Identification of the immune checkpoint signature of multiple myeloma using mass cytometry‐based single‐cell analysis

Abstract Objectives New targets or strategies are needed to increase the success of immune checkpoint‐based immunotherapy for multiple myeloma (MM). However, immune checkpoint signals in MM microenvironment have not been fully elucidated. Here, we aimed to have a broad overview of the different immune subsets and their immune checkpoint status, within the MM microenvironment, and to provide novel immunotherapeutic targets to treat MM patients. Methods We performed immune checkpoint profiling of bone marrow (BM) samples from MM patients and healthy controls using mass cytometry. With high‐dimensional single‐cell analysis of 30 immune proteins containing 10 pairs of immune checkpoint axes in 0.55 million of BM cells, an immune landscape of MM was mapped. Results We identified an abnormality of immune cell composition by demonstrating a significant increase in activated CD4 T, CD8 T, CD8+ natural killer T‐like and NK cells in MM BM. Our data suggest a correlation between MM cells and immune checkpoint phenotypes and expand the view of MM immune signatures. Specifically, several critical immune checkpoints, such as programmed cell death 1 (PD‐1)/PD ligand 2, galectin‐9/T‐cell immunoglobulin mucin‐3, and inducible T‐cell costimulator (ICOS)/ICOS ligand, on both MM and immune effector cells and a number of activated PD‐1+ CD8 T cells lacking CD28 were distinguished in MM patients. Conclusion A clear interaction between MM cells and the surrounding immune cells was established, leading to immune checkpoint dysregulation. The analysis of the immune landscape enhances our understanding of the MM immunological milieu and proposes novel targets for improving immune checkpoint blockade‐based MM immunotherapy.

Objectives. New targets or strategies are needed to increase the success of immune checkpoint-based immunotherapy for multiple myeloma (MM). However, immune checkpoint signals in MM microenvironment have not been fully elucidated. Here, we aimed to have a broad overview of the different immune subsets and their immune checkpoint status, within the MM microenvironment, and to provide novel immunotherapeutic targets to treat MM patients. Methods. We performed immune checkpoint profiling of bone marrow (BM) samples from MM patients and healthy controls using mass cytometry. With highdimensional single-cell analysis of 30 immune proteins containing 10 pairs of immune checkpoint axes in 0.55 million of BM cells, an immune landscape of MM was mapped. Results. We identified an abnormality of immune cell composition by demonstrating a significant increase in activated CD4 T, CD8 T, CD8 + natural killer T-like and NK cells in MM BM. Our data suggest a correlation between MM cells and immune checkpoint phenotypes and expand the view of MM immune signatures. Specifically, several critical immune checkpoints, such as programmed cell death 1 (PD-1)/PD ligand 2, galectin-9/T-cell immunoglobulin mucin-3, and inducible T-cell costimulator (ICOS)/ICOS ligand, on both MM and immune effector cells and a number of activated PD-1 + CD8 T cells lacking CD28 were distinguished in MM patients. Conclusion. A clear interaction between MM cells and the surrounding immune cells was established, leading to immune checkpoint dysregulation. The analysis of the immune landscape enhances our understanding of the MM immunological milieu and proposes novel targets for improving immune checkpoint blockade-based MM immunotherapy.

