NK cells CD56^bright and CD56^dim subset cytokine loss and exhaustion is associated with impaired survival in myeloma

The importance of the immune microenvironment and functional cellular cytotoxicity in the control of multiple myeloma has become increasingly well recognized. Both chimeric antigen receptor T-cell products¹  and ex vivo modified natural killer (NK) cells²  have antimyeloma activity, whereas established pharmaceutical therapies, including immunomodulatory drugs such as lenalidomide and monoclonal antibodies such as daratumumab, function via both direct plasma cell–mediated toxicity and through their effects on cytotoxic immune cells.³ 

NK cells are innate lymphoid cells that share many phenotypic and functional features with CD8⁺ T lymphocytes. NK cell behavior is determined by a complex interplay between activating and inhibitory signals, including the loss of target cell major histocompatibility complex (MHC) class 1 expression and upregulation of stress ligands such as MHC class I chain-related protein and UL16 binding protein.

NK cells are classically divided into 2 subpopulations based on their expression of CD56. The NK 56^bright subset constitutes 10% and the NK 56^dim subset 90% of circulating NK cells.⁴  The NK CD56^bright subset is immature and predominantly secretes cytokines in response to interleukin-12 (IL-12) and IL-18. In contrast, the NK 56^dim subset is considered more mature and releases cytotoxic granules in addition to cytokines after encounters with NK receptor ligands. Healthy NK cells predominantly produce tumor necrosis factor α (TNFα) and interferon γ (IFNγ).⁵  The dominant cytokine produced depends on the nature of the activating stimulus received by the NK cells, with TNFα production requiring the engagement of fewer receptor classes compared with IFNγ production.⁴ 

Malignant plasma cells differ from many other transformed cells because they continue to express MHC class 1⁶  and therefore have a degree of protection against NK cell–mediated lysis. Nonetheless, NK cells have the potential to target malignant plasma cells, provided sufficient levels of NK cell–activating ligands are expressed by myeloma cells. Plasma cells express cell surface MICA, the ligand for the NK receptor NKG2D, although the level of expression decreases with progressive disease.⁷  The NK cell–activating receptors NKG2D,^8,9  DNAM1,¹⁰  and 2B4⁹  are reduced in myeloma, suggesting that the response to plasma cell ligands will be suboptimal, and indeed NK cells in myeloma exhibit reduced in vitro cytotoxicity.^11,12  Furthermore, NK cell expression of the immune checkpoint inhibitor programmed cell death protein 1 (PD1) is elevated, leading to impaired NK cytotoxicity and IFNγ production.¹³ 

We used mass cytometry to characterize NK cells in newly diagnosed multiple myeloma (NDMM) and show that features of NK cell exhaustion are present at the time of diagnosis, affecting both the NK CD56^bright and NK CD56^dim populations and that this can influence prognosis.

Bone marrow samples obtained from 18 NDMM patients and control samples from discarded femoral heads of healthy individuals undergoing elective surgery (n = 6) or noninvolved staging bone marrows from individuals with high-grade lymphoma (n = 3) were stored and maintained in the Barts Cancer Institute tissue bank.

All NDMM bone marrow samples were obtained at diagnosis and before the initiation of treatment. NDMM samples represented a range of immunoglobulin A (n = 7), immunoglobulin G (n = 7), and light chain (n = 4) disease. Median age of NDMM patients was 64 years (range, 41-90 years); control individuals had a median age of 67 years (range, 40-85 years).

All bone marrow samples underwent density gradient centrifugation, were cryopreserved, thawed, and tested in batches. Ethical approval was provided by the East London and the City Health Authority Local Research Ethics Committee. Written informed consent was obtained from all patients in accordance with the Declaration of Helsinki.

Metal conjugated monoclonal antibodies were obtained from Fluidigm (supplemental Table 1). The antibodies 2B4 (Cl.7), CD38 (HIT2), IL-2 (MQ1.17H2), IL-10 (Jes3-9D7), Ki67 (Ki67), PD1 (9EH.12.2H7), programmed cell death-ligand 1 (29E.2AS), and T-cell immunoglobulin and mucin domain 3 (TIM3) (F38-ZE2) were purchased from BioLegend and metal conjugated by using Fluidigm MaxparX8 conjugation kits as per manufacturer’s instructions.

Unselected bone marrow mononuclear cells (BMMCs) were thawed and rested overnight in complete culture medium before stimulation for 72 hours with plate-bound CD3 (OKT3; eBioscience) and suspension CD28 (CD28.2; eBioscience), resulting in the indirect, cytokine-driven stimulation of the entire BMMC compartment. Protein transport inhibitor (2 µL/mL; eBioscience) and metal tagged CD107a (2 µL/mL; Fluidigm) were added for the final 4 hours of stimulation.

