CyTOF profiling of bone marrow immune dynamics across myeloma stages

ABSTRACT

Multiple myeloma (MM) orchestrates a profound disruption of immune balance within the bone marrow (BM) microenvironment, driving disease progression and therapeutic resistance. To better understand these complex immune dynamics, we used high-dimensional mass cytometry (CyTOF) profiling to comprehensively characterize the immune landscape of the BM across different stages of myeloma progression, including MGUS (n = 16), smoldering MM (SMM; n = 25), and active MM, both newly diagnosed (n = 43) and relapsed/refractory (n = 104). Our analysis revealed substantial immune remodeling during disease progression, characterized by adaptive immune suppression and extensive infiltration of innate immune populations. Transformation from MGUS to SMM was marked by significant alterations in central and effector memory T helper cells, effector cytotoxic T cells, and an enrichment of monocytic and neutrophil subsets. Active MM stages were further distinguished by increased expansion of myeloid and monocytic lineages, alongside a pronounced reduction in progenitor and transitional B cells. Correspondence analysis demonstrated that specific immune profiles were significantly associated with clinical outcomes, including progression-free survival and overall survival. This study highlights the potential of CyTOF-based molecular profiling to unravel the intricate immune dynamics of the BM microenvironment across MM disease stages, enhancing our understanding of MM pathogenesis and providing a foundation for identifying prognostic biomarkers and tailoring precision immunotherapeutic strategies.

Introduction

Multiple myeloma (MM) is a hematologic malignancy characterized by the clonal expansion of plasma cells (PC) within the bone marrow (BM), resulting in profound disruption of normal hematopoiesis and a range of clinical manifestations, including anemia, bone lesions, hypercalcemia, and renal dysfunction.¹ The pathogenesis of MM involves a complex interplay between genetic alterations, such as chromosomal translocations, copy number abnormalities, and somatic mutations, and the immune microenvironment, which is reprogrammed to support tumor growth through immune suppression and evasion.² Premalignant stages of MM, including monoclonal gammopathy of undetermined significance (MGUS) and smoldering MM (SMM), are marked by the presence of monoclonal plasma cells and associated immune dysregulation, serving as critical windows to understand early disease progression.³ Despite novel therapeutic strategies that have transformed MM landscape,^4,5 MM remains an incurable disease for most patients, underscoring the need for continued research to develop more effective and durable treatment strategies.

The MM microenvironment is a dynamic, organized ecosystem of phenotypically, transcriptomically, and functionally diverse immune and stromal cells whose interactions with malignant plasma cells, shaped by genetic alterations, treatment strategies, disease stage, and aging, disrupt innate – adaptive immune crosstalk, drive an immunosuppressive niche, and render the bone marrow susceptible to clonal hematopoiesis.^6–8 During the premalignant stage of MGUS, T cells show stem-like traits and increased expression of markers like TCF1, CXCR4, NR4A2, and CD69, indicating bone marrow residency.⁹ The progression from MGUS and SMM to symptomatic MM involves immune changes, including increased Tregs and non-classical CD16+ monocytes that promote an immunosuppressive, tumor-supportive ecosystem.^10–12 As MM progresses, adaptive immune dysfunction is evident with increased Tregs and decreased memory T cells hindering anti-tumor responses, while B cells act as reservoirs for therapy-resistant tumor cells.^13–15 In the innate compartment, active MM entails increased natural killer (NK) and CD16⁺ cells, with concurrent loss of plasmacytoid dendritic cells (pDCs), immature neutrophils, and CD14⁺ monocytes; the defective antigen presentation by these cells, together with expanded M2 macrophages, impairs T-cell activation and promotes tumor growth, and, along with NK cell overactivation and exhaustion marked by TIM3 and TIGIT, reinforces immune evasion.^10,13,14,16 A deeper understanding of PC biology and their tumor immune microenvironment (TIME) in precursor conditions and MM, compared to age-matched healthy controls, could inform novel therapeutic strategies to enhance anti-MM immunity or disrupt tumor-promoting signals within the immunosuppressive BM niche.

In this study, we comprehensively examined the immune landscape of the BMmicroenvironment across various stages of myeloma progression, including MGUS, SMM, newly diagnosed MM (NDMM), and relapsed or relapsed/refractory MM (RRMM), and compared it to healthy donors (HD). By employing advanced high-dimensional mass cytometry (CyTOF) profiling techniques, we analyzed innate and adaptive immune subsets to uncover stage-specific alterations and immune remodeling patterns associated with disease progression. Our findings provide insights into the expansion of myeloid and monocytic lineages, suppression of T cell subsets, and differential expression of immune checkpoint molecules, correlating these changes with clinical outcomes. These results highlight the intricate interplay between immune dysfunction and myeloma pathogenesis and may, upon further validation, inform prognostic biomarker development or the design of immune-modulatory interventions in multiple myeloma.

Materials and methods

Patient enrollment

BM aspirates were obtained from 188 patients with MM, including those with premalignant/asymptomatic MGUS (n = 16) and SMM (n = 25), as well as active symptomatic stages NDMM (n = 43) and RRMM (n = 104), along with HD (n = 10). Clinical characteristics of MM patients are detailed in Supplementary Table 1. Comprehensive clinical and treatment history was collected at the time of BM sampling. Samples were acquired during routine diagnostic BM aspirations at the Dana-Farber Cancer Institute (Boston, USA) and the Biomedical Research Center (Bratislava, Slovakia). Control BM samples from age-matched healthy donors (n = 10) were obtained from AllCells under an identical collection protocol. Ethical approval was granted by the Biomedical Research Center’s local ethics committee (Protocol No. Myelom 001) and the Dana-Farber Cancer Institute (Protocol No. 10–106), adhering to the Declaration of Helsinki. All participants, including patients and healthy donors, provided written informed consent for the use of their samples in this study.