INTRODUCTION
Multiple myeloma (MM) is a cancer of clonal plasma cells preferentially localised in the bone marrow (BM). The proliferation of MM cells, together with an MM cell-changed BM microenvironment, suppresses local and systemic immunity, eventually leading to an escape from immune surveillance. 1 Mechanisms involved in MM-induced immunosuppression include dysfunction of T and natural killer (NK) cells, 2 disruption of antigen presentation processes, 3 activation of immunosuppressive cells, 3,4 upregulation of inhibitory immune checkpoints 5,6 and release of immunosuppressive mediators. 7 Comprehensively uncovering the immune status in the BM microenvironment of MM patients will largely facilitate the understanding of the ongoing process of immunosuppression in MM progression and therefore promote the development of novel immunotherapeutic strategies.
Immunotherapy that involves stimulating and provoking a patients' own immune system against cancer has proven to be very encouraging as dramatic and durable anticancer responses are well documented in many cancer types. 8,9 Blocking inhibitory immune checkpoints on immune effector cells results in the reactivation of anticancer immunity. 10 Immune checkpoints contain a series of costimulatory and coinhibitory receptors or ligands expressed on T, NK or antigen-presenting cells and mainly function as switches of immune activation or suppression. 11 Under normal physiological conditions, immune checkpoints maintain self-tolerance and immune homeostasis, whereas malignant cells take advantage of these molecules to achieve immune evasion. 12 The most prominent immune checkpoint blocking strategies, such as targeting cytotoxic T lymphocyte-associated protein 4 (CTLA-4) and blocking the interaction between programmed cell death 1 (PD-1) and PD ligand 1 (PD-L1), are able to enlist and strengthen the immune system to attack cancer cells and have achieved clinical success in several cancer types, even in metastatic and chemoresistant cancer. 13,14 However, these immunotherapies are unable to control malignancy in a significant proportion of patients, largely because of the fact that inhibitory signals inducing the exhaustion and dysfunction of anticancer immune cells are not fully and sustainably blocked. 10,15 Indeed, as reported by a phase 1b clinical study, PD-1/PD-L1 axis-based immune checkpoint blockade failed to control MM progression, 16,17 suggesting that this checkpoint may not be the major mediator of failing anti-MM immunity. Besides PD-1 and CTLA-4, many other immune checkpoints have been discovered and are used for improved immune checkpoint-based immunotherapy. 18 However, immune checkpoint signals in the MM microenvironment have not been fully elucidated. The analysis of immune checkpoints will help us to better understand the mechanism of immune evasion of MM cells and would allow the development of potent strategies, focused on the checkpoint signals that are actually used by MM cells to evade the immune system.
The most commonly used technique for immune phenotyping, flow cytometry, suffers from the limited detection channels (generally < 15) and cumbersome compensation because of spectral overlap, making it difficult to simultaneously detect all immune checkpoint phenotypes. As a cutting-edge single-cell technology, current mass cytometry merging mass spectrometry with flow cytometry permits up to 50 metal isotope tags to be measured simultaneously on a single cell with minimal/no compensation. 19,20 Such high multiparametric detection provides an unprecedented opportunity for deep phenotyping of the tumor immune microenvironment at the single-cell level. For now, this powerful innovation has offered insights into the heterogeneity and complexity of biology and has been used to understand the complex processes in cellular development, 21 differentiation 22 and tumor immunology, [23][24][25] and to explore the potential immunotherapeutic targets. 23 In this study, 0.55 million BM cells from 10 MM patients and five healthy donors (HD) were analysed using mass cytometry to elucidate the phenotypic diversity and immune checkpoint signature in MM BM ecosystems. Our data reveal vast phenotypic heterogeneity among both malignant and immune cells, identify an abnormality of immune cell composition and suggest links between MM cells and immune checkpoint phenotypes. Through in-depth analyses of 10 pairs of immune checkpoint axes in 12 identified immune cell types at the single-cell level, a picture of the immune checkpoint interaction network that exists in the MM BM microenvironment of these patients was established. Several critical immune checkpoints were identified in the MM BM and may serve as novel targets for developing more potent and efficacious checkpoint blockade-based MM immunotherapeutic strategies.

