Introduction

Autologous stem cell transplantation (ASCT) and maintenance form the treatment backbone for most patients with multiple myeloma (MM) [1]. Along with the incorporation of novel immune-based treatment strategies, this paradigm has led to a consistent improvement in the outcomes of patients over the past two decades [2]. Despite this, a cure for MM remains elusive and most patients relapse. In addition, 2–4% of patients develop therapy-related myeloid neoplasms (t-MN) which are associated with a dismal survival, significantly contributing to post-ASCT mortality [3,4,5]. Thus, relapse and t-MN represent major barriers to improve the post-ASCT outcomes in MM. Therapy-related myeloid neoplasms are highly enriched in pathogenic variants in the tumor suppressor gene TP53 (TP53mut), which explains their aggressive course and the lack of response to the currently available therapies [6]. There is emerging concern that chimeric antigen receptor T cell therapy (CAR-T) can be complicated by the development of t-MN in upto 10–20% of patients [7, 8].

Given that ASCT is associated with significant morbidity and resource utilization, there is a substantial interest in identifying high-risk patients for relapse and t-MN before ASCT. While the focus thus far has been the neoplastic cells, recently the role of tumor immunosurveillance has emerged. A permissive immune tumor microenvironment (iTME) contributes to treatment resistance and progression of MM [9, 10]. Increased T-cell exhaustion and immune senescence have been associated with poor outcomes after standard MM therapies, ASCT and novel T cell redirecting therapies [11,12,13,14,15,16]. Immune reconstitution following autologous [17] and allogeneic SCT [18] impacts the post-transplant outcomes and immunomodulation is being pursued to improve long-term outcomes.

Similarly, TP53mut MN are characterized by a highly immunosuppressed immune milieu associated with a reduced number of immune effector cells (IEC) including cytotoxic T- and natural killer (NK)-cells, as well as the expansion of regulatory T-cells (Tregs), myeloid-derived suppressor cells (MDSC), and exhausted T-cells [19,20,21,22]. The expression of various exhaustion receptors is increased following hypomethylating agents and some patients with myelodysplastic syndromes can respond to checkpoint blockade [21, 22]. Finally, clonal expansion of terminally differentiated, immune senescent T-cells is seen in all MDS subgroups indicative of an ineffective T-cell response that fails to control disease progression [23, 24]. Peripherally mobilized blood-derived stem cell (PBSC) grafts are responsible for the reconstitution of the myeloid and immune compartments following the myeloablative doses of melphalan used in ASCT. We hypothesized that pre-existing immune abnormalities in the mobilized PBSC will lead to reconstitution of an immunosupressive iTME and contribute to early relapse and/or t-MN development. To that end, we explored the immune composition of mobilized PBSC and its relationship with subsequent MM relapse and t-MN development.

Methods

Study population

After institutional review board approval, we screened patients with active MM that were evaluated at Mayo Clinic, Rochester, MN between 01/01/2003 and 12/31/2020. Patients who underwent an ASCT and had cryopreserved PBSC products available for research were included. We enriched this cohort with patients who developed t-MN to study this group. t-MN was defined using the 2016 World Health Organization criteria [25]. The revised international staging system (R-ISS) was used to risk stratify MM patients [26]. Hematologic response to therapy and engraftment syndrome were defined per internationally accepted criteria [27, 28]. Short duration of remission for MM following ASCT was defined as <24 months if no post-ASCT maintenance was utilized or <48 months if post-ASCT maintenance was not utilized.

Mass cytometry

The antibody panels included 37 lymphoid- and myeloid-based markers each (Supplementary Table 1). For primary conjugations, purified antibodies were obtained in carrier protein-free phosphate-buffered saline and labeled using the X8 polymer MaxPAR antibody conjugation kit (Fluidigm) according to the manufacturer’s protocol. All antibodies were titrated to optimal staining concentrations using peripheral blood mononuclear cells. Antibody master mixes were prepared fresh for each experiment. All samples were processed identically. PBSC were collected after mobilization using peripheral blood leukapheresis. Using validated processes in a Current Good Manufacturing Practice (cGMP) laboratory, the cells were concentrated and mixed with a cryopreservation solution to give a final white blood cell concentration of 300 × 106 nucleated cells/ml in 10% DMSO, 10% plasma, and 30% PlasmaLyte-A (Baxter, Deerfield, IL) or Normosol-R (Hospira, Lake Forest, IL). The cells were then frozen using a controlled rate freezing process and stored in vapor phase liquid nitrogen tanks. The main products were stored in cryobags ranging from 50–100 ml in addition to 1.5 ml cryovials for additional testing as needed. Cryopreserved cells were resuscitated for mass cytometry analyses by rapid thawing and were rested in RPMI 1640 (20% fetal bovine serum) for 60 min prior to staining. Staining was performed using Fluidigm’s protocol. Briefly, 1–3 million cells were stained for viability with 5 mM cisplatin for 5 min at room temperature and quenched with cell staining medium (CSM; Fluidigm). Cells were then incubated for 10 min at room temperature with human FcR blocking reagent (Biolegend) and stained with the surface antibody cocktail for 60 min at 4 °C with gentle agitation. Finally, cells were washed twice with CSM, fixed with 1.6% paraformaldehyde, washed with CSM, and resuspended in 1:1000 solution of Iridium intercalator diluted in MaxPar Fix and Perm buffer (Fluidigm) for 20 min at room temperature. Prior to the acquisition, cells were washed twice in CSM and twice in deionized water and were then diluted to a concentration of 0.5 million cells per milliliter in water containing 10% of EQ 4 Element Beads (Fluidigm). Cells were filtered through a 35-μm membrane prior to mass cytometry acquisition. Samples were then acquired on a Helios mass cytometer.

