Expression and function of immune ligand-receptor pairs in NK cells and cancer stem cells: therapeutic implications

Abstract

Background

The interplay between the immune system and cancer cells has come to the forefront of cancer therapeutics, with novel immune blockade inhibitors being approved for the treatment of an increasing list of cancers. However, the majority of cancer patients still display or develop resistance to these promising drugs. It is possible that cancer stem cells (CSCs) are contributing to this therapeutic resistance. Although CSCs usually represent a small percentage of the total number of cancer cells, they are endowed with the ability of self-renewal and to produce differentiated progeny. Additionally, they have shown the capacity to establish tumors after transplantation to animals, even in small numbers. CSCs have also been found to be resistant to various anti-cancer therapies, including chemotherapy, radiation therapy and, more recently, immunotherapy. This is true despite the sensitivity of CSCs to lysis in vitro by natural killer (NK) cells, the main effector cells of the innate immune system. In this paper the expression of ligands specific for NK cells on CSCs, the intracellular network responsible for the expression of the NK cytotoxicity receptors, and the status of activation of NK cells in the tumor micro-environment are reviewed. The aim of this review is to highlight potential strategies for overcoming CSC immune resistance, thereby enhancing the efficacy of current and future anti-cancer therapies.

Therapeutic implications

NK cell activation in the tumor micro-environment through drugs neutralizing inhibitory immune receptors, and combined with other drugs harnessing the potential of the adaptive immune system, could be the most effective approach for attacking both stem cell and non-stem cell cancer populations.

Introduction

Both elements of the immune system, the innate and the adaptive, cooperate at many levels with one another in order to provide optimal protective responses [1]. Natural killer (NK) cells play a central role in the innate immune system. Despite the fact that they do not possess T-cell receptors (TCRs) and do not depend on specific antigens for their cytotoxicity, NK cells are still guided and restricted in their actions by a multitude of receptors that define the final result of their interaction with their targets. In addition, they shape the adaptive immune response by secreting cytokines, such as interferon γ (IFNγ) and tumor necrosis factor α (TNFα) [2]. The reactivity of NK cells against infected or transformed cells depends on the net effects of stimulatory and inhibitory interactions of surface receptors. The ligands involved act on their targets in a manner very similar to cytotoxic T-lymphocytes (CTLs), the main effectors of the adaptive immune system.

Neoplastic cells are targets for NK cell cytotoxicity, and the anti-tumor effects of the innate immune system have previously been discussed in both leukemias and solid tumors [3]. In the case of progressive cancers, it is evident that both the innate and adaptive immune systems fail to curtail cancer growth, thus allowing the tumor cells to escape immune surveillance. Moreover, it appears that intra-tumor heterogeneity is at the root of the development of resistance to various oncologic treatments, such as chemotherapy and radiotherapy. This heterogeneity seems to be responsible for the escape from immune surveillance as well [4]. Resistance has been traced back to a subset of neoplastic cells, known as cancer stem cells (CSCs) [5]. These cells have characteristics allowing for self-renewal, and the propagation of their progeny is what gives rise to the bulk of tumors [6]. Experimentally, CSCs have been identified in most common cancers through the expression of different cell surface molecules. Examples include CD44 in breast, colorectal and bladder cancers, CD133 (prominin) in colorectal cancer and CD34 in leukemia [7, 8]. In some cases, the definition of stem cells also includes an absence of expression of specific additional markers, which increases their selection purity. This is e.g. seen in breast CSCs exhibiting negativity for the surface molecule CD24. In this paper the interactions of CSCs with NK cells through various ligand-receptor couples that either result in lysis or resistance to cytotoxicity are reviewed. The intra-cellular networks in CSCs that regulate the expression of NK ligands are also reviewed. Finally, the state of activation of NK cells in the tumor micro-environment is discussed as a basis for the in vivo resistance of tumors to the immune system.

Basic NK cell physiology

NK cells represent an organism’s first line of immune defence against infected or transformed cells. They kill target cells after contact and immune synapse formation through perforin-facilitated infusion of granzymes, which are serine proteases, thereby inducing apoptosis by cleavage of pro-caspases [9]. Additionally, NK cells produce cytokines that promote activation of other immune cells and contribute to the induction of adaptive immune system responses. Bone marrow precursor hematopoietic stem cells (HSCs) give rise to the progenitors that are destined to differentiate into both NK cells and T-cells. These progenitors differentiate into precursor NK cells under the influence of interleukin (IL)-2 and IL-15 signaling. Following this differentiation, they mature into NK cells that express both specialized NK stimulatory and inhibitory receptors. Notwithstanding these specialized receptors, NK cells also express receptors shared with other immune cells [9]. The killing machinery consisting of granzymes and perforin is acquired in parallel to the specialized receptors. Two main NK cell subsets exist, with different localization predominances and functional properties, exhibiting either CD56^bright/CD16^dim or CD56^dim/CD16^bright phenotypes [10]. The former phenotype is dominant within the circulation, while the latter predominates in tissues. The circulation-dominant NK cells (with the CD56^bright/CD16^dim phenotype) show a functional superiority in their capability to respond to soluble factors as compared to those with the CD56^dim/CD16^bright phenotype. However, since this latter phenotype predominates in tissues, it responds better to signals from surface receptors, and is more cytotoxic than its circulatory counterpart.

An important concept for the activation of the natural cytotoxicity of NK cells is that of the “missing self”. This principle states that NK cells are activated when their inhibitory receptors fail to ligate MHC I molecules on foreign, infected or transformed cells, allowing unopposed activity of stimulatory co-receptors [11]. It is currently clear, in view of the arming or licensing concept, that the “missing self” principle is more accurately described as a detection of a “lost self”, stemming from the fact that killer immunoglobulin-like receptor (KIR) presence and engagement is required for subsequent activation [12]. NK cells are instructed by inhibitory KIR receptors to attack cells that were previously expressing, but no longer express, MHC I molecules. However, “license to kill” is not provided against targets that do not express MHC I at the critical time of the individual NK cell clone development, or if that clone does not express any inhibitory receptors [13, 14]. In addition, licensing may be modified according to the environment in which NK cells evolve and, thus, it has been regarded as “revocable” [10]. This modification may have important implications in the tumor micro-environment, providing an optimal window of time for NK cell cytotoxicity to take place effectively. This opportunity would be at the time of neoplastic transformation, when tumor cells lose expression of MHC I (i.e., development of “missing self”). After tumor establishment, new NK cells incoming to the tumor micro-environment may have their killing license revoked. In addition to the “lost self” concept, the “induced stressed self” concept operates in transformed or infected cells, which may newly express ligands for stimulatory receptors such as NKG2D (MICA, MICB and ULBPs) that contribute to these cells becoming targets for NK attack, if the NK cells express the appropriate receptors [15].