In-depth immune checkpoint phenotyping of MM cells using mass cytometry
To map the immune checkpoint signatures in the BM microenvironment of MM patients, we implemented a clinical high-dimensional singlecell profiling study of freshly collected BM from newly diagnosed and untreated MM patients using mass cytometry. Ten MM BM samples and five healthy BM samples were included for a large-scale mass cytometry analysis (Figure 1a). We stained prebarcoded BM cells with 30 antibodies to simultaneously determine the expression of 30 markers used to define cell populations and immune checkpoint phenotypes at the single-cell level ( Figure 1b). As the loss of CD138 caused by the cold storage and processing frequently occurs, 26,27 cells with a CD38 ++ CD45 À/dim phenotype were defined as malignant MM cells ( Figure 1c). To comprehensively view the immune checkpoint profile of MM cells from all patients, we generated a single-cell viSNE map to visualise high-dimensional data in two dimensions. 28 (Figure 2c). Natural killer T (NKT) cells are identified with a CD3 + CD56 + phenotype in many studies. [30][31][32] However, only a small proportion of CD3 + CD56 + are CD1drestricted, which is a unique feature of invariant NKT (iNKT) cells. Thus, this population is frequently referred to as 'NKT-like'. 32 Here, we gated two CD3 + CD56 + cell subsets, namely NKTlike and CD8 + NKT-like cells, after excluding CD4, CD8 and double-negative (DN) T cells from all CD3 + cells. As shown by heatmap, the expression of surface markers in each population was identical to the phenotype of indicated immune lineages ( Figure 2d). After gating on viSNE map, the immune lineages in individual samples were analysed (Figure 2e), which revealed a heterogeneity across HD or MM patients. Although wide variation existed in the frequencies of each immune cell type in different individuals, several significant changes between HD and MM patients were detected. In the BM of MM patients, the proportion of CD4 T, CD8 T, CD16 + NK and CD8 + NKT-like cells in CD45 + immune cells was significantly increased along with the significant decrease in granulocytes, as compared to those in HD BM cells (Figure 2f). The average percentage of CD8 T cells increased from 7.77% in HD to 14.82% in MM and that of CD4 T cells rose from 9.49% to 15.36%. Importantly, CD8 + NKT-like cells only accounted for 0.92% of HD BM immune cells in average, whereas it increased to 4.86% in MM patients ( Figure 2f). iNKT cells have been shown to be associated with MM and are important for antitumor immunity. 33 We also examined the proportion of iNKT cells with the T-cell receptor Va24Ja18 antibody. We found that they constitute a minor fraction of BM T cells and there is no significant difference in their percentages between HD and MM patients (Supplementary figure 2a). Moreover, the MM burden was positively correlated with the frequency of CD8 T cells in MM patients and negatively correlated with the frequency of CD16 + NK cells with a trend close to significance (Supplementary figure 2b). We also compared the intensity of the expression of these checkpoints in the corresponding positive cells. The expression of CD28 was significantly stronger in CD28 + CD4 T, CD8 T and DNT cells of MM patients. In PD-1 + CD8 + NKT-like and CD8 T cells, the PD-1 expression was also significantly increased in MM patients. Many significant changes in the expression of immune checkpoints in CD4 T, CD8 T, NKT-like or CD8 + NKT-like cells were discovered (Figure 3g).

The immune checkpoint atlas of MM BM non-T cells
The frequencies of PD-1 + , PD-L2 + , CTLA-4 + , ICOS + , 4-1BBL + , OX40 + and Tim-3 + cells in granulocytes were significantly increased in MM patients, although most of them were < 10% (Figure 4a and b). Granulocytes accounted for the major provider of ICOSL as more than 85% of them express ICOSL in both HD and MM patients. The frequencies of PD-1 + and 4-1BB + cells in undefined (the rest of) CD45 + , Tim-3 + cells in DC, LAG-3 + cells in CD16 À NK cells, and PD-L2 + cells in CD16 + NK cells were also significantly increased in MM patients (Figure 4c-e and Supplementary figure 5a-c). Although significant differences in the percentages of immune checkpoint-positive cells were not detected in many cell types, the intensity of their expression in several immune cell populations was significantly altered in MM patients (Figure 4f and g).

