Overview of the Soluble and Membrane-bound Tumor Factors Limiting NK- mediated Immune Surveillance

Many evidences suggest that NK cells are effective in patrolling for and eliminating tumors in their onset phase, but hardly limit the progression of large established solid tumors. Beside the transition of tumor cells towards a more aggressive phenotype, the NK cell efficacy might be limited by a complex immunosuppressive milieu present in the tumor microenvironment. Indeed, different mechanisms damping NK cell function have been shown in these last years. These include a plethora of tumor-derived immunomodulatory soluble factors (TGF-β, MIF, adenosine, LKynurenin, PGE2) as well as soluble ligands (MICA, ULBP-2, PVR, B7-H6) that compete with membrane-bound tumor ligands for binding to activating NK receptors. During NK-tumor cell contact the NK cell function can also be inhibited by the engagement on NK cells of different inhibitory receptors. The specific ligands might be either constitutively expressed at the tumor cell surface (HLA-I, B7-H3, PVR) or de novo induced/up-regulated (PD-Ls) by immunostimulatory factors (IFN-γ, TNF-α). These are largely released during the active phases of the immune responses and exert an unwanted side effect called “tumor adaptive immune resistance”. This review aims to summarize the best-known molecular mechanisms that, at various times and in different ways, can limit the efficacy of the NK-mediated immune surveillance of tumors.


INTRODUCTION
Natural killer cells (NK) are crucial cytolytic effectors belonging to the family of innate lymphoid cell (ILC) [1,2]. Originally described as cells exerting a "natural" cytolytic activity due to their capability to kill the highly susceptible K562 eritroleukemia cell line, it is now well established that they require activation to exert optimal effector functions. Moreover, the susceptibility to NK-mediated killing of established tumor cell lines is superior to that of tumors ex-vivo isolated from patients, as occurs in bone marrow metastases that are much more resistant to NK-mediated aggression [3]. This supports the concept that an effective NK-mediated anti-cancer activity can't dispense with an optimal activation of endogenous or adoptively transferred NK cells. It is also crucial to understand which NK cell subset, once activated, can exert the most effective anti-cancer activity in a particular immunotherapeutic setting. Indeed, it has been recently stressed that a great heterogeneity in NK cell phenotype and functions exists that goes beyond the classical CD56 dim CD16 high and CD56 bright CD16 low/neg NK cell dichotomy [4]. Beside the identification and description of CD56 neg NK cells [5] that are particularly abundant in peripheral blood of virus-infected donors, several studies highlighted the great heterogeneity of peripheral blood CD56 dim NK cells, which include subpopulations characterized by different capabilities of being activated by cytokines, antibodies or tumor contact [6,7]. Thus, adoptive transferred NK cell-based therapeutic protocols should combine optimal activation strategies, the selection of the best NK cell subpopulation, consider the in vivo persistence of the in vitro activated NK cells, and, last but not least, their chemokine receptor repertoire. Indeed, the scarce attitude of cytotoxic CD56 dim NK cells to reach and invade the tumor parenchyma represents a major obstacle for the effectiveness of NK cells, especially in the therapy of solid tumors [8,9]. This might occur also in a chemokine-rich tumor milieu, due to a defective expression in NK cells of the chemokine receptors involved in their migration toward peripheral tissues. This hypothesis matches with the scenario described in neuroblastoma patients whose peripheral blood CD56dim NK cells are characterized by an unusual reduced expression of CX3CR1, the fractalkine (CX3CL1) receptor [10]. Interestingly, CX3CR1 has been shown to be deeply downregulated by transforming growth factor beta 1 (TGF-β1), a pleiotropic soluble factor released by most cell types including tumors that is capable of modulating pivotal effector functions of different immune cells [11,12]. TGF-β1 is one among the several soluble factors present in the tumor microenvironment that recently emerged as potent immune-modulators. Moreover, beside the classical HLA class Imediated inhibition of the NK cell activity, several additional inhibitory signals have been recently described that, during NK-totumor contacts, limit the NK-mediated immune-surveillance. Thus, the choice of the most effective, activated NK cell subsets should not disregard their profile in terms of the cognate inhibitory receptorligand interactions. This review summarizes the best-known tumorderived soluble factors and tumor-associated surface molecules exerting an immunomodulatory role in NK cells.