Receptor-ligands of NK cytotoxicity

Activation of NK cells depends on their recognition of stimulatory receptors on the target cells and the absence of concomitant inhibitory receptor engagement that would interfere with activation. The balance of these stimulatory and inhibitory signals determines the outcome of the interaction of an NK cell with a target (Table 1). A brief overview of NK surface receptors for immune function is provided below.

Table 1.

Immune ligands for NK receptors with a potential role in cytotoxicity against CSCs. In the Type of ligand column: B, the ligand interacts with both stimulatory (S) and inhibitory (I) NK receptors as indicated in the last column; S, the ligand has only stimulatory interactions; I, the ligand has only inhibitory interactions

Ligand (alternative names)	Type of ligand	Chromosome	Complex partners	Expression/ comments	NK receptor
MHC I (HLA-A, HLA-B, HLA-C)	B	6p22.1, 6p21.33, 6p21.33	β2m/ antigen	Decreased expression in CSCs may be the main reason for increased NK susceptibility	KIR2/3DL (I), KIR2/3DS (S)
HLA-E	B	6p22.1	β2m/ restricted antigens	Expression in some CSCs may decrease NK susceptibility	CD94/ NKG2A (S), CD94/ NKG2C/E (I)
HLA-F	I	6p22.1	β2m/ restricted antigens		ILT2
HLA-G	B	6p22.1	β2m/ restricted antigens	Association with β2m modulates effect	KIR2DL4 (S), ILT2 (I)
MICA/B	S	6p21.33	–	Up-regulated in several types of CSC. Expression regulated by stem cell network	NKG2D
ULBP1–6	S	6q25.1	–	Up-regulated in several types of CSC	NKG2D
PVR (CD155)/ PVRL2 (CD112)	B	19q13.31 19q13.32	–	Expressed in various CSCs. Effect on NK cytotoxicity would depend on type of receptors expressed by TINKs	DNAM-1 (S), TIGIT (I), CD112R (I)
CD48	B	1q23.3	–	Expressed in MM	2B4 (B)
B7-H6	S	11p15.1	–	Expressed in cancers. Expression up-regulated by CSC network member c-Myc	NCR3
PD-L1 (B7-H1, CD274)	I	9p24.1	–	Expressed in CSCs from many common types of cancers	PD-1

Main stimulatory NK receptors

As mentioned above, in tissues NK cells may acquire the CD56^dim/CD16^bright phenotype, meaning that they up-regulate CD16, which is the receptor for the Fc region of IgG immunoglobulins, and which is alternatively called FcγRIII. Cells that express antigens recognized and coated by soluble antibodies are attacked by NK cells through CD16. CD16 is the only stimulatory NK receptor that has been described as being able to work autonomously without input from other receptors, and to effectuate cytotoxic granule polarization and degranulation in target cells [10]. In addition, CD16 represents a major co-operation point of the innate and adaptive immune systems, given that attack by NK cells depends on previous coating of target cells by antibodies produced by plasma cells. This process is termed antibody-dependent cell-mediated cytotoxicity (ADCC) [16, 17]. This process is also exploited in targeted monoclonal antibody treatment protocols. An example of this is treatment with trastuzumab, an anti-Her2 monoclonal antibody, of Her2 over-expressing and/or amplified breast and gastric cancers. The monoclonal antibody treatment-innate immunity synergistic effect results from direct cytotoxicity of the monoclonal antibody binding to its receptor (Her2) and subsequent activation of the ADCC. This activation is due to the ligation of CD16 receptors on NK cells to the Fc region of the monoclonal antibody-target cell complex [18, 19].

NKG2D (Natural Killer Group 2 member D, alternatively called CD314 or KLRK1 [Killer cell Lectin-like Receptor subfamily K member 1]) is a stimulatory receptor that recognizes the ligands MICA, MICB (MHC I Chain-related A and B) and ULBP1–6 (UL-16 Binding Proteins 1 to 6) (Table 1). These molecules are not expressed in normal cells, but are up-regulated in stressed and many malignant cells. Some of the ULBP ligands have soluble isoforms that are secreted from tumor cells and ligate NKG2D, resulting in its down-regulation from the surface and, thus, a decrease in NK cytotoxicity [20]. MICA is also shed from cancer cells, and its serum level has been proposed to have either a positive or a negative prognostic significance in various cancers [21, 22]. NKG2D, in common with all other stimulatory NK receptors besides CD16, requires ligation of other co-operating receptors such as NKp46 or 2B4 in order to complete target killing. Two other stimulatory killer cell lectin-like receptors are NKp80 (KLRF1, Killer cell Lectin-like Receptor subfamily F member 1) and NKp65 (KLRF2), which recognize C-type lectins AICL (Activation-Induced C-type Lectin) and KACL (Keratinocyte-Associated C-type Lectin), respectively [23].

The NCR family of stimulatory receptors has three members: NCR1 (also called NKp46 or CD335), NCR2 (also called NKp44 or CD336) and NCR3 (also called NKp30 or CD337). NCR1 and NCR3 are expressed in both resting and activated NK cells, while NCR2 is expressed only after activation [24]. All three NCRs recognize heparin and viral hemagglutinins. NCR3 also binds to a member of the B7 family, B7-H6 [25]. A carboxyterminal-truncated form of the lysine methyltransferase MLL5 (Mixed Lineage Leukemia 5), which is expressed on the cell surface of transformed cells, also acts as an activating receptor for NCR2 [26].

