Genomic Classification and Individualized Prognosis in Multiple Myeloma

Abstract

Purpose: Outcomes for patients with newly diagnosed multiple myeloma (NDMM) are heterogenous, with overall survival (OS) ranging from months to over 10 years.

Methods: To decipher and predict the molecular and clinical heterogeneity of NDMM, we assembled a series of 1,933 patients with available clinical, genomic, and therapeutic data.

Results: Leveraging a comprehensive catalog of genomic drivers, we identified 12 groups, expanding on previous gene expression-based molecular classifications. To build a model predicting individualized risk in NDMM (IRMMa), we integrated clinical, genomic, and treatment variables. To correct for time-dependent variables, including high-dose melphalan followed by autologous stem-cell transplantation (HDM-ASCT), and maintenance therapy, a multi-state model was designed. The IRMMa model accuracy was significantly higher than all comparator prognostic models, with a c-index for OS of 0.726, compared with International Staging System (ISS; 0.61), revised-ISS (0.572), and R2-ISS (0.625). Integral to model accuracy was 20 genomic features, including 1q21 gain/amp, del 1p, TP53 loss, NSD2 translocations, APOBEC mutational signatures, and copy-number signatures (reflecting the complex structural variant chromothripsis). IRMMa accuracy and superiority compared with other prognostic models were validated on 256 patients enrolled in the GMMG-HD6 (ClinicalTrials.gov identifier: NCT02495922) clinical trial. Individualized patient risks were significantly affected across the 12 genomic groups by different treatment strategies (ie, treatment variance), which was used to identify patients for whom HDM-ASCT is particularly effective versus patients for whom the impact is limited.

Conclusion: Integrating clinical, demographic, genomic, and therapeutic data, to our knowledge, we have developed the first individualized risk-prediction model enabling personally tailored therapeutic decisions for patients with NDMM.

FIG 1.

Genomic driver landscape in newly diagnosed multiple myeloma. (A) Driver genes significantly involved by single-nucleotide variants and indels using four different driver discovery tools (Fishhook, Oncodriver, MutSigCV, and dNdScv). X-axis label colors represent the COSMIC census annotation for each driver gene: red = oncogenes; blue = TSG; black = unknown. (B) Significant broad and focal copy-number changes detected by GISTIC: red = gain, blue = loss. (C-E) Kaplan-Meier curves for OS according to (C) RB1 allelic status, (D) presence of chromothripsis-CNV.Sig, and (E) APOBEC activity. CNV.Sig, CNV signature; GISTIC, Genomic Identification of Significant Targets in Cancer; OS, overall survival; TSG, tumor-suppressor genes; WT, wild type.

FIG 2.

Newly diagnosed multiple myeloma genomic classification. The features defining 12 genomic clusters are defined, including GEP70 status, FISH-TC6, UAMS gene expression groups, IGH translocations, and all key mutational, copy number, and structural variant features reported in Figure 1. Gray: wild-type; red: single allele event or APOBEC; brown: biallelic event; hyper-APOBEC; 1q amplification, and IGH canonical translocations.

FIG 3.

IRMMa. (A) Multistate model of a patient with NDMM. The six colored shapes correspond to different stages across the two phases (phase I: induction and phase II: postinduction), with different possible transitions (arrows). The number and percentage in each shape corresponds to the total number of patients who entered during the study follow-up. Among lost to follow-up in phase I group, 66/137 (48%) died in phase I for other causes. IRMMa was developed using NCNPH. (B) Sediment plot showing the risk over time for a single patient with NDMM from diagnosis to 5 years. The patient's clinical and genomic profile was predicted as high risk with a predicted probability be alive and in remission of 17.7% in line with the clinical and genomic high-risk presentation. (C) Boxplots comparing c-index for ISS, R-ISS, R2-ISS, and IRMMa for OS and EFS. Boxplots are generated using stratified cross-validation (5-fold × 10-random-repeats = 50 splits). As expected, R-ISS showed a lower accuracy than ISS and R2-ISS. This is because R-ISS was developed with a focus on high sensitivity, which means it is good at identifying patients with high-risk NDMM. However, this focus on sensitivity comes at the cost of specificity, which means that it is more likely to incorrectly classify high- and intermediate-risk patients as intermediate- and low-risk, respectively. (D) Model performance comparing predicted risk of progression (EFS) or death (OS) and observed (ie, bar colors). EFS, event-free survival; IRMMa, individualized prediction model for newly diagnosed multiple myeloma; ISS, International Staging System; NCNPH, neural Cox nonproportional hazards; NDMM, newly diagnosed multiple myeloma; OS, overall survival; PD, progression disease; R-ISS, revised international staging system.

FIG 4.

IRMMa anatomy. (A) The relative weight of age, ISS, genomics, treatment, other clinical (ie, ECOG, sex, race, and LDH level), HDM-ASCT, and maintenance/continuous treatment on our prediction model (IRMMa) for each of the multistate phase. (B and C) Predicted and observed outcome for the GMMG-HD6 cohort. The observed clinical outcome for patients included in the training treated with VRd + HDM-ASCT + MCT was also included. ECOG, Eastern Cooperative Oncology Group; HDM-ASCT, high-dose melphalan followed by autologous stem-cell transplantation; IRMMa, individualized prediction model for newly diagnosed multiple myeloma; ISS, International Staging System; LDH, lactate dehydrogenase; MCT, maintenance/continuous treatment; VRd, bortezomib, lenalidomide, and dexamethasone.

FIG 5.

Predicted treatment variance in patients with newly diagnosed multiple myeloma treated with VRd. (A) Heatmap showing the predicted treatment variance across 1,933 patients in case of treatment with VRd ± HDM-ASCT ± MCT. (B-D) Observed probability to be alive and in remission (PFS) across the treatment variance groups defined in (A). To correct for HDM-ASCT and maintenance/continuous treatment, time was calculated from the end of induction to the last follow up (ie, phase II). Statistical differences between all different treatment groups within each cluster was estimated by log-rank test and reported in the Data Supplement (Table S10). HDM-ASCT, high-dose melphalan followed by autologous stem-cell transplantation; IRMMa, individualized prediction model for newly diagnosed multiple myeloma; ISS, International Staging System; LDH, lactate dehydrogenase; MCT, maintenance/continuous treatment; PFS, progression-free survival; SNV, single nucleotide variant; VRd, bortezomib, lenalidomide, and dexamethasone; WT, wild type.

FIG 6.

Figure summarizing IRMMa potential integration into clinical applications. The figure was generated using BioRender. IRMMa, individualized prediction model for newly diagnosed multiple myeloma.