MIF, adenosine, L-Kynurenin, PGE2
In the past few years, many different tumor-associated regulatory/ suppressive mechanisms have been widely studied ( Figure 1). Among these, molecules shed from the tumor cell surface, such as the soluble ligands of NKG2D [13][14][15], DNAM-1 [16] and NKp30 [17], activating receptors crucial for T-and/or NK cell-mediated immune-surveillance [18]. sMICA, sULBP-2, sPVR and sB7-H6 compete for binding with the ligands expressed on the tumor cell surface, thus hindering the efficient target recognition. Moreover, different tumor-secreted soluble mediators have been shown to clearly suppress the efficacy of the immune system. Among these, the macrophage migration inhibitory factor (MIF, also known as Glycosylation-Inhibiting Factor) [19][20][21] and adenosine, an endogenous purine nucleoside highly produced by tumors expressing CD39 and CD73, ectonucleotidases converting ATP to adenosine. Both MIF and adenosine have been demonstrated to inhibit cytotoxicity and cytokine production in human NK cells, the second mainly through the engagement of the adenosine receptor 2A (AdoR2A), which is coupled to adenylyl cyclase via Gs protein [22,23].
In the mouse model the use of AdoR2A antagonist reduced the metastatic potential of CD73+ tumors [24] that was further reduced by blockade of AdoR2B. Since in vitro AdoR2B blockade had no significant effects on NK cell cytotoxicity, the benefit observed in tumor-injected mice treated with AdoR2 antagonist might depend on both NK cell-dependent and -independent mechanisms. Along this line, in mice selectively lacking AdoR2 expression in myeloid compartment it has been shown that adenosine can indirectly suppresses T and NK cell-mediated antitumor activity by shaping the functions of different myeloid cells. In particular, in this mouse model, a reduced melanoma growth was associated with a significant increase in MHC class II expression and IL-12 release in tumor-associated macrophages (TAM), features fitting with an M1-like proinflammatory macrophage polarization. Moreover, AdoR2neg TAM, dendritic cells (DC), and myeloid-derived suppressor cells (MDSC) showed a clear reduction in IL-10 expression, a cytokine originally described as a negative regulator of IL-12 production in LPSstimulated peripheral blood mononuclear cells [25]. To date IL-10 cannot be definitively included among the tumor-derived cytokines limiting NK cell activity since both stimulatory and inhibitory effects on NK cell functions have been described. On the contrary, a huge number of data have demonstrated the immunomodulatory function of L-kynurenine, the tryptophan catabolite derived from indoleamine 2,3-dioxigenase 1 (IDO1) pathway [26], and of Prostaglandin E2 (PGE2). Both factors deeply affect the cytokine-mediated upregulation of the expression and function of different activating NK receptors such as NKp46, NKp44 and NKG2D [27][28][29] a mechanism that seems to involve the c-Jun N-terminal Kinase (JNK) pathway [29]. The extent of in vivo suppression mediated by L-kynurenine and PGE2 might be considerable since these factors are released by several cell types colonizing tumor microenvironment, including the cancer-associated fibroblasts (CAF) [30,31], MDSC and DC [32]. In particular, IDOexpressing DC exert a deep immune-suppressive effect by affecting not only proliferation and effector function of NK cells, but also by inducing the conversion of CD4+ T cells into CD4+ CD25+ Foxp3+ regulatory T cell (Treg) [