Activation signature of T and NK cells in MM BM microenvironment
CD8 T and NK cells are major contributors to anticancer immunity and the main targets to be reinvigorated by immune checkpoint blockadebased immunotherapy. HLA-DR appears at the late stages of activated T and NK cells and has been widely used as an activation marker. [34][35][36] CD38 and HLA-DR are also primarily regarded as biomarkers for identifying activated T cells. 29,37 Here, the activation status of T-and NK cell subsets was systematically quantified using these markers. In the MM BM cells, significant increase in activated (HLA-DR + CD38 + ) cells was found in CD4 T, CD8 T, NKT-like and CD8 + NKT-like cells (Figure 5a). Specifically, the average frequency of activated cells in CD8 T cells was dramatically elevated from 11.66% in HD to 40.94% in MM patients (Figure 5b). Similarly, activated NK cells in   figure 6b and c). In addition, the expression of important immune checkpoints, including PD-1, CD28 and ICOS, was changed in activated (HLA-DR + ) CD4 and CD8 T cells (Figure 5g). Coexpression of CD39 and CD103 has been used to identify the tumor-specific CD8 + T cells in human tumors. 38,39 Here, we introduced these two markers to examine whether increased CD8 T cells in the BM are specific against MM cells. However, above 90% of CD8 T or activated CD8 T cells are CD39and CD103negative (Supplementary figure 6d), suggesting that bystander T cells instead of tumor-specific CD8 T cells are abundant in MM BM. To identify the immune checkpoint phenotypes in activated cells, we compared the frequencies of the immune marker-expressing cells in inactivated with In-depth and systematic analyses of the immune checkpoint profile of MM BM T cells As T cells are the primary anticancer contributor, we next systematically analysed the immune checkpoint phenotype of all possible exclusively and significantly changed T-cell clusters. From the viSNE containing all CD3 T cells, we observed a huge heterogeneity of the T-cell compartments, regarding the expression of immune modulatory proteins (Supplementary figure 7a). We next introduced spanning-tree progression analysis of density-normalised events (SPADE) analysis 40 to divide all T cells into 100 minor clusters (nodes) containing cells with similar phenotypes. On the SPADE tree, we were able to characterise the immune checkpoint phenotype of each cluster and clearly observe the differences in these clusters in each individual (Figure 6a and Supplementary figure 7b). Using cytoClusterR, the heterogeneity of immune checkpoint receptor signatures across 100 T-cell clusters from all 10 MM patients or five HD was obviously revealed on heatmaps (Figure 6b). Clusters 82, 92, 89, 68 and 42 were specifically presented in MM patients. In each cluster, different median expressions of immune checkpoint protein are summarised (Figure 6b). Among these 100 T-cell clusters, the frequencies of 42 clusters in MM patients were significantly different from those in HD (Figure 6c and Supplementary figure 7c). Twenty-eight clusters displayed an activated phenotype (HLA-DR + ) and were significantly increased in MM patients (Figure 6c), indicating that these T-cell clusters may play pivotal roles in remodelling the MM BM immune microenvironment. Among these 28 clusters, eight CD8 T-cell clusters, including clusters 37, 32, 39, 21, 73, 89, 68 and 42, were activated and PD-1 + , whereas all these clusters were deficient in CD28 expression, except cluster 89 (Figure 6d). In addition, MM burden was significantly correlated with the frequencies of clusters 32, 48, 76, 82, 92 and 96 in MM patients (Figure 6e), indicating that the changes in these Tcell clusters are MM cell-dependent.

Immune checkpoint network in the MM BM microenvironment
We summarised the top 3 cell types providing immune checkpoint-related receptors or ligands in MM patients (Figure 7a). Based on these main providers and the expression of immune checkpoint molecules in MM cells, a list and a network describing the interactions among MM and immune cells through immune checkpoints were established (Figure 7b and c). Considering the large heterogeneity among MM patients, we also built an immune checkpoint network for each MM patient (Supplementary figure 8).