Receptors for HLA ligands

The KIR (Killer Immunoglobulin-like Receptor) family encompasses both stimulatory and inhibitory members [27]. KIR stimulatory receptors have short intra-cytoplasmic domains lacking activation motifs. They are called KIR2DS or KIR3DS, depending on whether the extracellular part has two or three Ig-like domains, respectively. The intracellular tails of stimulatory KIRs are able to interact with DAP12, an adaptor protein that contains an ITAM (Immunoreceptor Tyrosine-based Activation Motif) through which it transduces NK activation signals. Its ligands include HLA-A, HLA-B and HLA-C isoforms. The allotypes of HLA alleles that are expressed by an individual person influence the complement of KIR receptors expressed in that person’s NK cell subsets. The only KIR receptor with a long cytoplasmic tail that is stimulatory is KIR2DL4, which is expressed by all NK cells. This receptor uses a type Ib HLA molecule, HLA-G, as its ligand signaling for cytokine production [27].

Inhibitory receptors of the KIR family represent the main human inhibitory NK cell receptors [28]. Their ligands are HLA class I molecules that, if expressed on target cells, inhibit NK cytotoxicity. In contrast to the stimulatory KIRs, inhibitory KIRs possess longer intra-cytoplasmic tails that contain ITIMs (Immunoreceptor Tyrosine-based Inhibition Motifs) through which they transduce their inhibitory signals after ligand interaction. Thus, they are called KIR2DL or KIR3DL. Similar to their stimulatory counterparts, their ligands include specific HLA I allotypes. Due to the dependence of HLA I molecules on the invariable light chain β2m for their surface expression, cells (including embryonic stem cells not expressing β2m) have been found to be sensitive to NK cytotoxicity in vitro [29].

CD94 is a glycoprotein that forms stimulatory heterodimers with NKG2C (CD159c) and NKG2E (KLRC3, Killer cell Lectin-like Receptor C3) or inhibitory heterodimers with NKG2A (CD159a). Both types of heterodimers ligate HLA-E, an atypical HLA molecule, on target cells. NKG2A expression decreases as CD56^bright NK cells become CD56^dim, while KIR expression increases in parallel [10]. However, all CD94 heterodimers are dispensable for NK development and function, given that CD94 knock-down mice exhibit normal NK cell development and a normal responses to viral challenges [30].

ILT-2 (Immunoglobulin-like Transcript 2, also called LILRB1 or CD85j) and ILT-4 (also called LILRB2 or CD85d) are inhibitory immune receptors expressed by NK cells, and they ligate classic (HLA-A, -B and -C) and atypical (HLA-E and -G) HLA class I molecules on target cells [31]. NK cytotoxicity has been shown to be defective in triple negative breast cancer patients and this defect has been traced down to the expression of ILT-2 by NK cells [32]. Antibodies blocking ILT-2 were found to be able to restore NK cytotoxicity. Circulating NK cells from colorectal cancer patients have been shown to exhibit up-regulated ILT-2, as well as inhibitory NKG2A receptors, and down-regulated stimulatory receptors such as NKG2D, DNAM-1, NCR1 and NCR3 [33]. This combination of altered expression patterns results in NK hypo-responsiveness.

Receptors shared with other immune cells

2B4 (CD244 or SLAMF4) is a receptor of the SLAM (Signaling Lymphocyte Activating Molecule) family and ligates CD48 on target cells. Besides being expressed in NK cells, it is also expressed in subsets of T-cells, γδ T-cells, as well as monocytes and basophils. Furthermore, 2B4 can act as a promoter or inhibitor of lymphocyte activation depending on the robustness of the signal and the availability of the intracellular interacting protein SAP (SLAM-Associated Protein). In both situations, the limiting factor leading to inhibition is SAP availability [34]. In addition, at least in mice, two 2B4 isoforms do exist: one with a long cytoplasmic tail that is inhibitory, and another one with a short cytoplasmic tail that is stimulatory, reminiscent of the KIR inhibitory and stimulatory receptors discussed above [35]. Ligand CD48 is also expressed in NK cells, and it has been proposed that interaction of the 2B4/CD48 pair on NK cells coming in contact with each other prevents their mutual killing [36].

Several other receptors are shared between NK cells and T-cells, including DNAM-1, LAG3, TIM-3 and PD-1. DNAM-1 (also called CD226) is a stimulatory receptor expressed by both NK cells and cytotoxic T-cells [37]. Its ligands are molecules of the nectin family called PVR (Polyoma Virus Receptor, CD155) and PVRL2 (PVR-Like 2, CD112). These ligands may also ligate an alternative inhibitory receptor, TIGIT, which is also expressed in NK cells [38]. PD-1 is an inhibitory receptor of the CD28 family that is ligated by PD-L1 and PD-L2. Both these ligands are members of the B7 family. PD-1 is expressed in NK cells, which may contribute to the success of monoclonal antibody inhibitors in cancer therapy, given that these inhibitors may activate both NK cells and CTL cytotoxicity in the tumor micro-environment. PD-1 expression contributes to NK cell exhaustion and hypo-responsiveness to the stimulatory ligands NCR1, NCR3 and CD16, in a manner similar to CTL exhaustion and hypo-responsiveness to the respective stimulatory ligands [39]. In addition, it has been found that PD-L1 expression in NK cells of tumor-bearing mice interferes with the ability of dendritic cells to present tumor antigens to CD8+ T-cells, thus orchestrating a more global immune paralysis that spreads into the adaptive immune response [40]. Upon activation, NK cells express CTLA-4, the other clinically exploited inhibitory receptor of the CD28 family, possibly leading to their inhibition. This inhibition affects both NK cells and CD8+ T-cells in vivo, potentially contributing to the clinical effectiveness of anti-CTLA-4 monoclonal antibodies, which neutralize this effect [41, 42]. TIM-3 is an inhibitory receptor with several ligands, including galectin 9, CEACAM-1 and HMGB1. LAG3 (CD223) is an inhibitory co-receptor of the immunoglobulin family related to CD4 that binds MHC II molecules. It essentially interferes with the binding activation of these molecules by TCRs and CD3 [43]. If expressed by NK cells, it may interfere with the activation of CTLs, resulting in a decreased killing of MHC II-expressing tumor cells by these CTLs in the tumor micro-environment (Fig. 1).