Conjugation strategy for antibodies

Antibodies, including clones, providers, and concentrations, are listed in Supplementary Table S2. Using the MaxPAR antibody conjugation kit (Fluidigm Sciences), antibodies (100 μg) were conjugated with mass isotopes following the manufacturer’s instructions. Briefly, lanthanide metal solutions (2.5 mM) were loaded onto the polymer for 40 min at 37°C. The antibody buffer was exchanged, and partial reduction was performed using 4 mM TCEP for 30 min at 37°C. After purification, antibodies were conjugated to lanthanide-loaded polymers for 1 hour, followed by recovery through four washing steps.

Assessment of metal-conjugated antibody performance

Metal-conjugated antibodies were digested in 2% HCl and then diluted in 2% HNO3. Metal content (≥100 metal atoms per antibody) was analyzed using a CyTOF 2 mass cytometer (Fluidigm) with blank and tuning solutions as normalization standards. Antibody concentration was measured at 280 nm using a NanoDrop, adjusted to 0.5 mg/mL in PBS antibody stabilization solution with 0.05% sodium azide (NaN₃), and stored at 4°C. Antibody efficacy was titrated and validated using positive controls, including human peripheral blood mononuclear cells or cell lines (see Supplementary Table S2).

Bone marrow sample processing

Fresh BM aspirates were collected in sodium heparinized tubes and fixed with proteomic stabilizer buffer (1.4 mL buffer per 1 mL BM) for 10 minutes at room temperature, then frozen at −80°C. Before analysis, samples were thawed in a 4°C water bath, and erythrocytes were lysed with hypotonic “1x thaw-lyse” buffer (4:1 buffer-to-sample ratio) for 10 minutes at room temperature (RT). Leukocytes were pelleted by centrifugation at 600 × g for 6 minutes, and lysis steps were repeated as needed. Cells were washed in cell staining medium (CSM; PBS with 0.5% BSA and 0.02% NaN3) and collected by centrifugation.

CyTOF workflow: sample preparation and acquisition

Cells were washed in CSM and blocked with 5 µL of Fc receptor blocking solution for 10 minutes at RT. A total of 5 × 10⁶ cells were then stained with specific antibody panels (B or TIME) using a 100 µL CSM-based cocktail for 30 minutes at RT. After surface staining, cells were washed and, for the TIME panel, subjected to intracellular staining using the FOXP3 Perm buffer. Intracellular antibody cocktails and/or 191/193 Iridium DNA intercalator were used for staining in 100 µL CSM for 1 hour at RT. After final washes, cells were collected by centrifugation. For CyTOF acquisition, cells were washed with PBS, then ddH₂O, and diluted to 0.5 × 10⁶ cells/mL in ddH₂O with 10% EQ Four Element Calibration Beads. Samples were acquired on a CyTOF 2 mass cytometer at a consistent rate of 300–500 events/second.

High-dimensional CyTOF data processing

Individual.fcs files from each sample set were concatenated using Cytobank’s concatenation tool (Mountain View, CA) and normalized with the Normalizer tool and EQ Four Element Calibration Beads to account for signal fluctuations across experiments. Signal intensities were arcsinh transformed with a cofactor of 5 (x_transf = asinh(x/5)). Gating strategies and median expression extraction were performed using Cytobank. Populations of interest were manually gated based on biaxial marker expression. Debris was excluded using 191Ir and 193Ir markers, singlets were identified, and beads were removed. Viable cells were gated by excluding those positive for cleaved caspase-3 and cleaved PARP. Viable cells were then subjected to high-dimensional clustering, such as SPADE analysis.

High-dimensional analysis with SPADE

High-dimensional data analysis was performed using spanning-tree progression analysis of density-normalized events (SPADE) in Cytobank software (Mountain View, CA). SPADE organizes cells into a branched tree structure based on phenotypic similarities, with nodes representing cell clusters. Node color indicates the median marker intensity (blue for low, red for high), while node size reflects the proportion of cells within each cluster. The SPADE algorithm comprises four key steps: (i) density-based down-sampling to equalize cell distributions, (ii) agglomerative clustering to group cells into clusters, (iii) construction of a minimum spanning tree to link clusters, and (iv) up-sampling to map all cells onto the tree. Bone marrow samples from MM patients across all disease stages and healthy donors were analyzed simultaneously to ensure consistent tree structures for each panel. Processed data from the B and TIME panels were exported to R for additional analysis (Supplementary Table 3). Marker intensities were visualized across samples to aid in phenotypic interpretation. Cluster boundaries were manually annotated based on marker expression in biaxial plots and prior knowledge, enabling the identification of biologically relevant cell populations.