DISCUSSION
The BM contains a complex environment and is filled with numerous kinds of immunoregulatory signal from both immune and non-immune cells.
In the MM BM microenvironment, non-immune cells, such as stromal cells, regulate immunosuppression through cell-to-cell contact and release extracellular vesicles and thus favor immune evasion of MM cells. 41 Immune checkpoints expressed on immune cells maintain the immune homeostasis, whereas MM cells enhance the suppression signal to escape from immune surveillance. Immune checkpoint blockade can break this malignant cell-induced inhibitory communication and thus lead to the reinvigoration of anticancer immunity. Success of immune checkpoint therapies largely relies on the targets responsible for cancer-induced immune suppression. To improve our understanding of the immune signature and immune checkpoint abnormalities in the MM BM microenvironment, we performed a high-dimensional single-cell analysis of the immune checkpoint molecules in healthy and MM BM samples. This high-quality data set identifies an unambiguous immune checkpoint network in the MM immunologic milieu of these patients (Figure 7b and c) and establishes a powerful new level of insights into MM checkpoint immunotherapy.
Mass cytometry has been recently used to identify T-cell heterogeneity and early alterations in resident T cells, and innate and myeloid cells in the BM of MM. 42,43 Kourelis et al. 42   results, they also found a very low level of CTLA-4 in both CD4 and CD8 T cells and that PD-1 is expressed by several T-cell clusters, but not by all T cells. All identified BM cell types, except myeloid DC, express very low level of PD-L1, further confirming the lack of PD-1/PD-L1 checkpoint signalling. The other recent study also analysed the BM T cells from 7 HD and 10 MM patients and the BM myeloid cells from 4 HD and 8 MM patients using mass cytometry. 43 They discovered greater terminal effector differentiation in memory T cells and an increased PD-L1 expression on myeloid cells from MM patients than healthy donors. However, detailed status of immune checkpoints, as well as the cell types providing checkpoint signals, has not been identified in these previous studies. Here, we devoted to systemically delineate the immune checkpoint signature of MM by measuring 10 pairs of immune checkpoint axes in freshly isolated BM samples from MM patients without treatment and our data would maximally reflect the real immune status of MM BM microenvironment.
Malignant cells offer a variety of immune checkpoint ligands to match receptors on immune cells and thus regulate anticancer immunity. With the successful application of PD-1/PD-L1 axis inhibitors in solid tumor immunotherapy, this blocking strategy has also become a focus of MM immunotherapy and plenty of clinical trials are conducted. 44 However, single-agent therapy with PD-1 inhibitors fails to induce significant clinical responses in a phase 1b study, 16 suggesting that PD-1 blockade alone is insufficient to reinvigorate a clinically meaningful anti-MM immunity. Discrepant results concerning PD-L1 expression on MM cells have been reported. 45 Several studies have confirmed the limited expression of PD-L1 on MM cells [46][47][48] ; in contrast, higher PD-L1 has been also found in MM cells than plasma cells from HD. 5,49 Our comprehensive data revealed a low frequency (< 12%) of PD-L1 expression in MM cells from all 10 MM patients. However, the expression of PD-L2, another ligand for PD-1, on MM cells was relatively higher than PD-L1. Anyhow, ligands of PD-1 were not widely expressed by MM cells, implicating the existence of other possible participants in inhibitory immunity. We validated here that several immune checkpoint ligands, including GAL9, ICOSL, HLA-DR, CD86, PD-L2 and 4-1BBL, were more generally presented on MM cells and these ligands are able to largely influence the immune response through binding to their receptors on immune effector cells.
A significant positive correlation between MM burden and GAL9 expression, together with the high frequency of GAL9 expression on MM cells, emphasises the possible contribution of this ligand to the MM immune microenvironment. In addition, Tim-3, a receptor of GAL9, 50 was expressed by activated CD8 T, NKT-like, DNT cells and DC in MM patients. Tim-3-GAL9 axis provides inhibitory immune signals to activated T cells, 51 and immunotherapy targeting Tim-3 and PD-1 pathways enables the reversion of T-cell exhaustion and restoration of antitumor immunity, 52 thus suggesting a possible use of this strategy to reconstruct anti-MM immunity.
ICOSL was also expressed by most of MM cells, and its receptor ICOS was increasingly detected in 20-40% of CD4 or CD8 T cells of MM patients. Being in line with this mechanism, a higher percentage of ICOS + cells in follicular helper T cells has been found in MM patients than healthy controls. 53 The ICOS/ ICOSL signal can mediate helper T-cell immunity and regulate effector T-cell differentiation. 54 In vitro, ICOS/ICOSL blockade significantly reduced the generation of MM cell-induced inhibitory CD4 + Treg cells, 55,56 and lenalidomide, a clinically approved anti-MM immunomodulatory drug, could inhibit ICOSL expression in MM cells 57 and enhance PD-1/ PD-L1 blockade-induced anticancer immunity in MM patients. 58 These evidences, together with our results, underline ICOS/ICOSL blockade as a possible enhancer for anti-MM immunotherapeutic strategies.
T and NK cells are at the forefront of anticancer immune responses, and quantitative and functional abnormalities in these cells' subsets have been well identified in the MM BM microenvironment. 2,59,60 The discovery of significant increases in CD4 T, CD8 T, CD16 + NK and CD8 + NKT-like cells in MM BM compared with HD BM confirms an abnormal immune cell composition induced by MM cells. Remarkably, these increased T or NK cells are activated in the MM samples, but with a suppressive phenotype as several inhibitory receptors, such as PD-1 and Tim-3, were increased. Because of the fact that CTLA-4, 4-1BB and LAG-3 were expressed only by very few CD8 T cells, targeting those checkpoints might be less effective.
Deep analysis of T-cell profiling identified several specifically activated CD8 T-cell clusters highly expressing PD-1 in MM patients, whereas most of them are deficient in CD28 expression, a critical Tcell costimulatory receptor that binds to B7 molecules, including CD80 and CD86. 61 The failure of PD-1 inhibitors in MM immunotherapy may result from the deficiency of CD28 in activated CD8 + T cells as substantial evidences have demonstrated that successful reinvigoration of exhausted CD8 + T cells by PD-1/PD-L1 blockade is dependent on CD28. 62,63 Most likely, once CD28 signalling is restored in these increased numbers of activated CD8 + T cells, strong anti-MM immunity will be achieved for controlling MM growth.
New targets or strategies are needed to increase the success of immune checkpoint-based immunotherapy for MM. By fine-grained analysis of the immune cells in the MM BM microenvironment, this study provides a detailed atlas of the infiltrating immune cells in MM, identifies immune checkpoints change that are unique to the MM immunologic milieu, and reveals distinct immune subsets that may be responsible for anti-MM immunosuppression. These data will be a valuable resource for future research to explore more efficient immunotherapy strategies tailored to restore anti-MM immunity through inhibition of immune checkpoints. The large individual heterogeneity in immune checkpoint networks among MM patients also emphasises the necessity of personalised strategies for a successful MM immunotherapy. Our findings demonstrating several potential immune checkpoint targets warrant further functional investigation into developing novel strategies for MM immunotherapy. In addition, non-immune cell components, such as stromal cells and extracellular vesicles, which also play an important role in regulating immunosuppression in the MM BM, also need to be taken into account in discovering novel targets for MM treatment in future.