Fig. 1.

The balance of engagement of stimulatory and inhibitory receptors in NK cells determines the outcome of an encounter with targets. The expression of these receptors depends on several factors, including stage of NK development and cytokine signaling. The ultimate result of an encounter of an activated NK cell with a target cancer cell also depends on the expression of NK ligands on the target. When stimulatory ligands (sticks on the cancer cell on the left) predominate, target lysis will be favored. When inhibitory signals (spikes on the cancer cell on the right) predominate, the encounter will not result in lysis

In summary, cooperation of multiple signals is required for NK cell anti-tumor activity to occur. In a mouse model of c-myc-induced lymphoma, for example, a priming signal is provided by the down-regulation of MHC I expression to avoid engagement of inhibitory KIR receptors. A triggering signal is provided by expression of NKG2D ligands [44]. In addition, in the tumor micro-environment, immunosuppressive signals such as those generated by TGF-β, IDO (indoleamine 2,3-dioxygenase) and immune suppressive cells such as Tregs, need to be overruled in order for an immune attack to proceed, even against a sensitive target [45, 46]. This process will be discussed in more detail below .

NK cell-mediated cytotoxic effects on normal stem cells and CSCs

Accumulating data show that CSCs are particularly sensitive in vitro to activated NK cells, in many occasions more so than their non-stem cell counterparts. This property is concomitantly shared with their normal stem cell counterparts of various origins. Normal mesenchymal stem cells and dental pulp stem cells, both able to differentiate into osteocyte, chondrocyte and adipocyte lineages, have been found to be particularly sensitive to lysis by IL-2 activated allogeneic NK cells, and to a lesser extend to lysis by non-activated NK cells [47]. This lysis could be inhibited by co-culture of NK cells with monocytes or by blocking CD16 with a monoclonal antibody (Fig. 2). In addition, stem cell differentiation has been found to result in a decreased susceptibility to NK cell attack [48]. The same effect may hold true for CSCs that become NK resistant when differentiated. It has been found that direct NK cytotoxicity is reduced when NK cells come into contact with sensitive cell targets, but not with resistant targets. This feedback mechanism has been proposed to play a role in NK elimination of a sub-set of CSCs, while the remaining resistant CSCs may differentiate and consolidate their resistance to NK lysis through their progeny [49].

Fig. 2.

Mesenchymal stem cells (MSCs) may be lysed by activated NK cells (solid arrow). Differentiation turns MSCs into less optimal NK cell targets (dashed arrow). NK cytotoxicity against MSCs is decreased after co-culture of NK cells with monocytes or incubation with antibodies blocking the CD16 receptor

Human induced pluripotent stem cells (hiPSCs) produced from hair keratinocytes of healthy donors by lentiviral transfection with the stem cell transcription factors Oct4, Sox2, Klf4 and c-Myc, have been found to be susceptible to IL-2 activated NK cell lysis [50]. Allogeneic NK cells were more effective than autologous NK cells, while non-activated NK cells were not effective against hiPSCs, in contrast to their effectiveness against K562 cells. These latter cells, which are of chronic myeloid leukemia origin, are a well-known NK cell target and have been historically used in NK cell cytotoxicity assays as a positive control [51]. hiPSCs have been shown to express ligands for the activating NK receptors NKG2D (MICA and MICB) and DNAM-1 (nectins PVR and PVRL2), as well as low levels of MHC I [50].

Mouse adult germ line stem cells are targets of NK cells and express RAE-1 family molecules that are ligands for NK NKG2D receptors [52]. DNAM-1 ligands (PVR, PVRL2) are also expressed but at lower levels, while ligands for NCR1 and 2B4 are mostly absent. Both mouse embryonic stem cells (ESCs) and cardiomyocytes derived from them express low levels of MHC I molecules, but the former are much better targets than the latter for syngeneic NK cell-mediated lysis [53]. This differential lysis was traced back to the expression of NKG2D ligands and the adhesion molecule ICAM-1 by stem cells, while differentiation to cardiomyocytes leads to loss of expression of both. NKG2D ligands expressed in mouse ESCs were similarly found to be down-regulated upon differentiation to neuronal cells [54]. Upon transplantation to immunodeficient mice or rats, both differentiated and undifferentiated cells were found to form teratomas despite the fact that neither type is teratogenic in immunocompetent hosts, which argues for an important role of the immune system in suppressing teratoma formation. When treated with cyclosporine A, which suppresses T-cell activity but not NK cells, differentiated cells could form teratomas, while ESCs could not. This suggests that NK cells are important for suppression of tumorigenicity by NKG2D ligand-expressing ESCs [54]. Interestingly, in another study it was found that when strains of mice lacking B- and T-cells but retaining NK cells were transplanted with adult mouse multipotent germ line stem cells and treated with cyclosporine A, they displayed teratoma formation. This implies a direct effect of the drug on teratoma promotion independent of the immune system or, alternatively, that ESCs are resistant to NK cells in this model [55]. In addition, this study confirmed that activation of host NK cells did occur despite the cyclosporine A treatment, and that the resistance was actually induced from inflammation related to the transplantation operation rather than from the transplanted cells per se. Overall, normal mouse and human embryonic stem cells are quite sensitive in vitro to cytokine-activated NK cells, as they express activating ligands for the various NK receptors. Moreover, these data underscore the importance of the in vivo micro-environment of stem cell niches for protection against NK cell attacks, which includes both normal stem cells and CSCs [56].