Interactive data visualization for CyTOF analyses

We developed an interactive web portal using R software and packages such as shiny, shinyBS, igraph, ggplot2, Cairo, gplots, colorRamps, plotly, and DT (Supplementary Table 3). The portal facilitates the annotation of cell population clusters in B and TIME SPADE analyses and normalizes cell and marker data. Normalization of cell numbers was performed using the formula x = c_B/c_A, where c_B is the sum of cells in a cluster and c_A is the total cells across clusters. Marker expression was normalized using y=Σ(c₁⋅e₁+c₂⋅e₂+ … +c_n⋅e_n)/c_B, where c represents node cell counts and e median marker expression.

We modified the mergeClusters, identifyDAC, and volcanoViewer functions from the SPADEVizR package to analyze marker levels among clusters. A second web portal was developed using R shiny and additional packages (e.g., shinydashboard, htmlwidgets, queryBuilder) to visualize normalized SPADE results via SPADE trees, heatmaps, dot-plots, and box plots. Based on these visualizations, data groups were generated and statistically compared using Dunn’s multiple comparison test following the Kruskal-Wallis one-way analysis of variance by ranks test, with results displayed as volcano plots. Both portals were designed for in-house use and hosted on a Linux-based web server. Venn diagrams were generated with the VennDiagram R package to show common marker changes across MM stages. Principal component analysis (PCA), correspondence analysis (CA), and their visualizations were created using FactoMineR, factoextra, survminer, and survival R libraries. Correlation heatmaps were generated with the ggcorrplot R package (Supplementary Table 3).

Statistical analysis

The tests of normality and the Kolmogorov-Smirnov and Shapiro-Wilk tests were used to assess distribution of data. The outliers were identified by Tukey’s test. Statistical significance of 2 groups was determined by the nonparametric Mann-Whitney U test. Multiple comparisons were conducted using the Holm-Sidak and Tukey methods. The differences in median values among 4 MM stages versus the HD control groups were evaluated by Dunn’s multiple comparison test after the Kruskal-Wallis 1-way ANOVA by ranks, with significant p-values (*p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001) indicated in graphs and figure legends.

Results

Immune landscape profiling of the myeloma microenvironment

To characterize the immune landscape of the myeloma microenvironment, we conducted high-dimensional single-cell profiling using CyTOF analysis on a cohort comprising 16 patients with MGUS, 25 patients with SMM, 43 NDMM patients, and 104 patients with RRMM, compared to 10 HD. We designed a multiscale adaptive and innate immune CyTOF TIME panel for the simultaneous mapping of immune subsets, utilizing 33 heavy metal-conjugated antibodies to identify immune subsets and lineages, as well as a CyTOF B-cell panel with 13 antibodies to evaluate B-cell maturation stages. Bone marrow cells from our cohort were simultaneously stained with both panels and analyzed using CyTOF technology (Figure 1A). Adaptive immunity, encompassing several subsets of T and B cells, and innate immunity, represented by NK cells, DC, myeloid/monocyte lineages, erythroid cells, and thrombocyte lineages, were evaluated in our MM cohort. We analyzed the distribution of various immune cell lineages using comprehensive spanning-tree progression analysis of density-normalized events (SPADE) clustering to illustrate the expression of pan-hematopoietic marker CD45 on immune cell subsets defined by TIME clusters (Figure 1B, middle). Moreover, the expression of CD7 (primarily identifying NK cells), HLA-DR (defining monocytes and DC), CD11b (granulocytes), and CD71 (expressed on erythrocytes) was used to characterize the innate immune ecosystem (Figure 1B, left). Similarly, the expression of CD19 (B cells) and CD3, CD4, and CD8 (T cells and their subsets) was utilized to profile adaptive immunity (Figure 1B, right) in a representative NDMM BM sample. Other clustering markers, differentially expressed in specific immune subsets, were similarly examined using SPADE tree (Supplementary Figure S1). TIME clusters were visualized in z-score clustered heatmaps based on the normalized expression profiles of immune markers, as well as by the expression of selected positive markers to define lineage-specific immune subsets of innate immunity (Supplementary Figure S2) or adaptive immunity (Supplementary Figure S3) across different MM stages and HD samples.

Figure 1.

CyTOF profiling of the immune ecosystem within the myeloma microenvironment. (A) experimental design schema employed in this study. (B) SPADE analysis of immune cell clusters in the TIME of a representative BM sample from an NDMM patient. Each node in the SPADE tree is color-coded based on the median expression of the CD45 marker (center), with node size reflecting the number of cells. Innate immunity is represented by CD7, primarily identifying NK and T cells; HLA-DR expression highlights monocytes and dendritic cells (DC); CD11b marks granulocytes, while CD71 is expressed on erythrocytes (left). Adaptive immunity is characterized by CD19 expression on B cells, and CD3 identifies T cell subsets, including CD4 T helper cells and CD8 cytotoxic T cells (right). The legend provides definitions for innate and adaptive immune clusters.