TGF-β1
TGF-β1 is the prototypic tumor-derived immunomodulatory soluble mediator, although many additional functions have been described over the years, highlighting its pleiotropic activity. Different studies reported a significant contribution of TGF-β1 in the epithelial to mesenchymal transition (EMT), a process that allows tumors of epithelial origin to acquire a less differentiated, invasive and prometastatic phenotype [40,41]. TGF-β1 has been shown to suppress the differentiation process and the effector functions of several immune cells [12]. In particular, it represses the development of human NK cells from CD34+ progenitors and inhibits differentiation of CD16 pos NK cells [42], TGF-β1 promotes the conversion of peripheral NK cells to a decidual NK-like phenotype [43] and, as shown in mouse salivary gland, it drives the differentiation of a particular ILC subpopulation sharing NK and ILC1 features [43]. Released as a large latent complex, TGF-β1 remains biologically unavailable until its activation in inflammatory sites such as the tumor microenvironment by signals including low pH, heat, proteases and members of the integrin receptor family [11]. Once activated, TGF-β1 shows a potent inhibitory effect on NK cells, both in vitro and in vivo, limiting the main NK cell effector functions including IFN-γ production and cytotoxicity [44][45][46][47][48]. In this context, in vitro conditioning of NK cells with recombinant TGF-β1 (rTGF-β1) caused severe downregulation of the surface expression of NKp30 and NKG2D, activating NK receptors cooperating in recognition and killing of several tumor histotypes [44]. The same occurred with rTGF-β2 [10], although knockout mice lacking TGF-β1 or TGF-β2 showed distinct phenotypic features suggesting that the two isoforms could also have specific, nonoverlapping functions [49]. Interestingly, while the TGF-β1-mediated downregulation of NKp30 occurred at the transcriptional levels no significant changes in NKG2D transcript was observed in TGF-β1treated NK cells [44]. Accordingly, it has been shown that the reduced NKG2D surface expression observed in the presence of TGF-β1 both in vitro and in vivo, is due to its capability of downregulating at transcriptional and translational level DAP10, the signaling subunit associated with NKG2D [50][51][52]. Very recently it has been shown that TGF-β1 also inhibits the IL-15-induced NK cell activation, particularly by selectively and quickly repressing the mTOR pathway [53], a crucial integrator of both pro-and anti-inflammatory signals. Recent data also suggest that the tumor-derived TGF-β1 might modify the migratory capability of NK cells. Indeed, it has been shown that neuroblastoma (NB) cell lines spontaneously release amounts of TGF-β1 capable of modulating the chemokine receptor repertoire of NK cells [10]. In particular NB-derived TGF-β1 increases CXCR4 and CXCR3 surface expression in all NK cells whereas it decreases that of CX3CR1 in the CD56 dim NK cell subset. Notably, unusual CX3CR1 low CD56 dim and CXCR3 high CD56 bright NK cell populations were observed in peripheral blood of patients with high risk NB (stage 4 or M) [10]. Thus, tumorderived TGF-β1 can affect the expression of chemokine receptors that play a key role in the bone marrow homing, egress, interaction with endothelium and recruitment into peripheral tissues of NK cells. Recently, in a mouse model of pulmonary allergic responses, it has been shown that TGF-β1 is crucial for the generation of allergic response acting as chemotactic factor recruiting ILC2 and eosinophils [54]. Importantly, it has been demonstrated that TGF-β1 modulates in tumor cells, the expression of specific microRNAs and up-regulates B7-H3 (CD276) [55], a molecule belonging to the family of immune checkpoint proteins, which acts as co-inhibitory factor and limits the function of NK cells. Overall data suggest that TGF-β1 antagonists, capable of overcoming or blocking its immunomodulatory effect might represent a valuable adjuvant therapy in the cure of different tumors. In this context, it has been recently shown that blocking of TGF-β1R in combination with antibodies targeting the NB-associated antigen GD2, potentiates the NK-mediated anti-NB activity leading to a reduced tumor growth and increased survival of mice injected with NB cell lines or patient-derived neuroblasts [56]. Moreover, ongoing clinical trials will evaluate the benefit of TGF-β or TGF-βR blockade in neoplastic patients (ClinicalTrials.gov NCT02452008; NCT02581787). It is of note however that, due to the wide homeostatic regulatory role of TGF-β, in order to obtain specific and restrained therapeutic effects, light should be made on the signaling pathways mediated by the different TGF-β isoforms and on the mechanisms regulating the immunomodulatory effects.