The ability of NK cells to specifically target CSCs has now been documented in several studies (Table 2). CD44+/CD24- breast cancer stem cells and CD133+ colorectal cancer stem cells have been found to be more sensitive to IL-2 and IL-15-activated NK cells in an in vitro assay than their non-CSC counterparts [57]. It was also found that breast CSCs express the NKG2D ligands MICA, ULBP1 and ULBP2, and that blockade of NKG2D with a monoclonal antibody reduces the cytotoxicity of NK cells against CSCs more effectively than blockade with monoclonal antibodies directed against NCR1 (NKp46) and NCR2 (NKp44). The colorectal cancer cell line KM12C has been found to contain a smaller percentage of cells expressing CSC markers and the pluripotency network transcription factors Sox2, Nanog and Oct4 than the highly metastatic cell lines KM12L4A and KM12SM [58]. A higher percentage of cells of these highly metastatic cell lines expresses the NKG2D ligands MICA and ULBP2, and they have been found to be more susceptible to lysis by the NK cell line NK92 in vitro. In another study, colorectal cancer initiating cells that were expanded from primary tumors were found to be more sensitive in vitro to allogeneic NK lysis than their bulk tumor cell counterparts [59]. In this case, monoclonal antibody blockade experiments revealed a most important role in cytotoxicity of the NCRs NKp30, NKp44 and NKp46, although the NKG2D and DNAM-1 receptors were also found to play a role. Melanoma cell lines and patient-derived cells were found to be heterogeneous in expression of the stem cell marker CD133. However, when these cells were sorted into CD133+ and CD133- fractions in vitro, they were found to be equally susceptible to lysis by cytokine activated NK cells [60]. Various NK cell activating ligands were found to be expressed in subsets of melanoma cells, but the most consistently expressed stimulatory ligands were the nectin family members PVR and PVRL2.

Table 2.

Studies examining NK cell cytotoxicity to target CSCs and main ligands involved. Question marks denote that the ligand(s) involved were not examined in the respective studies. For further details see text

Reference	Type of target CSCs	Ligands	Comments
[56]	Breast, colon	MICA, ULBP1, ULBP2	Blockade of NKG2D more effective than NCR1 blockade
[57]	Colorectal	MICA, ULBP2	Stem-like cell lines more susceptible than parental cell line
[58]	Colorectal	Ligands for NCRs, NKG2D and DNAM-1	NCRs more important than NKG2D and DNAM-1
[59]	Melanoma	PVR and PVRL2	CSC-like cells equally sensitive to bulk cells
[60]	Uveal melanoma	?	miRs involved
[61]	Breast, pancreatic, Ewing’s sarcoma, glioblastoma	Ligand(s) for NKG2D	In vivo study in mice
[62]	Bladder	MICA/B and ULBP1, less PVR and PVRL2	Both CSCs and bulk cells highly sensitivity to donor NK cells, less so to autologous NK cells
[63]	Leukemia	PVR and PVRL2	Heterogeneous expression of DNAM-1 ligands in different leukemia cell lines
[64]	Glioblastoma cell line	?	Stem-like cells more sensitive
[65]	Primary glioblastoma	Several ligands were expressed on NK cells used	Differentiation decreased sensitivity to NK cell killing
[66]	Glioblastoma cell line	PD-1 were expressed in NK cells (PD-L1?)	Inhibition of PD-1 with monoclonal antibody enhanced NK cell cytotoxicity
[67]	Primary oral squamous carcinoma	Absence of PD-L1	Stem cells more sensitive to both activated and non-activated NK cells
[68]	Primary breast cancer	MICA and MICB down-regulated through miR20a	Stem-like cells less sensitive to autologous and allogeneic NK cells and more prone to metastasis formation

NK cytotoxicity may also be regulated by microRNAs (miRs), as observed in both uveal melanoma cell lines and stem cells [69, 70]. A differential NK cytotoxicity of uveal melanoma cell lines was found to be associated with the expression of different miRs, such as miR-146a and miR-155, both of which are secreted by these target cells. miR-155 was found to be produced and secreted by an aggressive uveal melanoma cell line, but not by a less aggressive one. Blockade of miR-155 production increased the sensitivity of uveal melanoma stem cells to NK cell lysis. miR-155 is involved in regulation of the TGF-β pathway, which is one of the major players in stemness and differentiation, as well as the process of epithelial to mesenchymal transition, EMT [61, 62].

Subsets of cells with a CSC phenotype and functional properties from various cancer cell lines (breast, pancreas, Ewing’s sarcoma, glioblastoma) were found to exhibit a higher NK sensitivity than subsets without stem cell properties, when assayed as xenografts in immunosuppressed mouse models [63]. In this latter study, the most important NK receptor that was shown to mediate cytotoxicity against CSCs was NKG2D. Comparison of NK ligand expression levels in bladder carcinoma parental cell lines and sphere-forming derivatives revealed that both MICA/B and ULBP1 were up-regulated in sphere forming stem cell-like cells. In some cases, this up-regulation also included PVR and PVRL2 [64]. In contrast, NK cells derived from healthy donors did not show any enhanced cytotoxicity against sphere forming cells in in vitro assays, despite their more robust expression of ligands. This finding is likely due to the high sensitivity of both the unsorted and sphere forming cells in these experiments. Interestingly, it was found that NK cells derived from urothelial carcinoma patients acted as less robust killers, especially of sphere forming urothelial cells. This finding suggests that previous exposure to the tumor environment or the presence of soluble factors interferes with a future recognition of similar carcinoma cells.

Various leukemia cell lines express the DNAM-1 ligands PVR and PVRL2 in varying degrees [64]. K562 cells, for example, express high levels of PVR and PVRL2, although with significant heterogeneity and with subsets of cells expressing significantly lower levels than others. Other leukemia cell lines such as PL-21, THP-1 and MV-411 were found to express low levels of DNAM-1 ligands, whereas a subset of the cells contradictorily showed a somewhat higher expression, especially of PVR [65]. This heterogeneity in expression implies that different clones exist with varying DNAM-1 expression levels, and that these clones may be related to the presence of stem cell subsets. Functionally, clones with a higher expression of DNAM-1 ligands were indeed found to be lysed by NK cells, while their low level-expressing counterparts were resistant.