Immune profiling of the bone marrow ecosystem across myeloma disease stages

To delineate myeloma-driven immune alterations in the BM microenvironment, we compared the immune compartment in premalignant stages (MGUS and SMM) and active disease stages (NDMM and RRMM) to HD. These findings build on growing evidence that the tumor microenvironment undergoes dynamic remodeling during myeloma progression, influencing both innate and adaptive immune compartments.⁷

In the innate immune compartment, we observed a significant reduction in pro-monocytes (pro-Mo) and myelocytes (My1) clusters across all stages of myeloma (MGUS, SMM, NDMM, and RRMM), while pro-erythroblasts (pro-Eb) and meta-myelocytes (meta-My1) clusters were decreased in SMM, NDMM, and RRMM. Contrastingly, innate immune clusters such as myeloblasts (Mb), non-canonical monocytes (non-canMo), myelocytes (My2), neutrophils (Neu3), erythroblasts (Eb), and platelets (Plt) were consistently enriched across MGUS, SMM, NDMM, and RRMM compared to HD, as demonstrated by volcano and box plots. Additionally, clusters of monocytes (Mo) and neutrophils (Neu1 and Neu2) showed an increased frequency primarily in SMM, NDMM, and RRMM, but not in MGUS (Figure 2A,C). Statistical analyses of the adaptive immune compartment demonstrated a significant reduction in the frequency of immature B (Bi), B cells (B), and naïve CD8 T cells (CD8Tn1) clusters across all disease stages (MGUS, SMM, NDMM, and RRMM), whereas CD8Te1 cluster was diminished specifically in SMM, NDMM, and RRMM, and γ/δT cluster in MGUS, NDMM, and RRMM. A decrease in immature immature T (Ti), naïve CD4 T (CD4Tn), effector memory CD4 T (CD4Tem1), and activated central memory T cell (Tcma) clusters was observed exclusively in NDMM and RRMM compared to HD. Conversely, nonspecified T cell (Tns) cluster was elevated across MGUS, SMM, NDMM, and RRMM, with central memory CD4 T (CD4Tcm2) and activated effector memory CD4 T (CD4Tema) clusters showing increased frequency in SMM, NDMM, and RRMM, but not in MGUS. As anticipated, the PC cluster was significantly expanded in NDMM (Figure 2B,D). Moreover, longitudinal analysis of normalized innate and adaptive immune cell frequencies over nine months in nine NDMM and nineteen RRMM follow‑up patients revealed largely stable profiles, with greater modulations observed in innate immune cell subsets (Supplementary Figures S4 and S5).

Figure 2.

Characterization of adaptive and innate immune dynamics in the myeloma microenvironment. Volcano plots display the normalized cell frequency of (A) innate and (B) adaptive immune cell clusters in MGUS, SMM, NDMM, and RRMM patients compared to HD, analyzed using Dunn’s multiple comparison test following the Kruskal-Wallis one-way analysis of variance by ranks test. Colored dots indicate clusters with statistically significant differences in cell counts between groups, identified by a p-value < 0.05 and a Fold change > 2 (right) or < −2 (left). Boxplots show significant differences in the normalized frequency of (C) innate and (D) adaptive immune cell clusters in MGUS, SMM, NDMM, and RRMM relative to HD samples, with *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001 determined by Dunn’s multiple comparison test following the Kruskal-Wallis test; data are presented as median and interquartile range (IQR; 25th-75th percentile) with Tukey whiskers. Sankey diagrams depict the distribution of normalized frequencies for (E) monocyte/myeloid lineage clusters and (F) T cell subsets across HD, MGUS, SMM, NDMM, and RRMM. Related clusters are grouped within columns and arranged sequentially from early to late B-cell maturation stages. (G) venn diagram illustrates 15 intersections among the four MM disease stages: MGUS, SMM, NDMM, and RRMM. Each intersection represents the shared expression of statistically significant immune cell subsets that are either downregulated (-) or upregulated (+) within the myeloma microenvironment compared to HD.

Briefly, the most notable differences included an increase in myeloid lineage clusters (Mb, My, and Neu), a reduction in pro-Mo, and an enrichment of non-canonical Mo clusters in the monocytic lineage (Figure 2E). Within the T cell compartment, we observed reduced CD8Tn clusters across all disease stages and decreased CD8Te clusters in SMM, NDMM, and RRMM but not in MGUS. Similarly, helper T cell clusters (CD4Tcm and CD4Tema) were expanded in SMM, NDMM, and RRMM but not in MGUS (Figure 2F). Collectively, these findings reveal stage-specific immune alterations in both innate and adaptive immunity within the myeloma microenvironment, marked by the progressive expansion of myeloid and monocytic lineages and the suppression of T cell subsets from premalignant to active disease stages (Figure 2G).

Mapping B lymphoid lineage dynamics across premalignant and active myeloma stages

To assess B lymphopoiesis, we analyzed the frequency of B cell clusters in premalignant (MGUS, SMM) and active myeloma (NDMM and RRMM) stages compared to HD. The normalized frequencies of B cell clusters across HD, MGUS, SMM, NDMM, and RRMM are visualized using a Sankey diagram, which highlights a significant decrease in pro-B cells, immature and transitional B cells, and an increase in switched memory B cells and plasmablasts across all myeloma stages relative to HD (Figure 3A).

Figure 3.

Distribution of myeloma B cell lymphopoiesis. (A) The Sankey diagram illustrates the distribution of normalized frequencies of B cell clusters across HD, MGUS, SMM, NDMM, and RRMM. Related clusters are grouped within columns and arranged sequentially from early to late B cell maturation stages. (B) The bar plot displays the normalized fold change in the statistically significant frequencies of B cell clusters in MGUS, SMM, NDMM, and RRMM compared to HD. Statistical analysis was performed using Dunn’s multiple comparison test following the Kruskal-Wallis one-way analysis of variance by ranks test, with *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001 shown. (C) Venn diagram illustrates 15 intersections among the four MM disease stages: MGUS, SMM, NDMM, and RRMM. Each intersection represents the shared expression of statistically significant immune cell subsets that are either downregulated (-) or upregulated (+) within B cell development stages in MGUS, SMM, NDMM, and RRMM compared to HD.

