architecture of MM, and will help identify molecular alterations associated with progression of disease and development of drug resistance.