Human leukocyte antigens (HLA) are the most polymorphic gene in the human genome, and expressed on all healthy nucleated cells at variable densities that reflect cellular health and HLA allelic diversity (71, 72). HLA-A, HLA-B and HLA-C are together classified as “conventional HLA I”, or class Ia molecules, and represent the most polymorphic HLA I alleles (72). Unconventional HLA I molecules, classified as class Ib, include HLA-E, HLA-F, and HLA-G (71). These exhibit relatively low polymorphism, and have important roles in tolerance, inhibition, and pregnancy, with HLA-E remaining relatively conserved between human and primates (72–76). Each HLA I, except for HLA-F, is comprised of three alpha domains, and stabilized by β2 microglobulin and by peptides presented in the binding groove of its variable domain (57). HLA-F is less understood, but can associate with β2 microglobulin and exhibits at least two configurations: closed, where it seems to not present peptides, and open, where it presents uncommonly long peptides in an open-ended binding groove (77).

HLA I is co-evolving with the killer immunoglobulin-like receptors (KIR) and members of the NKG2 family (78). Their interactions can signal for activation or inhibition, depending on the specific receptors ligated, bound peptides, immunologic experience, and compound allelic partnerships (3, 79) ( Figure 2A, B ). Groups of conventional HLA I molecules display conserved public epitopes that enable groupings of HLA I molecules as “KIR ligands”: HLA-C molecules expressing Asp or Lys at position 80 define the HLA-C1 or C2 subgroups, respectively, and account for all HLA-C molecules (80, 81). A subset of HLA-A and B molecules each contain the HLA-Bw4 motif, and some specific alleles of HLA-A (HLA-A*03 and *11) can act as NK cell ligands, and the balance of HLA-A and -B alleles are not known to engage with any KIR (79, 82, 83).

KIR molecules with long cytoplasmic tails signal for inhibition via ITIMs encoded in their transmembrane regions. HLA-C1 engages KIR2DL3 and HLA-C2 engages KIR2DL1 (84). KIR2DL2 preferentially binds with HLA-C1 molecules, but some alleles also exhibit binding affinity with HLA-C2 (85, 86). HLAs harbouring the Bw4 motif bind with KIR3DL1; the remainder of HLA-B molecules and most of the HLA-A molecules are not known to encode a ligand for KIR (87). KIR3DL2 binds to HLA-A*03 and A*11 molecules, but only when they present certain peptides, including exogenous CpG DNA and peptides derived from Epstein-Barr virus (82, 83, 88). Inhibitory KIR2DL4 and immunoglobulin-like transcript (ILT)-2 and -4 molecules on NK cells bind HLA-G and signal via ITIM (89). HLA-G is an unconventional HLA molecule whose expression is normally limited to immune privileged sites including the cornea (90) and placenta (91), but has also been found upregulated, and associated with poorer prognosis in several cancers, including colorectal cancer (92), pancreatic cancer (93), breast cancer (94), and ovarian cancer (95). Soluble HLA-G has been found to dampen NK cell function, including impairment of NK cell migration to inflamed tissues (96).

KIR molecules with a short cytoplasmic tail recruit the ITAM-containing molecule DAP12 to the immunologic synapse. Activating KIR, including KIR2DS1, KIR2DS2, and KIR3DS1 are each known to bind with HLA I molecules, albeit with lower affinity than their inhibitory counterparts (97, 98). Additional activating KIRs (KIR2DS3, KIR2DS5) also have ITAMs and may contribute to activation and may bind with specific alleles (99), though the impact of these ligands and their corresponding signals have not been well established.

KIR2DS1 binds HLA-C2 (Lys⁸⁰) and, at high levels, individuals that are homozygous for HLA-C2 have KIR2DS1+ NK cells that are hyporesponsive (97, 100, 101). This has implications in hematopoietic cell transplantation, where HLA-matching is only beneficial in the absence of homozygous HLA-C2 (102, 103). Moreover, for the treatment of B-cell malignancies, addition of rituximab to chemotherapy did not improve survival outcomes in patients with KIR2DS1-HLA-C2/C2 (104), suggesting that KIR2DS1 and HLA-C status could be a predictive marker for efficacy of rituximab due to the potential of hyporesponsive NK cells (104).