Our findings revealed that switched memory B cell (SM1,3) and plasmablast (PB/PC2) clusters were significantly upregulated even in MGUS, while progenitor clusters (pro-B and pre-BII cells), immature B cells (I1), and plasma cell cluster PC5 were downregulated. Increased switched memory B cells (SM1–3), plasmablasts (PB/PC1–2), and PC (PC1–4) were also detected in SMM versus HD, with the highest abundance of PB/PC1–2 and PC (PC1–3,5) clusters in NDMM. In contrast, reduction of progenies (pro-B, pre-BII and pre-pro-B cells), immature (I1 and I1–2) and transitional B cells (T1), and unswitched memory B cell clusters (UM1,3 in NDMM) and PC5 (in SMM) were noted in SMM and NDMM. Similar downregulation of progenies (pre-pro-B and pro-B), immature (I1–2) and transitional B cells (T1–3), unswitched memory B cell clusters (UM1–4) and PC5 was evident in RRMM compared to HD, with no corresponding upregulation of plasma cells (Figure 3B).

Notably, SMM exhibited differences in switched memory, plasmablast/plasma cell, and plasma cell distributions when compared to MGUS, mirroring the patterns observed when NDMM was compared to both premalignant stages. These differences were more pronounced in MGUS. Additionally, significant reductions in B cell subsets, including progenitors, immature-transitional-naïve B cells, unswitched and switched memory B cell clusters, as well as plasmablast/plasma cells and plasma cells, were observed in RRMM relative to MGUS, SMM, and NDMM (Supplementary Figure S6). Over a nine‑month longitudinal analysis of normalized B‑cell maturation trajectories in eight NDMM and nineteen RRMM patients, we observed subtle modulation of B‑cell progenitors and memory B‑cell subsets, whereas PB/PC1‑2 and PC1‑7 clusters predominantly declined (Supplementary Figure S7). In summary, we observed a decrease in pre-BII cells and an increase in SM1–2 and PC4 clusters in one or both premalignant stages. In active myeloma stages, there was a notable reduction in B cell progenitors (pre-pro-B and pro-B cells), immature and transitional B cells, and unswitched memory B cell clusters, alongside an increase in SM3, PB/PC1–2, and PC1–3,5 clusters (Figure 3C).

Immune checkpoint profiling on innate and adaptive immune subsets across myeloma disease stages

To investigate the regulation of immune responses in the myeloma microenvironment, we evaluated the expression of key immune checkpoints, including killer immunoglobulin-like receptors (KIRs), programmed death-1 (PD-1), and its ligand PD-L1, across innate and adaptive immune subsets in premalignant (MGUS, SMM) and active myeloma (NDMM and RRMM) stages compared to HD. KIRs are inhibitory receptors expressed predominantly on NK cells and some T cell subsets, playing a critical role in immune tolerance,¹⁷ while the PD-1/PD-L1 axis is a well-known regulator of immune evasion in cancer.¹⁸

In the innate immune compartment, KIR expression was significantly upregulated on My1, meta-My1–2, Neu1–3, Neua, Eb, and Plt clusters, particularly in MGUS, SMM, NDMM, and RRMM compared to HD, with the exception of Plt in MGUS. Both PD-1 and its ligand PD-L1 were elevated on mDC and pDC clusters but showed downregulation on Mb, non-canMo, and pro-My clusters. Additionally, differential downregulation of PD-L1 expression was observed on Neu1–3, Neua, and Ba in MGUS, SMM, NDMM, and RRMM; as well as on My1–3 and meta-My1–2 in MGUS and NDMM compared to HD (Figure 4A,B).

Figure 4.

Assessment of immune checkpoints on innate and adaptive immune cells in the myeloma microenvironment (A) SPADE analysis of immune cell clusters in the TIME of a representative BM sample from an NDMM patient. Each node in the SPADE tree is color-coded based on the median expression levels of KIR (CD158), PD-1 (CD279), and PD-L1 (CD274), with node size representing the number of cells. Violin plots show the normalized median expression levels of KIR (upper row), PD-1 (middle row), and PD-L1 (lower row) on (B) innate and (C) adaptive immune cell clusters in MGUS, SMM, NDMM, and RRMM compared to HD samples. Statistical significance is denoted by *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001, determined by Dunn’s multiple comparison test following the Kruskal-Wallis one-way analysis of variance by ranks test.

In the adaptive immune compartment, KIR expression was significantly increased on CD4Tem2, Tcma, and Tns clusters across all stages of myeloma. Furthermore, KIR upregulation was observed on CD4Tcm1, CD4Tema, CD8Tem, and γ/δT clusters specifically in SMM, NDMM, and RRMM. PD-1 expression was significantly increased on Bi and B clusters, while Tns cluster exhibited reduced PD-1 levels. PD-L1 expression was predominantly downregulated on PC clusters in MGUS and SMM (Figure 4A,C). In summary, these findings highlight distinct and stage-specific dysregulation of immune checkpoints in both innate and adaptive immune subsets, reflecting their potential role in immune evasion and disease progression in the myeloma microenvironment.

