Fig 1.
Overview of experimental methods.
Bulk T-cell receptor β sequencing of DNA extracted from peripheral blood mononuclear cells (PBMCs) was analyzed from 447 samples, including 80 individuals with monoclonal gammopathy of undetermined significance (MGUS), 55 with smoldering multiple myeloma (SMM), and 31 with newly diagnosed multiple myeloma (NDMM) and compared to data from 166 healthy donors obtained from the Adaptive ImmuneAccess database. In total, 612 samples were analyzed. This included follow-up samples from 72 MGUS patients (median follow-up 677 days) and longitudinal samples from high-risk SMM patients at the end of cycle 1 (n = 54), cycle 4 (n = 50), cycle 8 (n = 46), cycle 20 (n = 36), and cycle 32 (n = 22) of carfilzomib, lenalidomide, and dexamethasone (KRd).
Fig 2.
T-cell receptor β chain (TCRB) clonality is not significantly associated with disease burden or depth of treatment response. A) Comparison of TCRB clonality (Gini coefficient) between 44 healthy individuals (age > 40 years) and patients with monoclonal gammopathy of undetermined significance (MGUS), smoldering multiple myeloma (SMM), and multiple myeloma (MM). P-values of significant differences in the univariate analysis (Wilcoxon test) are shown. No significant differences were observed in the multivariate linear model that corrected for age. B) TCRB clonality in MGUS patients at baseline and follow-up (median follow-up 677 days). C) TCRB clonality in MM patients before treatment, stratified by post-treatment depth of response. D) TCRB clonality in SMM patients undergoing carfilzomib, lenalidomide, and dexamethasone (KRd) shown for the entire cohort, and E) stratified by depth of response. F) Line plot comparing the change in individual TCRB clonality among SMM patients undergoing KRd treatment. MRD, minimal residual disease; CR, complete response; VGPR, very good partial response; PR, partial response.
Fig 3.
TCRB clustering analysis differentiates plasma cell dyscrasias from healthy blood samples.
A) Venn diagram illustrating the overlap of all TCRB clusters between healthy, monoclonal gammopathy of undetermined significance (MGUS), smoldering multiple myeloma (SMM), and multiple myeloma (MM) baseline samples. B) Principal Component Analysis (PCA) plot visualizing the variance in frequency of differentially abundant TCRB clusters across sample types. C) Performance metrics and 95% confidence intervals for each classifier across five repeated stratified 80/20 train/test splits within the 70% training dataset. For each machine learning model, performance was evaluated using five independent train/test iterations. D) ROC curve showing the discriminative performance of the final random forest classifier when applied to the independent 30% test set that was not used during model training or feature selection. The model achieved an AUROC (area under the receiver operating characteristic curve) of 0.84, demonstrating its ability to generalize to unseen samples.
Fig 4.
Physical properties of differentially abundant TCRB clusters vary across healthy and plasma cell dyscrasia (PCD) samples.
A) Heatmap illustrating the normalized mean physical properties (columns) of CDR3β amino acid sequences within each of the differentially abundant TCRB clusters (rows). Bar plots adjacent to the heatmap display the variable importance in the neural network model, the count of CDR3β amino acid sequences per cluster, and the number of patients per diagnostic group. B) Box plots comparing the distributions of physical properties between sample types, with Wilcoxon rank-sum test p-values indicated. C) Circle packing plot illustrating the hierarchical relationship of TCRB clusters containing one or more CDR3β amino acid sequences with known antigen or pathologic condition specificity previously reported in the literature. Outer grey circles represent TCRB clusters, while inner circles denote CDR3β amino acid sequences colored by antigen specificity. Colored circles not enclosed by grey circles signify the sole clonotype within that cluster with established antigen specificity.