Gene Expression–Based prediction of 1q, 1p, 13q, and 17p chromosome ploidy in newly diagnosed and paired relapsed myeloma and across the plasma cell dyscrasia spectrum

Abstract

Abstract Gene Expression–Based Prediction of 1q, 1p, 13q, and 17p Chromosome Ploidy in Serial Newly Diagnosed and Relapsed Myeloma and Across the Plasma Cell Dyscrasia Spectrum Background: Recurrent chromosomal abnormalities such as 1q gain and deletions of 1p, 13q, and 17p are biologically and prognostically significant in multiple myeloma (MM) yet are not fully integrated into clinical risk models like mSMART or R2-ISS. FISH, while standard for detection, has limitations in resolution and tissue availability. We developed gene expression–based models to infer ploidy states across 16,145 plasma cell dyscrasia (PCD) cases, including 1,371 newly diagnosed (NDMM) and 382 paired relapsed (RRMM) samples from a single-institution (UAMS) cohort with over 25 years of follow-up. Methods: Using Affymetrix gene expression data from CD138+ purified plasma cells (≥90% purity), we trained models to classify copy number states of 1q (2, 3, ≥4 copies; n = 1,581), and 1p, 13q, and 17p (1 vs 2 copies; n = 1,581, 689, and 1,246, respectively), based on matched FISH data. Models were evaluated on a 2:1 train/test split, achieving balanced accuracies of 0.918 (1q), 0.887 (1p), 0.872 (13q), and 0.818 (17p). These models were applied to the full PCD cohort to evaluate clinical and genomic correlations. Multivariate Cox models assessed associations with progression-free (PFS) and overall survival (OS), and combinations of chromosomal ploidy were analyzed to define high-risk subgroups. Results: Paired relapsed samples showed enrichment of high-risk lesions, including 1q amplification (≥4 copies: 14% NDMM vs. 21% RRMM), del(13q) (54% vs. 61%), and del(17p) (28% vs. 32%). In NDMM, 1q ploidy stratified outcomes: median OS was 13.3, 8.4, and 4.0 years for 2, 3, and ≥4 copies, respectively; PFS followed a similar trend (7.3, 4.6, and 2.4 years). Poor post-relapse survival was observed among those acquiring 1q amplification. Combinatorial modeling of inferred lesions revealed additive and synergistic effects, particularly among combinations of 1q amplification with del(1p), del(13q), or del(17p), which conferred inferior outcomes beyond existing risk classifications. The gene expression–based prediction framework was applied to 16,145 plasma cell dyscrasia samples—including MGUS, SMM, NDMM, RRMM, PCL, and EMD—and accurately inferred chromosomal ploidy in cases lacking FISH, enabling large-scale modeling of high-risk lesion incidence and co-occurrence. Conclusions: Gene expression–based prediction of chromosomal ploidy enables scalable, FISH-independent identification of subclonal and dynamic high-risk features across plasma cell dyscrasias. We highlight the incidence of 1q, 1p, 13q, and 17p abnormalities across the disease spectrum, with emphasis on paired serial samples from MGUS, SMM, NDMM, and RRMM, including bone marrow and extramedullary relapses. The poor survival associated with 1q amplification—especially in the context of concurrent 1p, 13q, or 17p deletions—underscores the need to integrate these patterns into next-generation risk stratification systems. This transcriptomic approach enhances prognostication and supports efforts to capture disease evolution more comprehensively than traditional models like mSMART and R2-ISS.