Deciphering immune cluster patterns to predict clinical outcomes in myeloma

To assess the relationship between immune cluster distributions and clinical parameters, correspondence analysis (CA) was employed to reduce high-dimensional immune cell frequency data into a smaller set of explanatory components, facilitating the simultaneous analysis of immune subsets and individual patient data in a unified space (Figure 5A). Patients were grouped based on their tendency to harbor specific immune subsets, while immune subsets were organized according to their propensity to co-occur in the same patients. The first component (CA1) was primarily driven by the presence of B cell progenitors (pre-BI and pre-BII), immature B cell cluster (I1), myelocyte cluster (My2), naïve cytotoxic T cells (CD8Tn2), central memory T helper cells (CD4Tcm2), and two plasma cell clusters (PC2 and PC7). In contrast, the second component (CA2) was directed by a different set of clusters, including plasma cells (PC1–4,7), plasmablasts (PB/PC2), switched memory B cells (SM1), nonspecified T cells (Tns), and myelocytes (My2) (Figure 5B).

Figure 5.

Correlation of the immune ecosystem with PC clonal clusters and clinical outcomes. (A) the first two components (CA1 and CA2) of correspondence analysis, accounting for 22% of the co-association between immune clusters (derived from B and TIME panels) and MM patients, are shown. Immune clusters are represented as triangles. (B) contributions of immune clusters from B and TIME panels to CA1 and CA2 are depicted. (C) Kaplan-Meier analysis of 39 and 37 patients with high scores relative to HD (red curve, CA2 high) versus patients with scores similar to HD (blue curve, CA2 low) reveals statistically significant differences in overall survival (p = 0.029) and progression-free survival (p = 0.007), respectively.

Patterns of co-occurrence in these CA components were leveraged to establish statistical relationships between the immune landscape (analyzed using B and TIME panels) and clinical outcomes. While CA1 was not associated with clinical outcomes, CA2 exhibited a significant correlation with both progression-free survival (PFS) and overall survival (OS). Patients with high CA2 scores were stratified into groups, with 37 and 39 patients assigned as high scorers for PFS and OS, respectively. Kaplan-Meier survival analysis revealed that patients with high CA2 scores experienced significantly shorter OS (log-rank p = 0.029) and PFS (log-rank p = 0.007) (Figure 5C). In a multivariate Cox proportional hazards model of RRMM patients, significant associations with OS were identified for My1, meta-My1–2, Neu1, Neu2, pro-Eb, and NK cell subsets. Similarly, Neu1–3, meta-My1–2, My1–2, mDC, pro-Mo, CD8Tem, pro-Eb, Eb, Bi, and B cell subsets emerged as significant predictors of progression‑free survival (Supplementary Figure S8). These findings, derived from the comprehensive immune profiling of 188 MM patients using CyTOF, suggest that the immune landscape holds prognostic value and may provide insights into disease progression and outcomes.

Discussion

In the context of myeloma research, CyTOF (cytometry by time-of-flight) merges flow cytometry with mass spectrometry via metal isotope-tagged antibodies to quantify over 50 cellular parameters at single-cell resolution, enabling unparalleled phenotypic and functional profiling, signaling-pathway analysis, and rare-cell detection within complex ecosystems, and thereby revealing immune microenvironment dynamics and tumor heterogeneity beyond the reach of conventional cytometry and genomic approaches.^19–21 Using CyTOF, we analyzed BM samples from MM patients across disease stages to dissect innate and adaptive immune contributions to pathogenesis, employing two panels, 13 markers capturing B-cell lymphopoiesis stages and 33 markers profiling immune subsets and malignant PC, augmented with immune checkpoint molecules to assess subset activation and function. Myeloma evolves from the premalignant conditions MGUS and SMM³; by profiling BM samples across MGUS, SMM, NDMM, and RRMM using integrated immunophenotypic and functional analyses, we aimed to uncover the critical drivers of progression and map the underlying cellular ecosystems.

The dysregulated T cell landscape in MM is marked by a shift of CD8⁺ cells toward terminally differentiated effector memory (TEMRA) cells, a decline in effector memory (TEM) cells, and clonal PD 1⁺ T cell expansions indicative of exhaustion, alongside altered γδ T cell function that nevertheless correlates with improved outcomes.^22–24 Patients with long-term survival demonstrate higher proliferation rates of cytotoxic T cell clones and reduced Th17 and Tregs.²⁵ Among T follicular helper (Tfh) subsets, a decrease in Tfh2 cells and an increase in Tfh17 cells²⁶ further underscore immune suppression and tumor promoting dynamics, together revealing targets for therapeutic intervention. Our findings revealed reductions in CD8+ naïve and effector cytotoxic T cells (CD8⁺), and γδ T cells in all MM stages, except for preserved effector cytotoxic T cells in MGUS and γδ T cells in SMM stage. Conversely, central memory and HLA-DR⁺ activated effector memory CD4⁺ T helper cells were upregulated in SMM and active MM stages, but not in MGUS, suggesting that the MGUS-to-SMM transition may involve shifts in central and effector memory T helper cells as well as effector cytotoxic T cells. In active MM, significant adaptive immune suppression was marked by downregulation of immature T cells, naïve T cells, effector memory T helper cells, and active central memory T cells within the TIME. In summary, our findings highlight profound alterations in T cell subsets, including cytotoxic, helper, and regulatory populations, collectively contributing to a highly immunosuppressive tumor immune microenvironment in MM, which plays a pivotal role in disease progression and immune evasion.

