In healthy human blood and lungs, NK cells make up around 10% of the population of lymphocytes (25, 26). NK cells are categorized based on the level of expression of CD56 (bright (br) and dim) and CD16 and include two broad subsets: CD56^br/dimCD16⁻ and CD56^dim CD16⁺ (27). We will refer to these as CD16⁻ and CD16⁺ NK cells, respectively, throughout this review as it can be difficult to distinguish between CD56^br and CD56^dim subsets in lung tissue – particularly in lung tumors (28, 29). CD16⁻ NK cells are less differentiated and have low cytolytic ability, but produce greater amounts of IFNγ and TNFα than their CD16⁺ counterparts (30, 31). Conversely, CD16⁺ NK cells are more differentiated, with high cytolytic activity (30, 31) ( Figure 1 ). A key difference between the two subsets in the blood is that CD16⁻ NK cells lack the expression of KIRs (31, 32). NK cell expression of KIRs is acquired during their maturation, in a process known as NK cell education or licensing, and expression of KIRs is vital for the cytolytic activity of NK cells (33–35). However, expression of KIRs by CD16⁻ NK cells does occur in some tissues, such as the lung (21, 36).

The lung microenvironment regulates resident immune cells at homeostasis to prevent unwanted activation by harmless antigens. Alveolar macrophages are a prime example of this; these cells are regulated by anti-inflammatory soluble factors such as TGFβ and IL-10, and by ligation of inhibitory receptors such as CD200R and SIRPα (37). The lung NK cell population as a whole (including circulating and tissue-resident subsets) in both humans and mice similarly displays hypo-functional features in the healthy state compared to NK cells from the blood or other organs (38–41). In vitro, human lung-derived NK cells exhibit 25-fold lower cytotoxicity against K562 cells (a cell line particularly sensitive to NK-cell-mediated lysis) (39) and decreased degranulation and ADCC responses compared to blood-derived NK cells (38). In mice, lung NK cells also show decreased cytotoxicity and degranulation versus NK cells isolated from the spleen (40), and express higher levels of the inhibitory receptor NKG2A and lower levels of activating receptors NKp44 and NKG2D (41). Therefore, lung NK cells are thought to be more tightly regulated with a higher threshold for activation than NK cells in other tissues.

In the blood (where they are known as circulating NK (cNK) cells) approximately 10% of NK cells are CD16⁻ and 90% CD16⁺ (31). The CD16⁻ compartment can also contain CD16⁺ NK cells that transiently lose expression of CD16 following activation, when CD16 is cleaved from the cell surface (42–45). In the lung, the distribution of CD16⁺ to CD16⁻ NK cells is similar to that in the blood (38, 46); however, it does vary, for example in the liver and gut, where the CD16⁻ population predominates (25, 26). In addition, recent studies show the presence of possible ‘resident’ NK cells in certain tissues (particularly the liver) on the basis of the expression of the markers CD69, CD103 (αE integrin) and CD49a (α1 integrin), which cause their retention (20, 47, 48) ( Figure 1 ). CD69 interacts with sphingosine-1-phosphate receptor 1 (S1PR1) to promote its degradation; as a result, S1PR1 surface expression is decreased, causing a reduction in chemotactic cues to the blood where high levels of S1P are present. CD49a and CD103 have specific ligands: CD103 binds to E-cadherin, which is expressed by epithelial cells; and CD49a binds to collagen IV, a component of the ECM. For simplicity, we will refer to these cells as tissue-resident NK (trNK) cells in this review, but cells with this phenotype have also been described (in various tissues) as ILC1s (49–51), CD103⁺ ILC1s (50) and intra-epithelial ILC1s (ieILC1s) (2, 50, 52, 53). In the placenta there is also a population of NK cells that expresses CD49a and CD103 described as decidual NK cells (dNK cells) (2, 54). dNK cells have unique functions that are crucial for successful pregnancy which include maintaining maternal-fetal immune tolerance and promoting uterine spiral artery remodeling; the latter being necessary for allowing sufficient blood flow to both the placenta and the developing fetus (2, 54) [reviewed by Liu et al. (55)]. In both humans and mice, and between different organs, the distinction between trNK cells and ILC1s/ieILC1s is blurred, with few clear markers to distinguish these cell types and no clear differences in function [reviewed by O’Sullivan 2019 (56), and Peng and Tian 2017 (57)]. We have summarized the similarities and differences between trNK cells, ieILC1s and ILC1s in the lung specifically in Table 1 .

