Unsupervised Clustering of DNA Copy Number Profiles Identifies a High-Risk Subtype of Hyperdiploid Multiple Myeloma: An Mmrf Commpass Analysis

Multiple myeloma (MM) is a malignancy of the antibody producing plasma cell, which exhibits a high degree of genetic diversity between patients. As genetic analysis technologies have improved so has our understanding of the diverse genetic phenotypes underlying the disease. The MMRF CoMMpass study (NCT01454297) is using whole genome (WGS), exome (WES), and RNA (RNAseq) sequencing to provide a precise characterization of each patient before and after therapy. However, these advanced assays are not widely available to patients today limiting the utility of many observations to a small population of patients. To expand the utility of the data set to a broader patient population we focused on DNA copy number (CN) phenotypes that can be identified by the standard FISH assays widely used in the field.

To discover potential underlying phenotypes of myeloma beyond the known dichotomy of hyperdiploid (HRD) and non-hyperdiploid (NHRD) karyotypes, unsupervised consensus clustering was performed on 871 patients with CN profiles from WGS. Given the limited dynamic range of CN values, a Monte Carlo reference-based consensus clustering algorithm, M3C, was used to limit potential overfitting issues. Three independent replicates of this procedure identified an optimal solution of eight subtypes with no more than 6 patients having different class assignments between replicates.

The eight CN subtypes consisted of five HRD and three NHRD subtypes and were annotated based on common CN features. The HRD classic subtype had ubiquitous CN gains, trisomies, of classic HRD chromosomes, 3, 5, 7, 9, 11, 15, and 19. The remaining HRD subtypes were annotated based on deviations from the classic HRD phenotype. The HRD, ++15 subtype phenocopies classic HRD except tetrasomy, not trisomy, is observed on chr15. Two groups of HRD patients were identified lacking CN gains of chr7 which are split into two distinct subtypes: the HRD, diploid 7 subtype, which lacked gains of chr7; and the HRD diploid 3, 7 subtype lacking trisomies of both chr3 and chr7. This suggest some relationship between chromosomes 3 and 7 where trisomy 7 is not tolerated in the absence of trisomy 3. Finally, the HRD, +1q, diploid 11, -13 subtype had gains of the classic HRD chromosomes except chr11 with gains of chr1q and loss of chr13. This subtype suggests trisomy 11 is essential for an HRD phenotype but it can be phenocopied by the combination of 1q gains and 13 loss. Within the NHRD subtypes, the diploid subtype is almost devoid of CN abnormalities less a common gain of 11q initiating at the breakpoint the t(11;14) event, which is almost universally observed in this subtype. Unlike the diploid subtype, the remaining NHRD subtypes have more complex CN profiles with the -13 subtype defined by monosomy 13, and the +1q/-13 subtype defined by gains of 1q and monosomy 13.

Outcome analyses of the CN subtypes identified in CoMMpass revealed that both HRD and NHRD patients with gains of chr1q and loss of chr13 exhibited poor PFS and OS outcomes as compared to patients in other CN subtypes. Interestingly, the PFS curves split into three groups with a good risk group defined by the HRD classic and HRD ++15 subtypes. a high-risk group defined by 1q gain and monosomy 13 regardless of ploidy phenotype, and an intermediate group with all other subtypes. The distribution of HRD patients into these three outcome groups highlights the danger of assuming all HRD myeloma patients will have similar outcomes. Patients in the HRD, +1q, diploid 11, -13 subtype exhibited poor OS outcomes (median = 56 months) as compared to patients in the HRD, ++15 (p<0.01), HRD, classic (median = 65 months, p<0.05), diploid (p<0.01), and -13 (p<0.05) subtypes. Patients in the +1q, -13 subtype also exhibited poor OS outcomes (median = 57 months) as compared to patients in the diploid (p<0.01), -13 (p<0.05), HRD classic (p<0.05), and HRD, ++15 (p<0.01) subtypes. Overall, both HRD and NHRD patients with gain of 1q and loss of chr13 exhibit poor outcome as compared to patients with other genetic backgrounds (HR = 1.928, 95% CI = 1.435 - 2.59, p<0.001). Further, the observation that NHRD patients in the +1q, -13 subtype exhibit poor OS outcomes as compared to NHRD patients in the -13 subtype highlights the importance of 1q gains in determining patient prognosis. These results can easily be translated into clinics around the world by matching existing FISH data to each of these groups until more advanced testing is common practice.