B-cell regeneration in MM differs markedly from healthy individuals, shaped by disease stage and treatment regimens. At MM diagnosis, bone marrow B‑cell precursors, transitional/naive B cells, and plasma cells are depleted; post‑ASCT these populations partially rebound by day + 100, though memory B cells and plasma cells remain low.²⁷ Patients with long-term disease control show more robust recovery of B‑cell precursors, plasma cells, and peripheral pre – germinal center B cells, supporting improved immune surveillance.²⁸ In NDMM, our previous study has documented a pronounced expansion of malignant PC subsets, reflecting profound intra- and inter-clonal heterogeneity marked by overexpression of MMSET, Notch 1, and CD47; deregulation of key B cell signaling regulators (IRF4, CXCR4, Bcl6, c-MYC, MYD88, sXBP-1); aberrant PC markers (CD319, CD269, CD200, CD117, CD56, CD28); and emergence of subclonal stemness markers (Nestin, SOX2, KLF 4, Nanog) that are associated with prognosis.²⁹ Here, analysis of B-cell lymphopoiesis revealed notable reductions in B-cell progenitors, immature B cells, transitional B cells, and unswitched memory B cells in SMM and active MM stages. In contrast, switched memory B cells and plasmablasts were upregulated as early as MGUS, with this trend intensifying in SMM and NDMM. Furthermore, our prior study revealed dysregulated B cell development in MM, marked by elevated MMSET, MYD88, c Myc, CD243, Notch 1, and CD47 across myeloma B cell lymphopoiesis (including premalignant stages), along with aberrant expression of B cell regulators (IRF 4, Bcl-2, Bcl 6, sXBP-1), plasma cell markers (CD52, CD44, CD200, CD81, CD269, CD117, CXCR4), and stemness factors (Nanog, KLF 4, Oct3/4, Sox2, RARα2) within the BM microenvironment.²⁹ These findings collectively emphasize the profound impact of MM on B-cell development, with significant alterations in lymphopoiesis and immune signaling contributing to disease progression and immune dysfunction. Understanding these mechanisms may provide valuable insights for therapeutic strategies aimed at restoring immune competence and targeting malignant PCs.

Malignant PCs disrupt normal myeloid differentiation in the BM, impairing maturation, causing neutropenia, and triggering a compensatory “left shift” with elevated circulating immature myeloid cells (myeloblasts, myelocytes, and metamyelocytes), reflecting inflammation and marrow stress.³⁰ Extracellular vesicles (EVs) released by malignant PCs alter hematopoietic stem and progenitor cells (HSPCs), promoting early progenitors while depleting late-stage precursors like common myeloid progenitors (CMPs), which contribute to disrupted hematopoiesis.³¹ Tumor-associated monocytes and neutrophils infiltrating the myeloma BM contribute to immune suppression through interactions with the malignant clone and other components of the TIME. Conversely, although higher circulating neutrophil counts correlate with improved survival, these neutrophils exhibit elevated CD64 (FcγRI) and other aberrant markers, indicative of an activated yet dysfunctional state with impaired chemotaxis and phagocytosis. Moreover, reduced granulocyte proportions, diminished immature granulocyte populations, and disrupted granulocyte maturation within the TIME correlate with poorer clinical outcomes.^10,12,17,32 Our findings revealed an increase in myeloblasts and non-canonical monocytes, accompanied by a decrease in pro-monocytes, in both premalignant and active MM conditions. These observations confirm that innate immune cells infiltrate the BM as early as the MGUS stage. When comparing premalignant stages to active MM, we observed no increase in pro-myelocytes, neutrophils (Neu1–2), or monocytes in MGUS. Moreover, no decrease in meta-myelocytes was detected in MGUS. These findings suggest that immune cells, particularly those from monocytic and granulocytic lineages, play a significant role in the progression from MGUS to SMM.

MM cells impair erythropoiesis (e.g. by downregulating key erythroid transcription factors GATA1 and KLF1 in HSPCs), with most patients presenting moderate to severe anemia at diagnosis.^33,34 Elevated immunoglobulin levels, along with altered erythrocyte morphology and decreased deformability, increase erythrocyte aggregation, leading to hyperviscosity and microcirculation issues.^33,35 In MM patients, pro-erythroblasts (CD71⁺ Ter119⁺ low) and Ter119⁺ subpopulations are reduced, while late-stage erythroblasts accumulate due to blocked differentiation, with the inflammatory BM environment further suppressing erythroblast proliferation and maturation.^33,34 In MM, thrombocytopenia, due to BM infiltration and chemotherapy, combined with platelet dysfunction from abnormal plasma proteins increases bleeding risk, whereas MM-induced platelet hyperactivation may contribute to thrombotic events despite low platelet counts.³⁶ Our study observed an increase in erythroblasts and platelets across all stages of myeloma, confirming the early infiltration of innate immune cells in MGUS. In contrast, the unchanged frequency of pro-erythroblasts in MGUS suggests their potential involvement in the transformation from MGUS to SMM. MM disrupts erythropoiesis and platelet function through mechanisms involving inflammatory cytokines and transcription factor dysregulation, leading to anemia and thrombocytopenia, with our findings highlighting early innate immune cell infiltration and distinct erythroblast dynamics in MGUS.