KIRs/HLA-I and NKG2A/HLA-E
The most powerful inhibitory pathway affecting the NK cell activity is represented by the interaction between HLA class I molecules (HLA-I) on target and specific inhibitory receptors on NK cells (Figure 1). These inhibitory receptors include Killer Ig-like Receptors (iKIRs), clonally distributed receptors distinguishing among allotypic determinants of the classical HLA-A, -B and -C, and the CD94/ NKG2A heterodimer specific for HLA-E [57][58][59]. Mature NK cells can also express the activating counterpart of these HLA-I specific inhibitory receptors, i.e. activating KIRs (aKIRs) and CD94/NKG2C [60,61]. During NK cell maturation the engagement of inhibitory receptors by their self HLA-I ligands confers functional competence to NK cells, through a process referred to as "licensing" or "education" that has been explained by different models [62][63][64]. Based on the "rheostat" model, during NK cell education NK cell reactivity is tuned by the strength of the inhibitory signal induced by self-HLA-I molecules [65]. The inhibitory receptor repertoire acquired by NK cells during maturation guarantees that in normal conditions inhibitory signals prevail on the activating ones safeguarding HLA-I + autologous healthy cells from NK cell-mediated killing. The NK cytolytic activity is unleashed in pathologic conditions such as virus infection or tumors where transformed cells increase the expression of ligands for activating NK receptors while downregulating that of HLA-I [66]. NK cell reactivity can be limited when tumor cells retain high levels of HLA-I expression, as occur in hematological malignancies such as acute lymphoblastic leukemia (ALL). However, it has been shown that in allogeneic settings such as in the context of haploidentical hematopoietic stem cell transplantation (haplo-HSCT), the differentiation of alloreactive NK cells from the donor, i.e cells expressing KIRs specific for HLA-I molecules absent in the recipient, strongly improve anti-leukemic surveillance [67,68].
The studies describing the beneficial graft versus leukemia (GvL) effect of alloreactive NK cells in haplo-HSCT has inspired the design of new immunotherapies aimed to enhance anti-tumor NK cell reactivity by blocking the interactions between HLA-I and iKIR or CD94/ NKG2A. A fully human anti-KIR mAb (1-7F9, Lirilumab, IPH2101) that recognizes iKIRs (KIR2DL1, KIRDL2 and KIR2DL3) has been generated [69], which favors the NK cell-mediated killing of HLAmatched tumor cells as documented in vitro and in vivo in phase I/II clinical trials involving Acute Myeloid Leukemia (AML) and multiple myeloma (MM) patients, [69,70]. Interestingly, Lirilumab has been successfully combined with the anti-CD20 rituximab to augment NKmediated cytotoxicity against lymphoma cells in vitro [71]. Based on the broad expression of HLA-E on both solid and hematological malignances, a novel therapeutic approach has been designed to block the CD94/NKG2A-HLA-E interaction, by using the humanized anti-NKG2A Monalizumab, which is currently in a phase I/II clinical trial [72]. This novel approach was developed upon the observation that NKG2A+ NK cells predominate in the early period of immune reconstitution after HSCT, thus representing optimal targets to potentiate NK cell-mediated anti-leukemic activity [73]. In addition, NKG2A+ NK cells express higher levels of activating receptors such as NCRs as compared to more differentiated KIR+ NKG2A-NK cells [6]. The high expression level could compensate the low cytotoxic potential displayed by these less differentiated NK cells. KIRs, CD57 and LIR represent phenotypic hallmarks of terminally differentiated NK cells, which show a good cytolytic potential but poor responsiveness to cytokines [74]. Interestingly, in individuals exposed to pathogens such as cytomegalovirus (CMV), the mature NK cell population is characterized by a very high percentage of cells expressing CD94/ NKG2C or aKIR [75][76][77][78], which show features similar to cells of adaptive immunity including clonal expansion capability, strong effector functions and longevity. This "memory-like" population represents a powerful candidate for adoptive NK cell transfer therapy in cancer patients [79]. Along this line it is crucial to define whether its efficacy might be limited during NK-to-tumor contacts by additional inhibitory or co-inhibitory signals, whose activity might be particularly relevant in the context of HLA-I low or negative tumors.