Human glioblastoma U87 cells growing in stem-cell neurosphere-promoting conditions are more sensitive to ex vivo cytokine-expanded NK cell lysis than their counterparts growing under serum-containing culture conditions [66]. Glioblastoma (GBM) cells obtained from patients and growing in stem cell medium were found to be susceptible to both allogeneic and autologous cytokine-activated NK cells in vitro, but not to freshly isolated NK cells from the same individuals [67]. Autologous NK cells from patients and allogeneic NK cells from healthy donors express similar levels of the activating receptors CD16, NCR1, NCR2, NCR3, DNAM-1 and NKG2D, as well as the inhibitory KIR receptors and NKG2A after incubation with IL-2. When the GBM cells were grown under serum-containing culture conditions, they underwent a partial differentiation and became NK resistant [67]. Using a mouse GBM cell model of CSCs derived from the GBM cell line GL261, Huang et al. [68] found that treatment of mice bearing these CSCs with murine NK cells showed a delay in tumor progression and a prolonged survival. In addition, blockade of the PD-1 inhibitory receptor on NK cells further increased their anti-tumor efficacy.

Primary oral squamous carcinoma cells with stem cell characteristics (CD133+ and CD44^bright) were found to be devoid of PD-L1 (B7-H1) and EGFR expression when compared to cells from the same tumors without stem cell properties [71]. Transcription factor NF-κB activity was also found to be suppressed in the stem-like cancer cells compared to the non-stem cells. Additionally, an associated increase was noted in the induction of interferon gamma (IFNγ) and a reduced secretion of the immunosuppressive cytokine IL-6 when stem-like tumor cells were co-cultured with NK cells. The stem-like cells were also more sensitive to lysis by IL-2 activated NK cells than their non-stem cell counterparts. A differential sensitivity to lysis by non-activated NK cells, but at significant lower levels, was also observed [71].

Somewhat contrasting results were obtained in a study of a series of 591 breast cancer patients evaluated by immunohistochemistry (IHC) and ex vivo cytotoxicity assays. The patients were divided into two groups based on higher or lower expression of ALDH, a stem cell marker [72]. In both groups CD56+ NK cells infiltrating the tumors were noted. In the low ALDH group a higher CD56+ cell infiltration was found to be associated with a better overall and metastasis-free survival. In the high ALDH group, no such effects of CD56+ infiltration were observed. In ex vivo studies, tumor cells from patients with a high ALDH expression showed a lower expression of cytotoxic NK ligands, and these cells were more resistant to autologous and allogeneic NK cells in cytotoxicity assays than tumor cells from breast cancer patients with a low ALDH expression [72]. Although the patients had either higher or lower CSC fractions as measured by ALDH, in these experiments the targets cells were unfractionated and, therefore, both expression levels and cytotoxicity susceptibilities may have been influenced by the presence of non-CSC fractions. In addition, CD56+ NK cells may not be the most cytotoxic effectors, given that, as discussed above, the main cytotoxicity effect resides in the CD16^bright/CD56^dim subset of NK cells. In another report, the floating fraction of a tumor cell line derived from a patient with an unknown primary cancer, expressing a CSC surface marker profile (CD44^high/CD24^low) and able to produce tumors when transplanted in NOD/SCID mice, was found to be more resistant to allogeneic NK cells in vitro than the fraction of the same cell line growing as adherent layer and having the reverse (CD44^low/CD24^high) phenotype [73]. NK resistance in the floating CSC-like population was observed despite the fact that these cells exhibited a lower expression of MHC I molecules. Activating ligands for NK receptors were, however, not examined in this study. Therefore, it remains unclear whether either the CSCs from this patient’s cell line happened to not express these ligands or whether down-regulation of MHC I in the CSCs, compared to the bulk of cells, resulted in levels below the requirements needed for NK education to attack.

The overall picture that emerges from these studies is that CSCs are susceptible to NK cell mediated lysis in vitro and in vivo, especially if the latter are cytokine-activated. This susceptibility is achieved mainly via ligation of stimulatory ligands expressed by CSCs. The next section provides additional information on this expression and on pathways involved in NK ligand regulation, while in the subsequent section the receptor status of NK cells in patients will be discussed. In addition, explanations will be evaluated on why NK cytotoxicity against CSCs appears to be less robust within the tumor micro-environment than in vitro.

Pathways regulating NK ligand expression in CSCs

Expression of classic MHC class I molecules in CSCs is important for both CTL and NK cell cytotoxicity, as they are the main signal for CTL cytotoxicity after recognition by a complementary TCR. Additionally, they may provide inhibitory or stimulatory signals for KIRs in NK cells. MHC I molecules are expressed on the cell surface as a trimeric complex with β2-microglobulin (β2m) and different antigens. Thus, their presence is affected by the availability of β2m and the production of antigen peptides for presentation [74]. Antigen production and presentation require a cellular machinery that includes the proteasome, endoplasmic reticulum lumen transfer proteins and up-loading facilitators. Any derangement of these components in CSCs may influence MHC I abundance at the cell surface (Fig. 3). It has been found, for example, that proteasome function may be decreased in CSCs compared to non-stem cancer cells, thereby contributing to a decreased MHC I surface expression [75].

Fig. 3.

Expression of HLA Ia and HLA Ib molecules on target cell surfaces is influenced by the pluripotency network. β2-microglobulin (β2m) and the antigen production machinery also play important roles in HLA expression and NK receptor engagement. Proteasome activity is often down-regulated in CSCs

Expression of the non-classic HLA Ib molecules HLA-E and HLA-G has been observed in a subset of cancers and may have prognostic implications. Specifically, in cases with down-regulated classic HLA molecules, cells expressing non-classic HLA molecules may be able to inhibit NK cytotoxicity by engaging CD94/NKG2A, ILT2 and ILT4 [76]. HLA-E has been shown to be expressed at higher levels in renal carcinoma cells than in normal kidney tissues [77]. Its expression is induced by IFNγ and leads to decreased NK infiltration. Thus, it is conceivable that activated NK cells impede their own tumor infiltration in an auto-regulatory manner by inducing HLA-E expression through IFNγ secretion. HLA-E expression has been found to be associated with decreased immune cell infiltration and poorer outcomes in colorectal cancer patients [78]. Both HLA-E and HLA-G have been found to be up-regulated together with the CSC transcription factors Oct4 and Nanog in colorectal cancer cells compared to normal colonic tissues [79] (Fig. 3). Moreover, it has been reported that β2m expression is relevant for NK cytotoxicity involving HLA-G interactions. While HLA-G in complex with β2m engages the inhibitory receptors ILT2 and ILT4, when expressed alone it engages KIR2DL4, which is ubiquitously present on NK cells and promotes cytokine production and cytotoxicity [80] (Fig. 3). HLA-E is expressed in GBM CSCs and plays a role in inhibiting NK mediated lysis of these cells, as confirmed by blockade of its expression through siRNA [81]. It is conceivable, though, that this lysis may depend on the expression of stimulatory CD94/NKG2C versus inhibitory CD94/NKG2A by NK cells in vivo. HLA-E expression may favor lysis if there is a preponderance of CD94/NKG2C in incoming activated NK cells (Fig. 3).