Cryopreserved BMMCs were thawed and rested overnight in complete culture medium. For cell surface staining, cells were washed twice and incubated for 30 minutes with a pretitrated antibody mastermix. Cisplatin (2 µL of 10× dilution stock solution) (Fluidigm) was added for the final 5 minutes of incubation, and cells were washed twice. Cells were stained for intracellular and intranuclear expression according to the Fluidigm intranuclear staining protocol. Fully stained samples were fixed and stored for 12 to 72 hours at 4°C. Immediately before analysis, samples were incubated for 20 minutes with 1 µL of iridium intercalator (Fluidigm) at a 1000× dilution. A final 3 washes were undertaken before data acquisition.

Data were acquired by using a CyTOF2 mass cytometer (Fluidigm) with a Super Sampler (Victorian Airships) at a cell concentration of 2.5 × 10⁶/mL. Data were normalized against EQ bead standards using the Helios normalizer (version 6.7).

A cytotoxicity assay was performed as described in the supplemental Methods.

Data analysis was performed by using Cytobank. Live single-cell events were identified on the basis of Iridium 191⁺Iridium 193⁺cisplatin^–. NK cells were identified by using the phenotype CD45⁺CD3^–CD16^+/−CD56⁺. The CD56^dim subset was defined as CD16⁺CD56^low, and the CD56^bright subset was CD16^+/−CD56^high. Backgating on B-cell and plasma cell lineage markers was used to ensure that CD56⁺ plasma cells were not included within the NK gate. Statistical analysis was performed in GraphPad Prism version 9 (GraphPad Software). The t test was used to compare groups when data were parametric, and the Mann-Whitney U test was used for nonparametric data. viSNE analysis¹⁴  was used to further characterize subpopulations within the NK cell gate. The viSNE algorithm is based on the t-distributed stochastic neighbor embedding algorithm; it generates a two-dimensional plot of multi-dimensional data, with each cell positioned according to its similarity to the phenotype of surrounding cells.

Total NK cell numbers were comparable between NDMM and control samples (Figure 1A). A significant skew toward the CD56^bright population was observed in NDMM compared with control (CD56^bright NDMM mean, 24.08%; control mean, 11.34% [t test, P = .046]; CD56^dim NDMM mean, 58.97%; control mean, 80.22% [t test, P = .0213]) (Figure 1B). Importantly, a bimodal distribution was seen in myeloma, with some individuals having a pattern of distribution similar to that seen in the control samples.

Because this skew in subpopulations could be due to a relative expansion of the NK CD56^bright subset or loss of the CD56^dim population, we next examined expression of Ki67 as a measure of proliferative potential.

Expression of Ki67 across the total NK cell population in NDMM was one-half that seen in control samples (NDMM mean, 15.50%; control mean, 29.57%) (Figure 2). Ki67 expression was significantly lower in NDMM in the CD56^dim subset (NDMM mean, 11.12%; control mean, 29.51%; t test, P = .005), whereas no difference was seen in the CD56^bright subset (NDMM mean, 24.84%; control mean, 26.78%; t test, P = .84). This suggests that the relative expansion of the NK56^bright subset seen in NDMM is due to a relative failure of proliferation in the NK56^dim subset, and we hypothesized that this may indicate the development of exhaustion in the NK56^dim subset.

NK cells in the resting state were identified by using traditional Boolean gating, and a viSNE analysis was performed to further characterize the NK cell populations (Figure 3A). The CD56^bright and CD56^dim populations could be readily identified. Within these classical NK cell populations, further phenotypically and functionally distinct cell groups could be identified by mass cytometry analysis (Figure 3B).

Within control samples, 2 distinct CD56^bright populations were identified (Figure 3C). The first is an active population that is Ki67 positive, indicating proliferation, and coexpresses IL-10, transforming growth factor β (TGFβ), and a variable level of TNFα and IFNγ. Granzyme and perforin expression in this population was low, in keeping with the expected phenotype, but the activation markers HLA-DR and TIM3 were expressed, indicating that this is an activated population.

In contrast, the second control CD56^bright population was poorly proliferative, expressed TGFβ but not the classical NK cell cytokines, and was strongly positive for both granzyme and perforin. This indicates that although CD56^bright populations were predominantly cytokine producing, they did have cytotoxic capacity and that the cytotoxic population was phenotypically distinct from the cytokine-producing population.