PD-1/PD-Ls, LAG-3/HLA-II and TIGIT/PVR
The PD-1/PD-Ls axis is a well-known immune checkpoint, i.e. inhibitory pathways that physiologically maintain self-tolerance and limit the duration and amplitude of T cell immune responses, thus minimizing tissue damage [80][81][82]. The PD-1 receptor (CD279) has been demonstrated to limit T proliferation and switch off the T cell functions mostly in peripheral tissues. More recent reports show the presence of PD-1 pos NK cells in both cancer patients and in a relevant proportion of healthy donors who were serologically positive for human CMV [83][84][85][86]. PD-1 expression is confined to terminally differentiated NKG2A neg KIR pos CD57 pos NK cells and their antitumor activity can be partially restored in vitro by antibodies disrupting the interaction between PD-1 and its cellular ligands PD-L1 and PD-L2 [86,87]. PD-L1 (B7-H1, CD274) and PD-L2 (B7-DC, CD273) belong to the B7 family that consists of several members including B7-H3 (see below) ( Figure 1). PD-L1 is expressed in several normal tissues, whereas PD-L2 is mainly restricted to antigen presenting cells (APC) [80,81]. Importantly, both inhibitory ligands can be expressed by tumor cells in response to immunostimulatory factors such as IFN-γ and TNFα that are released by activated T and NK cells [80,85,88]. This phenomenon called "adaptive immune resistance" [88] is a mechanism of escape by which cancer cells adapt their phenotype to the pressure of immune responses. Interestingly, it has recently been shown that PD-L1 neg metastatic cells purified from bone marrow aspirates of high risk NB patients have different capabilities of up-regulating PD-L1 in response to IFN-γ [85]. The lack of PD-Ls upregulation in some patient might contribute to a reduced clinical response to therapy with anti-PD-1. However, it should not be disregarded that the therapeutic efficacy of anti-PD1 antibodies has been observed also in patients who at the time of therapeutic decision carried PD-L1 neg tumors. Indeed, the clinical benefit might also depend on the strengthening of the crosstalk between PD-Ls+ APC and PD-1+ immune effectors including T or NK cells. IFN-γ is also a potent inducer of HLA class II molecules that can be recognized by LAG-3, an additional inhibitory mechanism that recently emerged in human NK cells together with TIGIT/PVR and TIM-3/galectin-9 interactions (Figure 1) [89][90][91][92]. Interestingly, the heterogeneous expression of the TIGIT inhibitory receptor observed in NK cells from healthy individuals inversely correlates with their capability of performing degranulation and IFN-γ release in response to IL-12 stimulation [91]. Moreover, low TIGIT expression has been described in CMV-induced terminally differentiated NK cells that appear more resistant to the inhibitory effect mediated by PVR pos MDSC. Since studies explored the cytokine-induced expression of these coinhibitory receptors mainly in long term-cultured NK cell lines such as NK92 [93], data on primary NK cells are required to better understand the relative contribution of these inhibitory pathways and the kinetic that regulates their emergence.

B7-H3R/B7-H3
Another interesting NK-to-tumor contact inhibitory pathway is mediated by the B7-H3 ligand, a tumor-associated surface molecule, also present in tumor-derived exosomes [94], which is endowed with both immune-regulatory and pro-tumoral functions. B7-H3 is capable of inhibiting the cytolytic activity of human NK cells against neuroblasts purified from bone marrow aspirates of high risk NB patients, which are characterized by reduced levels of HLA-I [18] and adhesion molecules (unpublished observation). The B7-H3-mediated inhibitory effect, which depends on its interaction with a still unknown inhibitory receptor, is particularly evident when using xenogenic B7-H3 high transfectants [95]. This suggests that this inhibitory pathway might require peculiar conditions to be unleashed, which could be represented by poor engagement of potent inhibitory NK receptors and/or by the presence of weak activating signals. It should be mentioned that in mouse B7-H3 has been described as a "friend" in tumor immunology [96]. In particular, intratumoral injection of an expression plasmid encoding mouse B7-H3 led to a complete NK-(and T-) mediated regression in approximately half of tumor-bearing mice [97]. Interestingly, the mouse B7-H3 gene codes for a molecule characterized by two Ig-like domains while human cell tissues predominantly express a four Ig-like domains isoform resulting from exon duplication [98]. While in human B7-H3 is still an orphan ligand, an activating receptor has been identified in mice that is represented by TREM-like transcript 2 (TREML2, TLT-2), expressed by activated T cells and myeloid cells [99]. Importantly however, Leitner and coworkers who extensively faced this issue did not find evidence for B7-H3/TREML2 interaction in human [100]. Considering that B7-H3 belongs to the B7 family that includes members interacting with both activating and inhibitory receptors, it can't be excluded the existence of a complex scenario resulting from the capability of B7-H3 to engage receptors with opposite signal. However, to date most in vitro and in vivo data lean toward a B7-H3 inhibitory role in human and the B7-H3R/B7-H3 axis has been included among the immune checkpoints [81,101,102]. An adjuvant therapeutic strategy in cancer might be represented by antibodies disrupting the interaction between B7-H3 and its receptor/s. Different phase I Clinical trials are ongoing with humanized anti-B7-H3 mAbs (NCT02628535; NCT02982941; NCT02475213) [81] and encouraging results have been obtained in the first in-human intrathecal injection of radioiodinated anti-B7-H3 Ab (following standard therapy) in 21 neuroblastoma patients with recurrent Central nervous system (CNS) metastasis [103]. It is of note that the therapeutic efficacy of anti-B7-H3 mAbs might depend not only by the strengthening of the NK-(and T-) mediated anti-tumor responses, but also by the weakening of the direct pro-tumoral activity of B7-H3. Indeed, studies in tumor of different histotype showed that high B7-H3 expression drives tumor cell progression through different molecular mechanisms. These include promotion of migration and invasiveness [104] and reduction of sensibility to chemotherapyinduced apoptosis, as demonstrated in breast [105] and pancreatic carcinoma [106]. Accordingly, high expression of B7-H3 is a negative prognostic factor in several tumors including neuroblastoma [18,[107][108][109][110][111]. In particular, in primary neuroblastoma high B7-H3 surface expression, in terms of both intensity and percentage of positive cells, has been correlated with poor event-free survival also in patients with localized disease (stage 1-3), suggesting that high B7-H3 expression might discriminate between low-and high-risk patients who need a more careful follow-up. NK cell-based immunotherapy is becoming a promising approach for the treatment of both hematological malignances and solid tumors.