Although KIR are not known to recognize HLA-presented peptides specifically, the biochemical features and structural/allosteric variations that they introduce can influence receptor binding and consequent signaling (57, 105). This is the case for KIR2DS2, which binds conditionally to HLA-C1 and HLA-A*11 alleles, when they present viral peptides, including those derived from viral helicases (3, 106), though their role in cancer is unclear. Similarly, KIR3DS1 is activated specifically by the open (peptide presenting) configuration of HLA-F (107).

The NKG2 family members engage with HLA-E, but in a manner distinct from conventional KIR-HLA interactions (108). NKG2A, an ITIM-containing receptor, and NKG2C, which engages ITAM-containing DAP12 for signaling, both bind with HLA-E and heterodimerize with CD94 (49, 50, 109–111). HLA-E molecules present the leader peptides of conventional HLA molecules but also present antigenic peptides, including neoantigens, to both T and NK cells (105). In healthy tissues, HLA-E expression is low (112), but high HLA-E expression is linked to poor patient outcomes in several cancer types, including gastric cancer (113, 114), colorectal cancer (113), pancreatic cancer (93), breast cancer (94), renal cell carcinoma (115), ovarian cancer (116), and glioma (111, 117), suggesting a dominant impact of NK inhibition on limiting anti-cancer activity. Indeed, high expression of NKG2A, both alone and with high HLA-E expression, is associated with poorer patient survival in liver cancer (118). Likewise, co-expression of the high-density HLA-E allele (HLA-E*01:03) and NKG2A associates with Epstein-Barr virus-associated lymphoma (119). Given the known impacts of HLA-E expression on cancer development, it is likely that these are exacerbated by NKG2A expression on NK cells in other cancer types, though this remains to be studied extensively.

In addition to signaling during target cell engagement, KIR-HLA and NKG2A:HLA-E interactions are central to a process called NK cell “education”, “tuning” or “licensing” (120–122). Each of KIR and NKG2A are expressed on only a subset of NK cells, their HLA ligands may or may not be available in the host, and the avidity of receptor-ligand binding is variable. Therefore, the extent to which each NK cell can be inhibited by “self” HLA varies. An individual’s NK cell repertoire therefore consists of both educated and uneducated cells (85, 123). Resultantly, the reactive thresholds of individual NK cells differ, and create a spectrum of responsiveness and anti-tumor effector function that varies within and between individuals (3, 122, 124).

NK cells that are most sensitive to inhibition by “self” HLA I exhibit a greater mobilization of activating receptors in the actin meshwork (8), greater DNAM-1 expression (125), a greater density of granzyme B (9) and lower levels of SHP-1 (53)– making them more easily activated in the absence of strong inhibition. Sensitivity to inhibition, however, is the Achille’s heel of educated NK cells in cancer therapy because HLA I on tumors often persists, or becomes upregulated in response to IFN-γ (126). Strategic selection of KIR and HLA allelic combinations and NK education status, or blocking signals for inhibition, will therefore be critically important for the success of NK cell-based immunotherapy (3, 124).

During cancer development, there is extensive pressure on HLA expression driven by immune activity: T cell-mediated recognition may select for clones lacking HLA expression, but this may create a target for NK cells by interrupting signals for inhibition. Downregulation of classical HLA I expression is reported for several solid cancers, including melanoma, cervical, breast, colorectal and lung cancers (127–130). HLA I loss may represent partial or complete losses of HLA gene loci (129, 131), or components of the HLA processing and presentation pathway (127, 131). Frequently, loss of heterozygosity for the HLA-ABC genes, and other genes involved in HLA I processing and presentation, is observed following acquired resistance to immune checkpoint inhibitors like anti-PD1/PD-L1 (127, 132, 133).

KIR is highly polymorphic; an individual’s KIR configuration has been associated with both the risk of developing cancer and their ability to respond to cancer therapies. KIR expression is not known to change as a function of cancer, but the presence of self-sensitive KIR can enable inhibition of otherwise-activated NK cells. Beyond strategic donor selection to optimize KIR/HLA genotype for potent NK cell alloreactivity, both KIR and NKG2A can be directly targeted through monoclonal antibodies. Lirilumab, an anti-KIR2DL1/2/3, has been tested in a phase I clinical trial against multiple myeloma, and supports alloreactivity of NK cells (134). Monalizumab (anti-NKG2A), in combination with Cetuximab, has shown anti-tumor effects in a clinical trial of patients with squamous cell carcinoma of head and neck with high expression of HLA-E and NK cell infiltration. Taken together, this implies the important role of selecting/regulating KIR/NKG2A in cancer treatment.