Immune exhaustion mediated by the PD 1/PD L1 axis drives immune evasion and disease progression in MM by suppressing autoreactive T cell activation and proliferation, leading to T cell exhaustion, reduced cytokine production, and impaired cell lysis. In MM, PD-L1 expression is upregulated, driven by cytokines such as IL-6 from BM stromal cells, and is notably higher in advanced and relapsed cases.^37,38 However, our findings revealed downregulated PD-L1 expression on MM cells in MGUS and SMM. Within the BM, PD-L1+ DCs and myeloid-derived suppressor cells (MDSCs) further suppress immune responses, with pDCs exhibiting higher PD-L1 levels than PCs, thereby contributing to immune evasion. Additionally, PD-1 expression is increased on T cells, particularly effector memory T cells, and NK cells in MM, impairing their functions.^39,40 Our analyses showed that the expression of PD-1 and its ligand PD-L1 was significantly upregulated on pDCs in both premalignant and active MM stages and on mDCs in the RRMM stage. Moreover, our data revealed upregulated expression of PD-1 on both immature and mature B cells in SMM and active MM stages. Conversely, downregulation of PD-1 and PD-L1 was observed on myeloblasts, pro-myelocytes, non-canonical monocytes, and nonspecified T cells (only PD-1) in both premalignant and active MM. Similarly, PD-L1 expression was decreased on myelocytes and meta-myelocytes in MGUS and NDMM, as well as on neutrophils (including HLA-DR+ activated neutrophils) and basophils across premalignant and active MM stages. Likewise, chronic stimulation by malignant PC exhausts NK cells marked by reduced activating receptors (e.g., NKG2D, DNAM 1, CD16), upregulated PD 1, and diminished cytotoxicity and antibody-dependent cellular cytotoxicity (ADCC).^41,42 KIRs, highly polymorphic molecules expressed on NK cells and some T cells, regulate immune responses by binding HLA class I molecules and include both inhibitory and activating types that influence NK cell activity.^43,44 While high levels of KIR/CD158 expression were on NK cells, it was upregulated on central memory T helper cells, effector memory both T helper cells (including HLA-DR+ activated cells) and cytotoxic T cells, and γ/δT cells in SMM and both active MM stages. Interestingly, KIR expression was also increased on myelocytes, meta-myelocytes, neutrophils (including HLA-DR+ activated neutrophils), erythroblasts, and platelets in both premalignant and active MM stages, with the exception of no change in KIR expression on platelets in the MGUS stage. Immune evasion in MM is driven by PD-1/PD-L1-mediated immune exhaustion, with our findings revealing PD-L1 upregulation on DC subsets that suppress T cell activation, PD‑L1 down‑regulation in myeloid subsets, dynamic changes in B cells, and increased KIR expression on T‑cell and myeloid populations in both premalignant and active MM stages, all of which contribute to immune dysregulation and disease progression.

Comprehensive analysis of tumor cells within their immune landscape, encompassing simultaneous evaluation of all immune subsets alongside tumor clones, represents a robust approach for predicting patient outcomes. In our study, patients with immune landscapes characterized by higher frequencies of specific PC subsets (notably PC1–4 and PC7), plasmablasts, switched memory B cells, and myelocytes exhibited worse clinical outcomes, including shorter OS and PFS. In summary, our high-dimensional CyTOF analysis of 188 MM patients illuminates distinct immune landscapes that correlate with clinical outcomes, underscoring CyTOF’s power to dissect MM pathophysiology, stratify patient risk, and guide targeted or immune-based therapies by revealing mechanisms of tumor promotion, immune suppression, and functional modulation. Future studies should incorporate larger, external cohorts to validate these findings and establish clinical utility.

Funding Statement

This study was supported by the Slovak Research and Development Agency APVV-16-0484 (JJ), APVV-20-0183 (JJ), APVV-19-0212 (DC), and APVV-23-0482 (DC) grants; Scientific Grant Agency VEGA 2/0088/23 (JJ) and VEGA 2/0087/23 (DC) grants; and Research Executive Agency grant agreement No. 609427-SASPRO 0064/01/02 (JJ). Moreover, this project was funded by the EU NextGenerationEU through the Recovery and Resilience Plan for Slovakia under project No. 09I03-03-V04-00451 (JJ). This work was performed during the implementation of the Building-up Centre for advanced materials application of the Slovak Academy of Sciences, ITMS project code 313021T081 supported by the Research & Innovation Operational Programme funded by the ERDF (JJ).

Author contributions

DC performed experiments and analyzed data, and contributed to the writing of the manuscript. GB and LK performed bioinformatics analyses and contributed to the writing of the manuscript. ZV performed experiments. LD, ML, EK and DMD contributed with clinical specimens and collected clinical data. KCA edited the manuscript. JJ conceived and designed the study, performed experiments and analyzed data, collected clinical data and wrote the manuscript.

Disclosure statement

The authors declare the following competing interests: EK declares honoraria (Amgen, Janssen, Genesis Pharma, Pfizer, Takeda, GSK) and research support (Amgen, Janssen, Pfizer); KCA declares advisory role (Pfizer, Amgen, Astrazeneca, Janssen, Precision Biosciences), board membership (C4 Therapeutics, Raqia, NextRNA, Window, Mana, Starton) and ownership interests (C4 Therapeutics, Oncopep, NextRNA, Raqia). Other authors declare no competing financial interests.

Data availability statement

All the data that support this study are presented in the paper.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/2162402X.2025.2542333.