In NDMM, there was a relative loss of the proliferative NK56^bright cytokine-producing population with preservation of the cytotoxic subset. Unlike the equivalent population in control samples, this population expressed both PD1 and TGFβ. This indicates that in myeloma, there was selective loss of the “true” CD56^bright population with expansion of a functionally impaired CD56^bright cytotoxic subset with features of pro-tumor immune dysfunction.

The CD56^dim region contains multiple distinct populations with varied phenotype (Figure 3D), which would not be apparent using traditional grating techniques. These are characterized by a highly proliferative region with minimal cytotoxic activity (population A); a poorly proliferative NKG2D-expressing region (population B); and a poorly proliferative region with moderate DNAM-1 expression (population C), a region coexpressing NKG2D and DNAM1, granzyme, and perforin in keeping with the classically described mature, cytotoxic CD56^dim population. In NDMM, this has additional expression of PD1 indicating NK cell exhaustion (population D), a region with globally poor marker expression (population E), and an activated and proliferative population with high expression of HLA-DR and DNAM-1 that coexpresses elevated levels of TGF-β and is expanded in NDMM (population F). It is interesting to note that a differential distribution of the NK cell–activating receptors NKG2D and DNAM1 was seen across the subpopulations, with coexpression apparent on population D, whereas prominent NKG2D expression was seen on population B and DNAM-1 expression on population F. This suggests that these markers may define different NK cell populations with distinct functional activity.

To further explore the functional capabilities of NK cells in NDMM, we performed stimulation studies to assess granzyme, perforin, and cytokine production capacity.

The expression of granzyme, perforin, and CD107a and the proportion of NK cells triple-positive for all 3 markers were assessed before and after stimulation of BMMCs. The proportion of triple-positive cells rose in both control (resting NK median, 0.26%; stimulated NK median, 9.986%; Mann-Whitney U test, P = .012) and NDMM (resting NK median, 1.582%; stimulated NK median, 8.537%; Mann-Whitney U test, P = .045). No differences in peak triple-positive cell numbers were observed (Figure 4A). This suggests that NK cells in NDMM retain the capacity to upregulate granzyme and perforin production when faced with appropriate stimuli. An in vitro functional cytotoxicity assay revealed impaired target cell killing in NDMM (supplemental Figure 1A).

Interestingly, both NK cell perforin (Pearson correlation r = 0.57; P = .025) and granzyme (Pearson correlation r = 0.55; P = .03) expression (Figure 4B) positively correlated with overall survival from time of diagnosis. This suggests that persistent NK cell cytotoxicity is a key aspect of the immunologic control of NDMM.

NK cells produce the antitumor cytokines IFNγ and TNFα and the pro-tumor cytokine IL-10. The distribution of multiple cytokine phenotypes was examined in both the resting state and the stimulated state. Loss of IFNγ was particularly noted (resting NDMM mean, 0.54%; control mean, 2% [t test, P = .03]; stimulated NDMM, 3.2%; control, 11.45% [t test, P = .13]) (Figure 5A). As expected, in the resting state, the majority of NK cells were triple-negative for TNFα, IL-10, and IFNγ (NDMM mean, 78.82%; control mean, 87.15%). After stimulation, control samples exhibited a 10-fold increase in IFNγ⁺TNFα⁺IL10^– cell population, whereas in NDMM, there was minimal increase in this population (Figure 5B). This is in keeping with the loss of the cytokine-producing subset observed in the viSNE analysis. An in vitro functional cytotoxicity assay exhibited impaired cytokine production in NDMM, and this was particularly pronounced for IFNγ (supplemental Figure 1B).

Expression of TIM3 was significantly lower on NDMM NK cells than those of control samples (NDMM mean, 3.01%; control mean, 9.34%; t test, P = .002). This was predominantly due to a loss of expression on the CD56^dim subset (NDMM mean, 3.23%; control mean, 9.62%; t test, P = .002) (Figure 6A).

TIM3 expression increases with NK cell maturation and is considered to be a marker of NK cell activation.¹⁵  As expected, TIM3 levels rose on control samples with decreased CD56 expression (NK56^bright, 7.764%; NK56^dim, 9.620%); however, they fell in NDMM (NK56^bright, 4.705%; NK56^dim, 3.228%) (Figure 6A). Loss of NK cell TIM3 expression in NDMM may therefore represent the development of an exhausted state. When patients with NDMM were stratified according to survival, a higher level of TIM3 was seen in those surviving >36 months (mean survival <36 months, 2.13%; mean survival >36 months, 3.47%; Mann-Whitney U test, P = .031) (Figure 6B), again suggesting that NK cell activation may be important for long-term myeloma survival.