Perhaps the best known member of the NKG2 family, NKG2D, is unique in that it forms homodimers (135), and signals for activation via DAP10 (62). Ligands for NKG2D are structural homologs that co-evolved alongside HLA I molecules and include MHC class I chain-related molecule A and B (MICA/B), and UL16 binding proteins 1,2,3,4,5 and 6 (ULBP1-6) ( Figure 2C ) (136, 137). NKG2D ligands are typically expressed at low cell surface densities, but their density on the cell surface increases in response to DNA damage (138), oncogene activation (138), infection, excessive proliferation, and oxidative stress, earning them the title of “stress ligands” (138). The NKG2D receptor itself is responsive to cytokines associated with immune priming (ie. IL-2, IL-15), which increase basal phosphorylation of DAP10, priming the cell to deliver activating signals (139). Together, NKG2D and its ligands generate a robust system to detect hallmark features of stressed cells.

Presented on a target cell, NKG2D ligands signal for NK cell activation, and NKG2D is central in cancer immunosurveillance. In mouse models, antibody-mediated neutralization of NKG2D interrupted immunosurveillance that otherwise prevented carcinogen-induced tumors (140). Likewise, elimination of NKG2D using microRNA silencing rendered mice more susceptible to cancer growth (140–142). In humans, histological studies have confirmed high and co-expression of NKG2D ligands in cancers of the breast (142, 143), colon (144, 145), gastric system (146, 147), lung (148, 149), skin (150), ovary (151, 152), pancreas (153), prostate (148, 154), and kidney (148). For NKG2D, diminished receptor expression is known to occur in response to hypoxia and diminished STAT3 activity (155), and in response to shed soluble ligands released following protease-mediated cleavage, which block the receptors from signaling (156).

Existing and nascent approaches to cancer therapy alter the expression of NKG2D ligands and encourage NK cell-mediated tumor killing. For example, ionizing radiation (157, 158), histone deacetylase (HDAC) inhibitors (159), and chemotherapy (158, 160) each prompt increased levels of NKG2D ligands. Furthermore, agents that prevent ligand matrix metalloproteinase activity by small molecule inhibitors can potentiate NK cell response by preventing ligand shedding (161).

The NKG2D receptor is a logical target for immunotherapy, and it is now being incorporated into immunotherapeutic approaches including NKG2D-CAR-NK, NKG2D-CAR-T cells (162, 163), CAR-T cells with a DAP10 intracellular domain (164), and bi-specific killer-engagers (BiKEs) (165, 166). While these therapies have indeed enhanced NK cell mediated anti-tumor activity, complete tumor control will require combination therapies that extend beyond the targeting of NKG2D alone.

The tumor necrosis factor receptor superfamily (TNFRSF) is a group of proteins that primarily regulate cell activation, differentiation, and survival, either as membrane-bound factors or cleaved, soluble factors. Here, we focus our discussion on those studied for their roles in NK cell function and reactivity, including the death receptor ligands Fas and TNF-related apoptosis inducing ligand (TRAIL), and costimulatory members 4-1BB and CD40L ( Figure 3 ). TNFRSFs engage TNF superfamily (TNFSF) ligands, which can often be in membrane bound and soluble forms (167). This family of receptors and ligands signal primarily through NF-κB to induce proinflammatory function, or induce apoptosis of the target cell (168). Notably, while other TNFSF members have been reported to be expressed by NK cells, or to impact their function [i.e. GITR (CD357) (169), LIGHT (CD258) (170) and CD70 (171)], relatively little is known on their roles and targetability in cancer; these remain open questions and opportunities.

The natural cytotoxicity triggering receptors (NCRs) are a group of HLA III genes that primarily generate activating signals in response to ligand binding through a transmembrane-encoded ITAM, or association with ITAM-containing adapter molecules (199) ( Figure 4A ). The NCRs do not themselves encode ITAMs, and instead recruit and signal through ITAM adapter molecules, including DAP12, CD3ζ and FcϵRIγ (199). Each exist as splice variants, with some isoforms conveying inhibitory signals via an ITIM-like sequence (200). Although there are others, three NCRs that have been studied in the context of cancer are NKp30, NKp44, and NKp46.