Mass cytometry data analysis

Flow cytometry standard files were normalized and concatenated using the Fluidigm acquisition software. Flow cytometry standard files were uploaded to the OMIQ software from Dotmatics (www.omiq.ai, www.dotmatics.com), where transformation and cleaning (doublets, debris), were done as previously described [11]. To correct for between-sample technical variations (“batch effects”), we used fdaNorm [29] within the omiq.ai platform. Characteristic images of marker expression before and after normalization are shown in Supplementary Fig. 1. Clustering and visualization were performed within the omiq.ai platform using PhenoGraph [30] and UMAP [31], respectively. When clustering, 50,000 CD45+ events per file were used when using the lymphoid panel due to computational constraints. When using the myeloid panel, we manually excluded T-, B-, and NK-cells using canonical markers (CD3, CD19, and a combination of CD7 and CD56, respectively) and clustered the remaining cells and using 50,000 CD45+ events per file. Identified clusters were then exported for downstream statistical analyses. All immune subset frequencies are reported as a percent of CD45+ cells. All raw mass cytometry data are uploaded to FlowRepository (https://flowrepository.org/).

Statistical methods

Fisher’s exact test was used to compare categorical variables and Wilcoxon rank sum (2 groups) or Kruskal-Wallis test ( > 2 groups) for continuous variables. Kaplan-Meier survival analysis was used to estimate the overall survival (OS) from diagnosis of MM, and progression-free survival (PFS) and tMN-free survival (MNFS) from ASCT. The log-rank test was used to compare groups. A Cox proportional hazards model was used to perform multivariable analyses for time to event outcomes. Patients alive and without progression to MM or t-MN, were censored for PFS and MNFS analyses, respectively. Hierarchical clustering of patients according to the abundance of immune subsets was performed using Ward’s minimum variance method. Principal component analyses were performed using log transformed and scaled values of immune subset frequencies. Statistical analyses were performed using the JMP Pro statistical software version 14.1 (SAS Institute, Cary, NC). Correlation analyses were performed in R using the corrplot package. A 2-sided false discovery rate (FDR) adjusted P-value of <0.05 was considered significant when multiple comparisons were performed; otherwise, a P-value of <0.05 was considered significant.

Results

The baseline clinical characteristics of the 54 patients included in the study are shown in Table 1. Given the referral nature of our institution, patients with high-risk diseases were overrepresented. Also, given the limitations imposed by sample availability, we included patients that had PBSCs collected but did not receive lenalidomide maintenance, which became standard of care later in the study time frame. As a result, 19 patients did not receive post ASCT maintenance.

Table 1 Baseline Characteristics for patients in the study.
Full size table

The immune landscape of mobilized PBSC

We generated 2-dimensional UMAP maps of the data generated by the 2 panels (lymphoid and myeloid) in the manually gated CD45+ cells (Supplementary Figs. 2 and 3, respectively). These analyses demonstrated that major lineages separated well in the 2-dimensional space. The most abundant cell populations were T-cells (49% of the CD45+ cells) and myeloid cells (43% of the CD45+ cells, Supplementary Fig. 4), which was not surprising considering these were mobilized cells. We then clustered all CD45+ cells using the lymphoid panel and projected the identified clusters on the respective UMAP (Fig. 1A, B). A heatmap of the marker expression of the lymphoid populations identified with the lymphoid panel is shown in Fig. 1C. To better characterize T- and NK-cell diversity, we subclustered on the manually gated respective cell subsets and excluded the markers not expressed on T- and NK-cells, respectively. Subclustering on T-cells did not increase the T-cell cluster diversity (not shown). Therefore, only the T-cell clusters that resulted after clustering with the lymphoid panel on CD45+ cells were considered for further analyses. When subclustering on NK-cells however, an additional 16 NK-cell clusters were identified and were projected on a 2-dimensional UMAP map (Fig. 2A, B). The corresponding heatmap of their marker expression is shown in Fig. 2C. We then subclustered on manually gated myeloid cells (excluding markers not expressed on myeloid cells) and identified 27 myeloid clusters which were projected on a UMAP map (Fig. 3A, B). A heatmap of their marker expression is shown in Fig. 3C.

