Introduction: Multiple myeloma (MM) is an incurable plasma cell malignancy with a poor survival rate that is treated with combinations of immunomodulatory drugs (IMiDs), proteosome inhibitors (PIs), monoclonal antibodies (mAb), and increasingly antigen-based therapies like bispecific antibodies (BiTEs) and chimeric antigen receptor T-cell (CAR-T) therapies. Myeloma is hallmarked by high genetic diversity, which causes resistant subclones to proliferate after acquiring resistance to therapy. This pattern is not limited to IMiD, PI, and mAB treatment, and increasingly there is evidence of antigen escape in patients treated with anti-BCMA BiTE and CAR-T cell therapies. For these reasons, it is imperative to understand resistant subclones and subclones that are the most likely to acquire resistance in a patient sample regardless of the relative proportion of these subclones.

Methods: Using our single cell multiomic dataset of 325,025 plasma cells across 10 smoldering MM (SMM), 22 newly diagnosed MM (NDMM), and 17 relapsed refractory MM (RRMM) patients, we studied the subclonal heterogeneity to better understand clonal dynamics in response to treatment. We created a graph-based artificial intelligence approach to identify MM subclones using scRNA-seq, scATAC-seq, and inferred copy number alterations. This approach used the multiomic low dimensional representations (scRNA + scATAC) and the inferred copy number alterations (scRNA) to create two Euclidean similarity matrices. These matrices were used to merge high-resolution cell clusters into subclones based on Infomap network community detection. We trained a LASSO-Cox proportional hazard model on the Multiple Myeloma Research Foundation (MMRF) CoMMpass dataset using the gene sets from pathways related to treatment resistance and proliferation with progression free survival (PFS). We evaluated the risk of the subclones identified as it relates to our own patient PFS, overall survival (OS), disease progression, and treatment response.

Results: We identified 155 subclones from 49 patients with a mean of 2.30 in SSM, 3.59 in NDMM, and 3.12 in RRMM. There were significantly more subclones in NDMM patients than SMM patients (P = 0.020). When clustered, subclones from the same patient did not universally cluster together representing heterogeneity in the cells.After the LASSO-Cox model of PFS was trained and tested on the CoMMpass cohort (P < 0.001), the model was applied to the subclones from our study. We assigned an overall risk to each patient from either the lowest risk subclone or the highest risk subclone in that patient. The highest risk subclone in each patient was effective at identifying the lowest risk patients (25%) across all stages of myeloma (PFS P = 0.002 and OS P = 0.001), in NDMM (PFS P = 0.042), and in SMM (PFS P = 0.025). One NDMM patient in the high-risk strata later relapsed from PI and was refractory to BCMA BiTE. Upon further inspection, the highest risk subclone used to make these predictions had gain of 1q, deletion of 13, deletion of 14, and lower expression of BCMA (P < 0.001) compared to the other myeloma clones. When we used the lowest risk subclone from each patient as their overall risk, we could effectively identify the highest-risk patients (10-20%) across all stages of myeloma (OS P = 0.019), in NDMM (OS P = 0.040), and in SMM (PFS P = 0.005). In the SMM-specific analysis, we identified both SMM patients that progressed to MM and relapsed from therapy in the high-risk strata of patients. Both patients had high risk cytogenetic events like 1q across all subclones. This represents why the lowest risk subclone may be a good indicator of a high-risk patient, i.e., resistant and/or proliferative subclones could make up the entire myeloma compartment.

Conclusions: From our subclone atlas, we found that if a patient has even a single high risk subclone they are not low risk and that if all of the subclones in a patient are high risk that patient is in an especially high-risk category. We developed novel approaches to call myeloma subclones in an unbiased manner and used them to carefully annotate 155 subclones across 49 patients. We identified SMM patients who later progressed and NDMM patients who were later refractory to therapy based on hidden subclones. Building subclone atlases should become the norm as we attempt to address the complexities of our patients and personalize treatments that account for multiple subclones with different resistances.

Disclosures

Chopra:Genentech, Inc.: Current Employment; F. Hoffmann-La Roche Ltd.: Current equity holder in publicly-traded company. Zafar:Genentech: Current Employment.

This content is only available as a PDF.
Sign in via your Institution