NCR ligands typically become available as a result of cellular stress, inflammation, and transformation (199). They are often expressed on tumor cells, or released during oncogenesis, and likely evolved to respond to ligands available in disease contexts. Cancer-derived ligands are not the exclusive binding partners for the NCRs, but for simplicity, we focus on those involved in cancer pathology here.

NKp30 binds B7-H6 (205), BAT-3 (205), and heparin sulfates (214, 215). NKp30’s transmembrane domain associates with the ITAM-containing adaptors CD3ζ and FcϵRIγ. There are at least six splice variants of NKp30, which differ in their tissue distribution, engagement with adapter molecules, and outcomes upon ligand binding, notably on production of IFN-γ and IL-10 (216). Variants a-c are the most common, with a and b favouring IFN-γ production and NKp30c associating with IL-10 production (216). B7-H6 has been proposed as a biomarker for development and progression in an array of cancers (217–219); by extension, leveraging this expression as a target for immunotherapy may be possible.

NKp44 signals for activation when bound by PDGF and heparan sulfates, and signals for inhibition upon binding to PCNA (199, 215, 220, 221). Three splice variants are described for NKp44, where the transmembrane domain in the NKp44b and NKp44c isoforms associate with DAP12 to signal for NK cell activation (222). NKp44a signals via an ITIM-like domain and binds PCNA, and at high surface densities, this splice isoform can inhibit NK cell function (200, 223). Dominant expression of this “inhibitory” NKp44 isoform is associated with poorer survival in patients with acute myelogenous leukemia (223), and PCNA is currently being explored as a biomarker and target for monoclonal antibodies (224).

NKp46 associates with CD3ζ and FcεRΙγ (225, 226). Key cancer associated ligands for NKp46 include heparan sulfates (215), vimentin (199, 225), and ecto-calreticulin (18). NKp46 expression is correlated with the degree of NK cell response generated (227) and IFN-γ produced via NKp46 signaling alters the deposition of fibronectin limiting metastasis in murine melanoma (228). Like NKp30, the ligands for NKp46 have been proposed as biomarkers for cancer severity and progression (229), endorsing these as potential targets for NK cell-based immunotherapies.

Targeting NCRs or their ligands for cancer immunotherapy is an area of active exploration that warrants investigation in NK cell-based cancer immunotherapy. Current studies are investigating the potential of using BiKEs or CARs that use the NCRs to enhance tumor recognition (230, 231). For example, a trifunctional natural killer cell engager (TriKEs) targeting the AML antigen, CD123, while simultaneously binding NKp46 and CD16a on NK cells has demonstrated efficacy in murine models and nonhuman primates (232) and is currently being tested in early clinical trials (NCT05086315). A similar strategy – coupling NKp46 engagement to anti-CD20, CD16 and the IL-2R beta chain has likewise generated promising preclinical results (233). NKp30 has been targeted on tri-specific engagers, coupling NKp30 engagers with Fab and anti-EGFR to create a potent pathway for tumor cell lysis and NK cell cytokine production (234). These strategies are in their infancy, but illustrate the promise of targeting such conserved NK cell ligands for immunotherapeutic purposes.

CD16a is expressed on NK cells, monocytes, and macrophages, and is the major receptor engaged for ADCC ( Figure 4B ) (235). CD16a encodes an ITAM motif, which recruits the signaling adapters FcϵRIγ, and CD3ζ to signal NK cells for proliferation, survival, cytokine production and degranulation (43, 236). Notably, CD16a activates NK cell degranulation, even in the absence of other activating signals (237).

Antibodies can function by neutralizing targets or opsonizing them for phagocytosis or killing. Target cell binding by antibodies enables cross-linking of CD16a; these antibodies can be produced endogenously by plasma cells or delivered as therapy (21). In NK cells, CD16a Fc receptor crosslinking triggers ADCC (238). NK cell-mediated ADCC is a central mechanism to the killing primed by anti-HER2 (trastuzumab) (239) anti-GD2, (dinutuximab) (240), and anti-CD20 (rituximab) (241). In mice deficient for CD16a, or in which CD16a engagement with antibodies is blocked, tumor growth is exacerbated (239). Noteworthy, CD16a is susceptible to cleavage by metalloproteases within the tumor environment, and these soluble fragments could block CD16a on NK cells and create an additional opportunity to evade ADCC (242). Strategies to strengthen antibody binding and Fc receptor signaling are being explored and include antibody Fc engineering to maximize binding and activation, non-cleavable CD16 molecules, bi-specific antibodies, and antibody-drug conjugates (37, 238, 243).

