Multiple myeloma (MM) is a heterogeneous disease, and cytogenetic abnormalities such as t(4;14) and del17p are well-established independent high-risk features in newly diagnosed patients. In two clinical trials (N=1422), the IFM2009 and Determination, patients with del17p or t(4;14) had a hazard ratio of 1.9 (95% CI 1.5-2.4) and 1.6 (95% CI 1.2-2.1), respectively. However, the same study also showed that 25% of the remaining patients had a similar short PFS without any identified known high-risk features. To understand the genomic etiology of the high-risk in these patients, we generated WGS data from 640 newly diagnosed patients (GAMER dataset) in 3 categories (low risk [PFS > 36 months] without known risk features (LR), high-risk [PFS < 18 months] without known risk features (unknownHR, uHR) and high-risk with known risk features) (HR). Our primary focus was to find factors that contribute to high-risk without known risk features. As a validation set, we used WGS samples from newly diagnosed patients who enrolled in the DFCI 2009 study. Overall, we observed that t(14;16) was the only primary translocation that differs between uHR (7%) and the LR group (1%) (p value < 0.01). We found more frequent driver mutations in TP53(7%), ATM(6%), and EGR1(4%), in the uHR group (FDR < 0.05). In uHR group, we found that the cancer cell fraction carrying these mutations was significantly lower compared to double hits events observed in HR (p value = 6.4e-05), suggesting their presence in high-risk subclones. Chromosome 1q gain vs amplification were associated with similar outcomes in these patients (p value = 0.81), but both significantly differed from chr1q wild type (p value = 1.8e-07). High-risk patients with and without known risk features had a significant enrichment of mutations in the Genome Integrity pathways (24%) compared to the LR group. However, the enrichment in MAPK signaling pathway mutations (60%) was significantly different between the two high-risk groups. Features that contribute to genome integrity, such as structural variants (insertion, deletion, translocation and inversion, p value<0.001), mutational load (p=0.00025), and genomic scar scores (p=0.0007), were all significantly lower in the low-risk group compared to the uHR. The number of copy number alterations that have not been described before, such as del16, del8p and gain 9 and 19, were also significantly different between low-risk and high-risk patients without known risk features. We further searched for genomic factors that contribute to the classification of uHR using multiple statistical modeling and machine learning tools. All models were run using a 10-fold cross-validation for choosing variables to put into the final model. Using all variables in the final model, we created a classification model using the variables' significance and the model's accuracy. Out of 27 variables, we found that 7 (mutational load, structural variants, del8p, t(14;16), Genome Integrity Pathway, MAPK pathway and ISS > 2) significantly contributed to the classification model. On training and validation data, this model successfully detected the risk group (p value = 3.5e-09 and 0.0027, respectively). Our dataset gives the ability to identify genomic risk markers that are not currently considered in newly diagnosed MM and allows us to extend our understanding beyond traditional risk. The absence of unique and mutually exclusive markers between uHR and LR suggests the additive risk model in MM patients without known high risk markers. Additional risk features might be hidden in the transcriptome and regulation of the transcriptome. We are currently building an integrative model from the same patients with WGS, RNA-seq and methylation data. This will allow us to fine-tune our risk model and help us develop simplified and complex models on MM risk beyond known cytogenetic risk features.
Read full abstract