In the lung, trNK cells comprise 10–25% of the total NK cell population, with the overwhelming majority of trNK cells being CD16⁻ (46, 48, 58, 59). As the majority of the NK cells in the lung are circulating rather than tissue-resident, this provides an explanation for the predominance of CD16⁺ NK cells in this organ, whereas other tissues — namely the liver, skin and secondary lymphoid organs — have a larger population of CD16⁻ trNK cells (48). A caveat in the identification of trNK cells in the human lung is that all studies (21, 46, 58) use lung tissue obtained from patients with lung cancer (tissue that is distal from tumors). trNK cells (described as ieILC1-like cells) are also reported at a higher frequency in lung tumor regions than in adjacent non-cancerous lung tissue (2). A population of trNK cells is also present at low frequencies in the blood, although the frequency of these cells was increased in lung cancer patients compared to controls (21).

Whether lung trNK cells are truly resident remains unclear. In terms of their function, one study which separated CD16⁻ trNKs into CD56^br and CD56^dim trNK cells found that CD16⁻ CD56^br trNK cells have higher expression of the degranulation marker CD107a than CD16⁻ CD56^br cNK cells in response to influenza A virus in a human lung explant model (46). However, this difference is not observed between CD16⁻ CD56^dim trNK cells and CD16⁻ CD56^dim cNK cells. Additionally, no difference is observed in CD107a expression or production of granzyme B, IFNγ or TNFα between trNK cells and cNK cells in response to PMA and ionomycin (46, 58); however trNK do produce increased levels of GM-CSF and show decreased perforin expression (58). In addition, a sub-population of trNK cells described as adaptive-like trNK cells has recently been described in the human lung, characterized by expression of CD49a, NKG2C and KIRs (21). In line with their memory-like phenotype, these cells display greater cytokine production and cytotoxicity than CD49a⁺ NKG2C^- trNK cells (21). The discovery of these cells adds further confusion to the nomenclature surrounding trNK cells, and more research is needed to determine the similarities and differences between trNK cells and memory-like NK cells.

Differences are observed at the transcriptional level in lung trNK cells (in this study defined as CD69⁺CD49a⁺ and/or CD103⁺) compared to CD69⁺ single-positive and CD69-negative NK cells, with higher expression of ID3, IRF4 and RBPJ in the former (58). Genes involved in tissue retention such as RGS1, RGS2 and ZNF683 (Hobit) are also upregulated in the trNK cells compared to CD69⁻ NK cells. Different tissue resident immune cell subsets share some features, for example, lung trNK cells express high levels of CXCR6 and RGS1, and have reduced levels of expression of SELL, S1PR5 and FGFBP2 similar to lung and spleen CD8⁺ tissue-resident memory T cells (TRM) (58, 69). Liver trNK cells also express high levels of CXCR6 at both the mRNA and protein level (20) and produce high levels of IFNγ, TNFα and GM-CSF but lower levels of perforin upon stimulation; similar to trNK cells in the lung (70, 71).

Comparison of human NK cell subsets with those in mice are difficult as murine NK cells do not express CD56, but are instead identified using NK1.1 and/or CD49b (DX5), and with maturity determined using the expression of CD11b and CD27 ( Figure 1 ) (72, 73). Fortunately, most activating and inhibitory receptors are conserved between human and mouse; for example, NKp46 and NKG2D (74). CD49a is also used as a marker for tissue residency in mice, however most studies refer to trNK cells as ILC1s as, in mice, the population tends to lack expression of the transcription factor Eomes (75). Despite these differences, parabiosis mouse models show that 80–90% of lung NK cells are cNK cells (4, 76), with the remaining 10–20% presumed to be trNK cells, in agreement with the human studies described above. Further studies are needed to ascertain the functional role of this potential trNK cell population.

In the healthy lung, NK cells are localized to the interstitium rather than to the airways (38). However, NK cells are observed in the bronchoalveolar lavage (BAL) fluid after infection with influenza virus (79), Staphylococcus aureus (80), cytomegalovirus (CMV) (81), and SARS-CoV-2 (82). NK cells are also present in BAL fluid in various non-infectious diseases, such as asthma and chronic obstructive pulmonary disease (COPD) (47).

Despite an apparent hypo-functionality at homeostasis, NK cells respond and activate quickly in response to pulmonary infections. In humans, rare genetic disorders affecting NK cell function are associated with increased respiratory infections, particularly viral infections (13, 83). As ‘poised effector’ cells, NK cells are early responders to infection and animal models show that substantial numbers are recruited to the lungs 2–3 days after infection with influenza virus, S. aureus and Klebsiella pneumoniae (41, 84, 85). The increase is thought to be due to NK cell recruitment rather than local proliferation, as the numbers of splenic and circulating NK cells show concomitant decreases following infection (41, 86). This increase is transient, as levels return to normal by day 6–9 after infection (85, 87). Similarly in SARS-CoV-2 infection in humans, peripheral NK cells are reduced during acute infection (88–90) and return to normal levels as the infection is cleared; however, in patients with severe COVID-19 the NK cell count remains low at 3 weeks post infection (91).