Signalling lymphocytic activating molecule (SLAM)-family receptors are a group of type I transmembrane receptors, expressed on hematopoietic cells (244), especially during and shortly after differentiation ( Figure 4C ) (245). SLAM-family receptors expressed on NK cells include SLAMF1, SLAMF2 (CD48), SLAMF3, SLAMF4 (2B4), SLAMF6, SLAMF7 (CRACC) and SLAMF8– with CD48, 2B4, and CRACC being the most prominent (246).

These SLAM-family receptors, excluding CD48 and SLAMF8, signal through ITSMs (247). SLAM-family receptors are homotypic, except for the partnership of 2B4 and CD48 (248, 249). CD48 and 2B4 also, uniquely, bind in both trans (ie. with other ligand-expressing cells) or in cis (ie. on the NK cell surface) (250, 251). Cis binding conveys baseline 2B4 ITSM phosphorylation, and a higher signalling threshold for activation due to competition with ligands for binding in trans (248, 250, 251).

SLAM-family receptors can be expressed on hematologic (252) and solid tumors (245, 249). Indeed, CD48-2B4 signalling has been linked to early NK cell activation by monocytes, followed by exhaustion (253). NK cell monocytes isolated from hepatocellular carinoma highly express CD48, and blocking this CD48-2B4 interaction leads to relatively decreased NK cell activation and sequential exhaustion (253). Monocytes isolated from hepatocellular carcinoma have high CD48 expression and blocking the 2B4-CD48 interaction decreased NK cell activation and exhaustion (253). This provides evidence that a multitarget approach might be necessary for checkpoint therapy, and that targeting SLAM-family receptors (whether inhibiting or activating these receptors) is not a one-size-fits all for each tumor. Additionally, both CD48 and CRACC are highly expressed on multiple myeloma, and these are now being investigated as targets for monoclonal antibody therapies (254). Monoclonal CD48 has shown promising pre-clinical results at decreading multiple myeloma tumor growth (252). Undoubtedly, the extensive expression of SLAM family members and their ability to contribute to NK cell activation will promote further investigation toward their clinical use.

The DNAX-accessory molecule-1 (DNAM-1) and TIGIT receptors are commonly expressed on NK cells and T cells, and to interact with the same ligands: CD155 and CD112 (64, 255, 256). DNAM-1 is an activating receptor that signals via ITT; TIGIT is an inhibitory immune checkpoint that encodes an ITIM (65, 257) ( Figure 5 ). CD96 (TACTILE) also contains an ITIM, and binds to CD155, but not CD112 (257), and is thought to have secondary roles in cell-cell adhesion (257). Recently, KIR2DL5, an ITIM-containing KIR, has also been identified as a receptor for CD155 (258).

Each of CD155 and CD112 bind TIGIT and DNAM-1, signaling for inhibition through TIGIT or activation through DNAM-1. CD155 and CD112 both bind TIGIT with a higher affinity that DNAM-1 (259, 260). CD155 and CD112 are expressed at low levels on healthy cells and are upregulated in response to inflammation, cellular stress, reactive oxygen and nitrogen species (261–263), and following chemotherapy (264). Overexpression of these ligands on the tumor is usually associated with poorer prognosis for patients: increased tumor expression of CD112 correlates with increased tumor size and stage, in cancers of the gallbladder (265), colon (266), ovary (267), and pancreas (268). Likewise, tumor overexpression of CD155 is associated with poor prognosis and progression of multiple cancers (269) including breast cancer (270), gastric cancer (271), non-small cell lung cancer (272), melanoma (273), colorectal cancer (274), and sarcoma (275).

TIGIT modulates DNAM-1 mediated activation through the competitive binding of their shared ligands (276). Further, TIGIT directly inhibits DNAM-1 activation through cis interactions that interfere with DNAM-1 homodimerization (277). TIGIT blockade enhances NK cell responsiveness both in vitro and in vivo, particularly when DNAM-1/CD155 interactions remain intact (277, 278). In mouse models, blockade of CD96/CD155 binding results in reduced metastatic spread in vivo– in part by enabling DNAM-1/CD155 binding through loss of CD96 competitive binding (279, 280).