Fig. 1: Lymphoid Subsets.

A Phenograph Clusters projected on UMAP plot of all identified lymphoid subsets (NK, T, and B cells) and CD34+ cells using the lymphoid panel. Myeloid or lineage negative subsets were excluded for clarity. B Canonical marker expression (CD3, CD19, CD56, CD34) of major lineages. C Clustered heatmap showing the expression of all markers (lymphoid panel) expressed in identified lymphoid subsets across all files.

Full size image
Fig. 2: Natural Killer Cell Subsets.
figure 2

A Phenograph clusters projected on UMAP plot of all identified NK cell subsets after sub-clustering manually gated NK Cells and using the lymphoid panel. B Canonical NK-cell marker expression (CD16, CD56). C Clustered heatmap showing the expression of all markers expressed in NK cell subsets after sub-clustering of manually gated NK cells.

Full size image
Fig. 3: Myeloid Subsets.
figure 3

A Phenograph Clusters projected on UMAP output of all identified myeloid subsets identified by clustering all CD45+ non-T, B, or NK cells. Lineage-negative clusters not expressing canonical myeloid markers (CD11c, CD33) were excluded. B Canonical marker expression (CD33, CD11c, CD123). C Clustered heatmap showing the expression of all markers expressed in myeloid cell subsets after sub-clustering of manually gated myeloid cells.

Full size image

Next, we explored the associations between the immune subsets and the dichotomized clinical outcomes of interest. To identify if specific immune subsets correlated with short or long remissions, the frequencies of the identified immune subsets between the 2 groups were compared. We found no differences at our prespecified level of significance (FDR corrected P-value < 0.05). When considering trends (non-FDR corrected P-value < 0.05), the activated (CD69/CD161/KLRG1+) memory (CD27/CD28/CD127+) CD8 T-1 cell subset was enriched in the patients with long remission compared to those with short remission (0.9% vs. 0.3%, P = 0.02). Similarly, no significant differences between the patients who subsequently developed t-MN compared to those who did not were identified. Several trends of populations of potential interest became apparent (Supplementary Table 2). NK-cell subcluster-7, characterized by CD56high, CD159a+ (NKG2A) phenotype, was enriched in the patients who subsequently developed t-MN (0.33% vs. 0.17%, P = 0.006). Enrichment of these well-described inhibitory NKG2A+ NK-cells has been noted in the bone marrow of patients with MDS and AML [32]. A “terminally mature” [33, 34] CD56/CD16/CD27dim, CCR4/CXCR3/CCR7high NK-15 subset was enriched in patients developing t-MN (0.09% vs. 0.06%, P = 0.001), and is thought to have T-cell immunosuppressive properties [35]. The T-cell subsets T-5 and T-20, two CD8+ naive subsets with bone marrow homing (CXCR4 + ) [36] and cytotoxic (NKG2D + , CD226 + , respectively) receptors were decreased in patients who later developed tMN as was the T-1 T cell subset described above. These observations imply that changes associated with decreased immunosurveillance may be evident in G-CSF mobilized PBSC products years before the development of t-MN.

We then compared patients that had been mobilized with G-CSF alone or G-CSF and plerixafor and found no differences (including those with a non-FDR corrected p-value < 0.05), suggesting that plerixafor use does not significantly influence the immune contexture of mobilized PBSC.

The structure of the immune landscape correlates with clinical outcomes

We hypothesized that the immune subsets with highly correlated frequencies were more likely to be co-regulated. A correlation matrix and a list of all identified significant (P < 0.05) correlations sorted by their R values are shown in Supplementary Fig. 5 and provided as a Supplemental spreadsheet, respectively. This analysis identified 3 patients with unique immune compositions and poor outcomes (supplemental information). To further explore the major drivers of variability in the data, we performed a principal component (PC) analysis. The first 2 PCs (PC-1 and PC-2) explained 15.2% and 12.5% of the variability in the data, respectively. The top and bottom 10% of immune subsets with the highest and lowest loadings for each PC are shown in Fig. 4A. Conceptually, an immune subset with high loadings within a given PC, correlates the most with that PC and is most influential in defining its immune subset composition. Consequently, patients that have high scores for this PC are expected to have higher abundance of these immune subsets. Immune subsets with high loadings within PC-1 consisted exclusively of CD4+ cells with early memory phenotypes (CD27/CD28/CD127+), whereas those with low loadings for PC-1 consisted mostly of the rare myeloid and T cell subsets that were identified during the correlation analyses described above and overlapped with most subsets with high loadings within PC-2. Finally, immune subsets with the lowest loadings for PC-2 consisted exclusively of myeloid subsets, some of which with phenotypes consistent with plasmacytoid dendritic cells (pDCs; M-26:CD123/CD303 + ) or progenitor cells (M-22, M-11: CD13/CD90/CD7/CD15 + ). These data suggest that the major drivers of variability in PBSC include subsets associated with an improved or impaired immunosurveillance (e.g., early memory CD4 subsets), various myeloid subsets, and as expected, mobilized progenitor cells.