As discussed in the previous section, ligands for the NKG2D activating NK receptors MICA, MICB and ULBPs (mainly ULBP1 and ULBP2), have been found to be expressed in CSCs from diverse carcinomas and may play a role in mediating NK cytotoxicity [57, 58, 64]. MICA and MICB, as well as ULBP2 and ULBP3, are up-regulated during epithelial to mesenchymal transition (EMT), which promotes the mobility and metastatic potential of cancer cells [82–84]. EMT is associated with stem cell plasticity [85] and, thus, the two processes may have an in-built propensity for promoting the expression of NKG2D ligands. Interestingly, it has been found that MICA may serve as a prognostic factor in colorectal cancer based on a microarray study in which patients with a MICA expression level above the median intensity of the series had a better disease-specific survival than patients with a MICA expression level below the median [86]. This differential survival was mainly due to a better survival of patients with stage III tumors, as patients with other stages exhibited no significantly improved survival rates when MICA was expressed at high levels in their tumors [86]. It has been reported that the colorectal cancer cell line HT29 is a weaker target of NK cells in vitro than other colorectal cell lines, such as SW620 and Colo320, which trigger a stronger IFNγ production and NK de-granulation [87]. This difference is partially due to a weaker expression of MICA and MICB in HT29 cells. The transcription factor STAT3 is active in HT29 cells and interacts directly with binding sites in the MICA and MICB gene promoters, leading to transcription suppression [87]. Given that STAT3 is also active in CSCs that express MICA and MICB, it is probable that in some cases, and depending on the cellular context, either STAT3 acts as an activator of MICA/B gene promoters or, alternatively, that other transcriptions factors override its suppressive effects.

Additional cancer-activated pathways and transcription factors may act as inducers of NK ligands. The transcription factor E2F has been found to act as a promoter of neoplastic cell cycling and, concomitantly, as an inducer of the NKG2D ligand Rae-1 in mouse fibroblasts [88]. The DNA damage response kinases ATM and ATR were also found to be involved in Rae-1 induction, independent of p53 status. In another study, induction of the NKG2D ligands Rae-1α and Rae-1β, as well as ULB1, 2, and 3, was observed in Ras-transformed mouse and human breast cancer and embryonic kidney cells, respectively [89]. In human cells, NKG2D ligand induction was found to be dependent on PI3K kinase activation downstream of Ras, with a lesser contribution of the MAPK/MEK cascade. The PVR and PVRL2 ligands are involved in NK-mediated lysis of various CSCs, such as those from colorectal cancer, melanoma, bladder carcinoma and GBM, as discussed before [59, 60, 64, 67]. These studies imply that activated NK cells use the activating receptor DNAM-1 instead of TIGIT, which is the inhibitory receptor for the same ligands [90]. It remains to be investigated, however, whether tumor infiltrating NK cells express either one of these receptors for PVR and PVRL2. B7-H6 is a B7 family member and a ligand for NCR3 (NKp30). It is absent in most normal tissues, but its expression has been found to be induced in several cancers and to promote NK cell recognition and cytotoxicity [25]. Although no data exist specifically for B7-H6 expression in CSCs, its induction has been deduced from the action of the oncogene c-Myc, a well-established member of the pluripotency network [91]. Expression of CD48, the ligand for 2B4 (CD244), is observed in most multiple myeloma cells. A monoclonal antibody that blocks this ligand has been found to be effective in inhibiting tumor growth in mice through a complement-dependent and antibody-dependent cytotoxicity process, although a direct role on NK activation was not assessed [92]. CD48 is expressed in normal hematopoietic progenitors of mouse fetal and adult origin, but not in hematopoietic stem cells [93]. Whether CD48 is expressed by CSCs of hematopoietic or solid tumors remains to be investigated. PD-L1 is a purely inhibitory member of the B7 family and is expressed in several types of CSC [94]. Colorectal cancer cells, for example, expressing the stem cell marker CD133, as well as the pluripotency transcription factors Oct4 and Sox2, have been found to be positive for PD-L1 [95]. Breast cancer and GBM CSCs expressing the stem cell markers CD44 and CD133, respectively, also co-express PD-L1 [96, 97]. Induction of PD-L1 has been shown to be the result of several cancer-associated effector pathways, including the Ras/Raf/Erk, the PI3K/Akt, and the NF-κB pathways, which may explain its common expression in cancer [98]. Despite their PD-L1 expression, the sensitivity of CSCs to NK cells observed in vitro may imply that the inhibitory action of this ligand can be overruled by activating ligand-receptor pairs, or that robustly activated NK cells in vitro express a low level of the PD-1 receptor.