Through alternative splicing, CD155 can be released in a soluble form (sCD155), which binds to DNAM-1 with greater affinity than TIGIT or membrane-bound CD155, and block binding and signaling for activation via DNAM-1 (281). High expression of sCD155 is associated with poor prognosis and increased cancer progression in lung, breast, liver, and gynecological cancers (282–284). sCD155 binding can also result in the endocytosis of DNAM-1, making it less available to generate signals for activation (285, 286).

The opposing roles these receptors play is evident in cancer progression; in acute myeloid leukemia, high expression of DNAM-1 is associated with longer progression-free survival and overall survival (287), while increased TIGIT expression was correlated with NK cell dysfunction and poorer patient outcomes (288). Further, CD96 expression was linked with an increased immunosuppressive immune signature and poor patient prognosis in gastric cancer (289).

Targeting TIGIT, and thereby blocking its inhibitory interaction with CD155 and/or CD112, has been the focus of recent research, with several monoclonal antibodies currently undergoing clinical trial. A phase II trial assessing tiragolumab in combination with atezolizumab demonstrated an objective response rate of 37%, with an objective response rate of 66% in patients with a high PD-L1 status. However, some adverse events were observed, with two treatment-related patient deaths (290). Phase III trials are currently under way for both small cell and non-small cell lung carcinoma (NCT04256421; NCT04294810), as well as trials of tiragolumab against other malignancies, such as melanoma (NCT05116202; NCT03554083), pancreatic cancer (NCT03193190), gastric cancer (NCT0493322; NCT05251948), and cervical cancer (NCT04300647). Of course, administration of these antibodies can impact other lymphocytes, and understanding the specific roles of NK cells compared with other lymphocytes are needed.

Immunologic checkpoints are best studied in the context of T cells in cancer but have similar functions in NK cells (51). Immunologic checkpoints known to impact T cells may also inhibit NK cells. These include T cell immunoglobulin and mucin-domain containing-3 (TIM-3), lymphocyte activation gene (LAG-3), programmed cell death protein-1 (PD-1) and cytotoxic T-lymphocyte-associated antigen 4 (CTLA-4) ( Figure 5 ). Like the NCRs, these immunologic checkpoints have and share several ligands associated with cellular transformation and cancer. The ligands are discussed below, first alone, and then in the context of their receptors.

Reinvigoration of T cells has been the major goal of monoclonal antibodies against TIM-3, LAG-3 and PD-L1/L2, but NK cells may be also rescued by their inhibition in the cancer microenvironment. TIM-3 signals for inhibition via five conserved tyrosine residues, and Bat-3 is associated with its transmembrane domain at steady state (322, 323). Known ligands for TIM-3 include Gal-9, HMGB-1, CEACAM-1, and phosphatidylserine. NK cells expressing TIM-3 have suppressed cell-mediated cytotoxicity and can be rescued with TIM-3 blockade (324, 325).

The specific role for Gal-9’s interactions with TIM-3 on NK cells is unclear, with reports of both stimulation for IFN-γ production by the NK-92 cell line (296) and immunoregulation in viral infections and pregnancy (326, 327), suggesting that its role may be nuanced or influenced by microenvironmental features. In T cells, Gal-9/TIM-3 binding triggers release of Bat-3, which liberates an immunosuppressive signal (328). TIM-3 signaling results in inhibition of T cell proliferation and cytokine productions, potentially leading to T cell death (291, 292).

In cancer, the impacts of TIM3:Gal-9 interactions on NK cells are similarly controversial, with reports alternately ascribing pro- and anti-tumor roles. TIM-3:Gal-9 interactions are associated with NK cell exhaustion and decreased cytotoxicity in AML and gastrointestinal tumors (329, 330). Increased Gal-9 expression is associated with worse overall survival, decreased progression-free survival, and increased metastasis in cancers of the liver (331), kidney (332) and virus-associated cancers (333). Conversely, and consistent with an activating role for TIM-3:Gal-9 interactions, higher expression of Gal-9 is associated with less tumor dedifferentiation and metastasis in cervical cancer (334). Gal-9 drives apoptosis of melanoma, leukemia, and lymphoma cell lines (335, 336). Higher Gal-9 expression is associated with superior outcomes in patients with gastric cancer (337) and triple-negative breast cancer (338).

LAG-3 is a member of the immunoglobulin superfamily receptors, and binds MHC class II, and LSECtin (339–343). On T cells, LAG-3 is induced by cell activation, ostensibly to enable control of ongoing lymphocyte responses (344, 345), associates with the TCR and binds HLA II with higher affinity than CD4, to regulate signaling and inhibit proliferation (344–346). LAG-3 is known to be expressed by NK cells (347) with the highest expression being reported on activated, adaptive, and mature NK cells (348).