Human specimens
Multiple myeloma BM samples were collected from MM patients undergoing BM biopsy for diagnosis, and healthy BM samples were obtained from donors undergoing BM biopsy for BM donation. Informed consents in accordance with the Declaration of Helsinki were obtained from all participants. All participants were recruited at the Third

Sample processing
Bone marrow samples were collected into sodium heparin tubes. To maximally maintain the immune profile, freshly isolated BM cells were directly fixed using an optimised and well-established fixing method with minimal effects on target epitope. 64 About 1-2 mL of BM samples was fixed with Fix I Buffer (Fluidigm, South San Francisco, CA, USA) for 10 min at RT, and red blood lysis buffer was used to fully remove the erythrocytes. Cells were then resuspended in cell staining buffer (CSB) and Dulbecco's phosphate-buffered saline, supplemented with 0.5% bovine serum albumin and 0.02% sodium azide, containing 10% dimethyl sulphoxide, and stored at À80°C until cell staining was performed.

Barcoding
To eliminate sample-specific staining variation, all samples were barcoded first and then stained, processed and acquired as one multiplexed sample. A total of 0.5 9 10 6 fixed cells from each samples were washed thrice with CSB and washed twice with 19 Barcode Perm Buffer (Fluidigm). These samples were then barcoded using a 20-Plex Pd Barcoding Kit (Fluidigm). Each sample was washed thrice with CSB after incubation with different barcodes for 30 min at RT, and all samples were combined together into one tube for antibody staining.

Antibody staining
Combined samples were washed once with CSB and incubated with Human Fc Receptor Binding Inhibitor Antibody (Thermo Fisher, Waltham, MA, USA) for 10 min at RT to lower non-specific binding. Anti-human ICOSL-biotin (BioLegend, San Diego, CA, USA) was added to the samples for incubation for another 30 min at RT. These cells were washed twice with CSB and stained with 29 metal isotopetagged antibodies and 1 metal-labelled antibody against biotin (Supplementary table 2

CyTOF data acquisition
Samples were washed twice with CSB and twice with ultrapure water. Immediately prior to data acquisition, the sample was resuspended in ultrapure water containing 15% EQ Four Element Calibration Beads (Fluidigm) and filtered through a 38-lm cell strainer. The sample was acquired on a Helios mass cytometer (Fluidigm) at an acquisition rate of < 500 events/s. Bead-based normalisation and debarcoding were completed using CyTOF software 6.7 (Fluidigm).

Data analysis
Individual debarcoded files were uploaded to an online single-cell analyser, Cytobank (Beckman Coulter, Brea, CA, USA). 65 Beads, debris and doublets were excluded from the events, and the single-cell data were subsequently used for high-dimensional analyses. Contour plots, viSNE, SPADE and heatmaps were implemented using Cytobank. The frequency of positive cells in each gated population was determined using FlowJo (FlowJo LLC, Ashland, OR, USA). Bar plot, violin plot and heatmap of correlationship were generated using the GraphPad Prism software (GraphPad, San Diego, CA, USA). The comparison of HD and MM SPADE node was implemented using the cytoClustR R package developed in Kordasti Lab from King's College London. SPSS 20.0 software (IBM, Armonk, NY, USA) was used for the Pearson correlation analyses.

Statistical analysis
The Mann-Whitney U-test was used to determine the statistical significance between the two groups. A paired ttest was performed on the frequencies of different cell subsets in individuals. Error bars represent mean AE standard error of mean (sem). A P-value < 0.05 was considered as statistically significant.