Expression of PD1 on NDMM NK cells was more than twice than on controls (PD1 mean control, 7.5%; NDMM, 15.85%) and was increased on both control and NDMM NK CD56^dim cells compared with NK CD56^bright subsets (Figure 7). Upregulation of CD56^dim PD1 expression was more pronounced in NDMM (NDMM mean, 20.04%; control mean, 7.99%; t test, P = .039) populations. Increased expression of PD1 on NDMM CD56^dim populations is suggestive of NK cell exhaustion in this population.

Here we established that multiple defects in both CD56^dim and CD56^bright NK cell populations are present at the time of myeloma diagnosis and that there is a survival disadvantage associated with these features of NK cell derangement.

The CD56^bright defects result in selective loss of the classical cytokine-producing population with preservation of a functionally abnormal cytotoxic subset. The CD56^dim defects result in reduced proliferation, loss of IFNγ production, loss of NK cell activation, and elevated PD1 expression.

NK CD56^bright skewing has been reported in other malignancies. In non–small cell lung cancer, the predominant tumor infiltrating NK cell is CD56^bright, compared with CD56^dim being predominant in noncancerous lung tissue.¹⁶  Furthermore, CD56^dim tumor-infiltrating NK cells have defects in degranulation^16,17  and IFNγ production.¹⁷  In hematologic malignancies, Vari et al¹⁸  reported that NK cells in the peripheral blood of patients with Hodgkin lymphoma are shifted toward an NK CD56^bright subset. In Hodgkin lymphoma, however, it is these CD56^bright cells that have elevated expression of PD1. Unlike the work we present here, none of these previous studies has examined in detail the functional NK cell subpopulations within the CD56^bright and CD56^dim groups. It is important to remember that circulating NK cells may not reflect the functional changes seen within the tumor microenvironment, where NK cells are in direct contact with the malignant population, as will be the case with the bone marrow samples studies here.

By harnessing the deep phenotyping power of mass cytometry, we have extended the existing data on NK cells in myeloma by examining multiple parameters at the single-cell level and identifying key subpopulation abnormalities. This includes the selective loss of the CD56^bright cytokine-producing population, which would not be apparent using traditional analysis.

The role of tumor immune surveillance is often attributed to cytotoxic T cells, with the role of NK cells being overlooked. Our observation that features of exhaustion can be seen in NK CD56^dim populations, while those patients with superior long-term survival express multiple markers of NK cell activation at the time of myeloma diagnosis, highlights the crucial role that NK cells are playing in tumor surveillance in this disease.

These data have important implications for optimizing therapeutic strategies for NDMM. Traditionally, individuals with NDMM are offered treatment only once symptomatic disease has become apparent. Our data, however, clearly indicate that NK cell defects which may affect prognosis are already detectable at the time of myeloma diagnosis. Focusing on repairing these defects at an early disease time point may allow long-term myeloma control to be established, avoiding or delaying the need for more intensive therapeutic strategies and preventing end-organ damage. Possible therapeutic options would include the use of lenalidomide, which has been shown to restore IFNγ production in NK cells from individuals with chronic lymphocytic leukemia.¹⁹  Because this drug has an established use in myeloma, is a well-tolerated oral treatment, and has already been explored in the setting of smoldering myeloma, its role in restoring NK cell function in this setting should be examined.

The monoclonal antibody elotuzumab has a variety of immune stimulatory activities, including the activation of NK cells via targeting of SLAMF7. It has displayed antimyeloma activity when combined with an immunomodulatory drug and corticosteroids in the relapsed refractory setting^20,21  but has not yet been evaluated in earlier lines of treatment.

In summary, these data show that NK cells in NDMM exhibit multiple features of NK cell exhaustion that affect both the CD56^dim and CD56^bright subsets. Deleterious NK cell activation is associated with inferior long-term survival. NK cell subsets may serve as a useful therapeutic target at this early disease stage, and steps to optimize NK cell activation should be explored.

The authors thank the staff of the Barts Cancer Institute Tissue Bank and Flow Cytometry Core facility who assisted with this work.

This work was supported by research funding from The London Clinic Charity (ONAS1A2R) and from Barts Charity (MRD0203/G-001195, J.G.G. and F.S.).

Contribution: F.S. performed experiments and made the figures; F.S. and J.G.G. analyzed results; F.S., J.D.C., and J.G.G. designed the research and wrote the paper; and J.M. provided clinical data.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Frances Seymour, Department of Haematology, Leeds Cancer Centre, St James's Hospital, Leeds LS9 7TF, United Kingdom; e-mail: frances.seymour1@nhs.net.