The expression of MHC II in the tumor microenvironment is typically associated with improved immune cell infiltration and patient outcomes (349, 350). Expression of LAG-3, on the other hand, is well established as a marker of increased tumor progression and aggressiveness across cancer types (351–353). As a result, therapies directly targeting LAG-3 are currently being explored, both as monotherapy and in combination with immune checkpoint blockade. While studies evaluating LAG-3 therapies rarely profile NK cells, one study evaluating the response of patients with melanoma to a combination of anti-LAG-3 and anti-PD-1 found that adaptive LAG-3⁺ NK cells were most prominent in those who responded to immunotherapy (348). This data supports the role for NK cells influencing response to checkpoint inhibitors. Interestingly, this has also been suggested by earlier work where NK cells isolated from a murine model with LAG-3 deficiency exhibited defects in NK cell mediated anti-tumor immunity (354). Additional research is needed to better understand the role and importance of LAG-3 in NK cell cancer killing.

PD-1 is expressed most prominently on T cells, and has recently been described to be present, but at lower densities on mouse and human NK cells (355, 356). Interference with PD-1/PD-L1/2 signaling with monoclonal antibodies on these mouse NK cells increases their target cell cytotoxicity, confirming a parallel function to that described for T cells (357, 358). PD-1 expression on human NK cells is not yet well understood and seems to be limited to the adaptive NK cell subset (359–361). Nonetheless, blockade of the PD-1/PD-L1 pathway improves NK cell responsiveness both in vitro and in vivo (362, 363). Blocking PD-L1⁺ NK cells with anti-PD-L1 improves degranulation and cytokine production, as well as control tumor in vivo, indicating that NK cells may contribute to the success of checkpoint inhibition independent of PD-1 expression (364).

Monoclonal antibodies that interfere with PD-L1/L2-PD-1 interactions have become part of treatment for an array of tumors. Durvalumab and atezolizumab (both anti-PD-L1 monoclonal antibodies) are approved as first-line treatments for non-small cell lung cancer (365). Further, durvalumab is approved for bladder cancer and atezolizumab is approved for treatment of triple-negative breast cancer and liver cancer (366–369). Pembrolizumab (anti-PD-1) is also approved for use in MSI-H/dMMR cancers (319). Whether and to what extent these impact NK cell function with impacts on cancer control in patients is unknown.

CTLA-4 is a member of the B7/CD28 family and is constitutively expressed on regulatory T cells, as well as upregulated on other T cell subsets upon activation (370). CTLA-4 inhibits CD28 signalling (371), and T cell activation, through the competitive binding of its ligands CD80/CD86 (372). Expression of CTLA-4 has been reported in both activated mouse and human NK cells (355, 373, 374). In healthy human donors, CTLA-4 expression on NK cell populations is low, and associated with decreased production of activating cytokines and an increase in IL-10 production (355). CTLA-4 expression was found on mediastinal lymph node-derived NK cells of NSCLC patients (375) and on tumor infiltrating NK cells in early-stage lung cancer (376).

The exact role of CTLA-4 expressing NK cells has not yet been fully elucidated. Nonetheless, NK cells have been shown to play a role in the success of anti-CTLA-4 therapies. In melanoma, patients’ response to ipilimumab correlated with an activated NK cell signature (377). Further, ipilimumab was shown to directly bind NK cells through immunofluorescence microscopy. Along with directly binding ipilimumab, NK cells further target CTLA-4 expressing Tregs in vivo and target them through ADCC (378, 379).

Monoclonal antibodies against CTLA-4 are currently in the clinic both alone and in combination with other checkpoint inhibitors. Ipilimumab was first approved as a single-therapy treatment in melanoma (380) and has since been approved as a combination treatment with nivolumab in several cancer types, including colorectal (381), liver (382), renal cell (381), and lung cancer (383). Tremelimumab has recently been approved in combination with durvalumab for the treatment of patients with unresectable liver cancer (384).

Altogether, the immunologic checkpoints that impact T cells may also inhibit NK cells (and be rescued by immune checkpoint blockade). Whether these classical checkpoints are effective targets, or other mechanisms of NK cell inhibition would enable superior anticancer activity remains to be defined.