Expression of receptors on NK cells infiltrating the tumor micro-environment

Tumor Infiltrating NK cells (TINKs) may modify their cytotoxicity receptors after coming in contact with tumor cells and other cellular and soluble elements in the tumor micro-environment. The state of responsiveness of TINKs is equally important for the result of their encounter with the tumor as the expression of appropriate ligands in the target cells, which can be modified by cytokines such as TGFβ [99]. CD25+ Treg cells are important in suppressing NK cell activity in the tumor micro-environment through modulating the expression of stimulatory and inhibitory receptors. This suppression by CD25+ Treg cells is quite similar to the suppression of cytotoxic CD8+ T-cells [100]. KIR receptor expression by TINKs in patients with non-small cell lung cancer (NSCLC) has been found to be associated with a significantly decreased overall survival compared to patients with TINKs not expressing KIRs (mean survival of ~4 weeks versus 40 weeks in KIR non-expressors) [101]. Interestingly, immunohistochemistry was used for KIR detection in this study and KIRs were also found to be expressed in a subset of tumor cells. Furthermore, an association was found with a poor survival for patients in this cohort compared to patients without KIR expression in their tumor cells. NK cells from ovarian carcinoma effusions have been found to express lower levels of the stimulatory receptors DNAM-1, 2B4 and CD16 than autologous circulating NK cells [102]. This low expression was due to down-regulation of receptors after interaction with tumor cells, as was observed after co-incubation of NK cells with tumor cells expressing PVR, leading to DNAM-1 down-regulation. It has also been reported that the melanoma tumor micro-environment affects TINKs which, after contact with tumor cells, down-regulate several stimulatory receptors such as NCR1, NCR2 and NKG2D [103]. Other types of tumors have also been found to exhibit a low percentage of infiltration by NK cells [104]. In melanoma this has even been observed in tumors that exhibit a high infiltration of other types of immune cells [105]. NK cells infiltrating tumors may trigger their own functional neutralization through secretion of IFNγ, which induces both MHC I up-regulation in tumor cells and production of the immune suppressing mediators indoleamine 2,3-dioxygenase (IDO) and prostaglandin E2 (PGE₂) [106]. Similar results were obtained from tumor cells of a patient with NSCLC which displayed MHC I expression down-regulation due to absence of β2m expression [107]. Autologous NK cells from this patient were found to be unable to lyse tumor cells, despite the absence of MHC I. This inability was due to the down-regulation of NCRs and NKG2D. In contrast, it was found that allogeneic NK cells with a normal stimulatory receptor expression were able to lyse lung carcinoma cells from the same patient. TINKs from colorectal and breast cancer patients exhibit a CD56^bright/CD16^dim phenotype and express VEGF, thereby promoting neo-vascularization reminiscent of their normal physiologic role in the placenta during pregnancy [108].

Thus, it appears that the tumor immune-suppressive micro-environment has a profound effect on infiltrating TINKs, allowing even cells that would otherwise be susceptible to activated NK cells, such as CSCs, to survive.

Conclusions and perspectives

Accumulating data support a role for NK cell-mediated cytotoxicity against CSCs. Several studies have also revealed a positive influence of NK cell tumor infiltration on patient prognosis [109, 110]. Nevertheless, most currently available data are from in vitro systems using cytokines for activation of NK cells as well as high NK cell to target ratios, both of which are not easily achievable in vivo. The NK cell to target ratios in in vitro cytotoxicity experiments usually range from 10:1 to 1:1, which is evidently higher than the ratios observed in histologic tumor sections [59, 60]. When attempts were made to achieve a more pragmatic NK cell to target ratio (i.e., mimic an in vivo environment), it became evident that resistance developed similarly to what is typically seen in cancer patients with progressive tumors [111]. The in vitro resistance in these experiments was found to be associated with up-regulation of HLA I molecules and down-regulation of NKG2D ligands. Not only do primary tumors exhibit local effects on TINKs in patients, they also affect circulating NK cells by lowering their absolute number, as observed in patients with breast cancer compared to healthy female donors [112]. Patients whose tumors exhibited a pathologic response to neo-adjuvant chemotherapy showed an increase in circulating NK cells as opposed to patients with non-responding tumors. In another breast cancer patient study it was found that, although the numbers of circulating NK cells did not differ from those in healthy female donors, the patient-derived NK cells exhibited a relatively lower expression of most stimulatory receptors [113]. Thus, an opportunity exists for immunotherapy through induction and activation of tumor-suppressed NK cells. Activation of autologous NK cells ex vivo or in vivo has advantages over the use of allogeneic NK cells. The rationale behind this concept is mainly related to safety, due to the absence of graft-versus-host disease and the additional advantage of not being subject to potential rejection. Cytokine activation of immune effectors ex vivo represents an opportunity for activation without a systemic exposure to cytokines that may be particularly toxic, as has been illustrated by IL-2 treatment in patients with metastatic melanoma [114]. Use of cytokine combinations such as IL-2 and IL-15 may activate both major subsets of immune effector cells, i.e., CTLs and NK cells. This method of induction may be envisaged more easily with ex vivo exposure than with in vivo exposure [115]. Additional ex vivo pre-infusion manipulations using e.g. anti-CTLA-4 or anti-PD-1 antibodies may prevent up-regulation of inhibitory receptors during stimulation. The latter has already been shown to be effective in experimental mouse models [116].

Combination therapies aimed at inducing stimulatory ligands such as MICA/B in tumor cells and blocking inhibitory ligands on NK cells could be particularly effective. In this respect, histone deacetylase inhibitors (HDACi) such as trichostatin A, sodium butyrate and valproic acid have been shown to induce MICA and MICB in in vitro human ovarian and cervical cancer cell line models [117]. These drugs could add to the effects of anti-PD-L1 or anti-PD-1 antibodies currently used in tumor immunotherapy. This additive effect may be brought about by presenting NK cells expressing NKG2D with the appropriate stimulatory ligands and concomitantly freeing them from the throngs of the PD-L1/PD-1 inhibition. Monoclonal antibodies blocking KIRs are currently in clinical development, and may be combined with other drug-based or immunotherapies [118]. However, tumors may develop additional mechanisms to counteract their own expression of stimulatory NK ligands by either secreting or shedding them off the cell surface through the action of metalloproteases [119]. In the case of MICA, it has been found that metalloprotease cleavage is preceded by conformational changes mediated by the disulfide isomerase ERp5 [120]. These shedding enabling steps may provide additional therapeutic opportunities in appropriately selected patients.

NK cell ligand expression is independent of the mutation burden of a given tumor, which appears to be a major determinant of immunogenicity. Increased immunogenicity leads to CTL effectiveness, as it increases the chance of CTL clones recognizing tumor neo-antigens [121]. Less dependence on tumor mutation burden, together with an inherent sensitivity of CSCs to NK cells, may turn them into ideal candidates for immunotherapy. This may potentially hold true, even for tumors not traditionally thought of as suitable immune targets, if resistance produced by the immunosuppressive tumor micro-environment can be overcome.

Compliance with ethical standards

Conflicts of interest

None declared.