Ready to kill or killers who require to be armed?
Originally described as cells exerting a "natural" cytolytic activity it is now well established that, to exert optimal effector functions, NK cells require activation via immunostimulatory cytokines and tumor contact.
NK cell population is heterogeneous and includes subsets characterized by different capabilities of being activated and different effector functions.
Once activated, NK cells need a chemokine receptor repertoire ideal to their recruitment in inflamed tissues such as tumors.

How tumors may dump NK cell function in vivo?
Tumor cells purified from patients are less susceptible to NKmediated killing than established tumor cell lines commonly used in vitro.
A complex immunosuppressive milieu is present in the tumor microenvironment.
A plethora of tumor-derived immunomodulatory factors exists, either soluble or membrane bound, that might limit the NK-mediated immune-surveillance in vivo.

Soluble immunomodulatory mediators
Tumor cells may release soluble ligands (sMICA, sULBP-2, sPVR and sB7-H6) that compete with membrane-bound tumor isoforms for binding to activating NK receptors. TGF-β1 is the prototypic tumor-derived immunomodulatory soluble mediator. In NK cells, it represses development from CD34+ progenitors and differentiation, limits the main effector functions (IFN-γ production and cytotoxicity), reduces the expression level of activating receptors (NKp30, NKG2D) and alters the chemokine receptor repertoire.

NK-to-tumor contact immunomodulatory signals
NK cell reactivity can be limited when tumor cells retain high levels of HLA-I expression, as occur in hematological malignancies.
Tumor cells may express ligands (PD-L1, PD-L2, B7-H3) belonging to the immune checkpoint family, i.e. inhibitory pathways that limit the duration and amplitude of anti-tumor responses.
B7-H3 not only weakens the strengthening of the NK-(and T-) mediated anti-tumor responses, but also exerts a direct pro-tumoral activity.
Tumor cells can constitutively express inhibitory ligands or increase their expression in response to immunostimulatory cytokines (IFN-γ, TNF-α) that are released by activated NK (and T) cells, a phenomenon called "tumor adaptive immune resistance".
IFN-γ, besides inducing PD-Ls expression, is also a potent inducer of HLA-II that can be recognized by LAG-3, an additional inhibitory mechanism that recently emerged together with TIGIT/PVR and TIM-3/galectin-9 interactions.

Future directions
When designing immunotherapies, we can't neglect the complexity of the immune-suppressive milieu characterizing the tumor microenvironment. Thus, NK cell-based therapeutic protocols should combine optimal activation strategies, the selection of the best NK cell subpopulation, consider the in vivo persistence of NK cells, and their chemokine receptor repertoire. Valuable adjuvants may include the use of humanized mAb to disrupt immune checkpoints pathways and/or to hinder the effects of soluble factors that in vivo may dampen the NK cell activity against tumors.