Fig. 4: Correlation of Immune subsets with Clinical Outcomes.
figure 4

A Immune subsets with the highest/lowest loadings for each principal component (PC). B Patients with high PC-2 scores had worse tMN-free survival, C progression-free survival from transplant and D inferior overall survival from ASCT.

Full size image

Next, we analyzed these components to explore the relationships between immune system composition with clinical outcomes of interest. A high PC-2 score was associated with an inferior OS and PFS from ASCT and MNFS (Fig. 4B–D). Worse OS was also noted from MM diagnosis (not shown). Interestingly, despite the difference in outcomes, the baseline characteristics were largely comparable between the two cohorts, except for age and R-ISS stage 3 status. Patients with high PC-2 scores were older (63 vs. 59 years, P = 0.009) and were noted to have higher proportion of R-ISS stage 3 disease (54% vs. 14%, P = 0.005). In a multivariate analysis that include PC-2 score status (high vs. low), R-ISS stage 3 disease and age, PC-2 score status was not independently associated with PFS and OS from ASCT (not shown). In contrast, the high PC-2 cohort was associated with an inferior MNFS (Fig. 4B). The high PC-2 score remained an independent predictor of shorter MNFS on a multivariate analysis [HR 2.2 (95%CI: 1.1–4.4), P = 0.036] including lines of therapy received and whether lenalidomide maintenance was used or not. To understand the relationship between MNFS and MM PFS, we evaluated the extent of overlap between the development of tMN and progression of MM. Sixteen patients did not have MM progression during follow-up, of these 7 developed tMN. Thirty-eight patients had MM progression during follow-up, of which 13 also developed tMN. In these 13 cases, 12 patients developed tMN much later than the first MM progression with the exception of one patient who was diagnosed with tMN at the same time as their first MM relapse.

Immune subsets that were significantly different (FDR corrected P-value < 0.05) between the patients with high and low PC-2 scores are shown in Fig. 5. These, not unexpectedly, largely overlap with the dominant (high/low loadings) immune subsets within PC-2. Immune subsets enriched in the high PC-2 cohort included the “NKT-like” (CD56 + ) T-2 subset which expressed the inhibitory CD159a receptor [37] and TIGIT; the terminally differentiated (CD27/CD28-, CD57/KLRG1 + ) and exhausted (TIGIT/PD-1 + ) T-3 subset the exhausted (TIGIT/PD-1 + ) T-7 and T-14 subset, the immunosenescent (CD27/CD28-, CD57/KLRG1 + ) T8 subset as well as the terminally exhausted (C27/C28-, KLRG1/PD-1/TIGIT + ) TCRgd subset. Among subsets that were low within the patients with high scores for PC-2 included the HLA-DR expressing M2 subset; the CX3CR1 + M-3 subset that expressed markers of nonclassical monocytes (CD14dim/CD16 + ) and M1 polarization (CD86/CD16 +); the M-9, M-10 and M-26 subsets that expressed dendritic cell markers (CD1c/HLA-DR+ and CD123/CD303/HLA-DR + , respectively), the M-11 and M-22 subsets with a phenotype consistent with progenitor cells (CD90/CD7/CD13 + ). Given the known detrimental effects of malignant plasma cells on BM hematopoiesis [38, 39], and T cell immunosenescense [14] which could be reflected in the abundance of these subsets in PBSC, we evaluated differences in the quality of hematologic response after induction chemotherapy (VGPR or better versus not and degree of clearance of bone marrow plasma cells) between PC-2 high and PC-low patients but found no differences. These data suggest that T cell subsets associated with decreased tumor immunosurveillance are increased, whereas those associated with antigen presentation (DCs, HLA-DR + ) improved tumor outcomes (M1 monocytes) and progenitor cells are decreased in the PBSC products of patients with poor outcomes after ASCT.

<div class="c-article-section__figure js-c-reading-companion-figures-item" data-test="figure" data-container-section="figure" id="figure-5" data-title="Immune subsets significantly different (FDR p value 

Fig. 5: Immune subsets significantly different (FDR p value < 0.05) between patients with high (PC-2-High) and low (PC-2-Low) PC-2 scores.
figure 5

All p-values noted are FDR corrected.

Full size image