Activation of NK cells in the lung is facilitated by the upregulation of DAMPs, which are often ligands of NK cell activating receptors, particularly NKG2D (92). For example, Mycobacterium tuberculosis-infected monocytes and macrophages upregulate the NKG2D ligand ULBP1, and blocking ULBP1 prevents NK cell-mediated lysis of infected mononuclear cells (93). Airway epithelial cells also express the NKG2D ligands MICA and MICB (referred to as MICA/B hereafter) and ULBPs under stress (94). In addition, NK cells bind directly to viral proteins expressed on the surface of infected cells, such as the haemagglutinin (HA) glycoprotein of influenza virus, which is recognized by the activating receptor NKp46 (95). Once activated, NK cells produce pro-inflammatory cytokines including TNFα and IFNγ. The latter has a number of important roles, including induction of interferon-stimulated genes in nearby cells (96), recruitment of other immune cells to the site of infection, and aiding the activity of CD8⁺ T cells (34, 83). In addition to cytokine production, activated NK cells also release perforin and granzymes, which induce apoptosis in targeted cells ( Figure 2A ). NK cells can also kill virally infected cells through death-receptor-mediated pathways; for example, influenza A induces the expression of TRAIL on NK cells, and blocking TRAIL in vivo results in reduced viral clearance (97).

Animal models show that lung NK cells have both protective and damaging roles in infection. In influenza virus and S. aureus infection, mice that lack NK cells (Ncr ^−/−, Il15 ^−/−, or depleted using anti-NK1.1 antibodies) show a higher lung viral or bacterial burden at the peak of infection and a longer infection period (46, 80, 86). However, NK cells can also cause harm by contributing to inflammatory-mediated damage, particularly with high dose infections (85–87, 98) [reviewed by Frank and Paust 2020 (99)]. In a more regulatory role, in some tissues NK cells can also indirectly inhibit viral clearance by promoting the apoptosis of activated CD4⁺ T cells, which in turn reduces CD8⁺ T cell numbers ( Figure 2B ). However, in the same way, decreasing CD8⁺ T cell numbers can also prevent CD8⁺ T cell-mediated immunopathology ( Figure 2C ) (100). Therefore, timely induction of NK cell activation — but also contraction of their responses — is optimal for virus containment. Impaired NK cell responses occur in aged mice, which show decreased numbers in the lung following infection and reduced levels of the activation marker CD69 (101). Clearly, NK cells are important in the lung, but their impact on the outcome of an infection in mice is affected by factors such as viral titre and the strain and age of the mouse (85, 98, 101).

NK cell activity will alter depending on the physical and chemical properties of the tissue. These changes in tissue architecture are particularly evident in fibrotic lung diseases associated with chronic inflammation, and alterations in NK cell function are seen in a number of fibrotic lung diseases. For example in idiopathic pulmonary fibrosis (IPF), decreased expression of the activating receptor NKG2D on NK cells (and NKT cells and γδ T cells) isolated from the BAL fluid has been documented, indicating a decrease in NK cell functionality in IPF (115). In addition, NK cells are reduced in the blood of IPF patients (116). In COPD, which also contains a fibrotic element, lung NK cells have increased cytotoxic activity in comparison to NK cells from similar patients without COPD, and may actually drive destruction of the epithelium in COPD (113, 117). In a mouse model of pulmonary fibrosis (bleomycin-induced), the disease is more severe in mice where CXCR3-dependent immune cell (particularly NK cell) recruitment is reduced, resulting in increased mortality in Cxcr3^-/- animals (118). Indeed, IFNγ was shown to have a protective effect in this model (118), implying that NK cells as a key source of IFNγ could have a beneficial role in pulmonary fibrosis, in contrast to the studies described above in COPD. This is an understudied area - NK cells can have negative and positive consequences in different fibrotic lung diseases.

Abundant NK cells or a dominant NK cell gene signature within a tumor correlates with better overall survival in a number of cancers, including lung cancer [reviewed by Larsen, Gao and Basse (119)], but the prognostic importance of NK cells in lung cancer is unclear. Some studies link high tumor NK cell infiltration with increased survival (120, 121), whereas others show no correlation (29, 122). Early studies grouped NK cells, NKT cells, γδ T cells and/or ILC1s together [reviewed by Habif et al. (122)] which has now been largely overcome by using antibodies to NKp30 (36) or NKp46 (25, 29), or by co-staining of CD45⁺ CD3⁻ CD56⁺ cells (122). Discrepancies also arise owing to the different stages of tumor development examined and the precise microenvironment generated (78). Regardless of their prognostic significance, it is clear by both immunohistochemistry and flow cytometry that lung tumors have low NK cell infiltration (25, 28, 29, 36, 122–124). Additionally, NK cells tend to locate at the tumor edge, or in the tumor stroma, indicating a possible defect in their recruitment (25, 29, 36, 122). Factors affecting NK cell recruitment to both tumors and sites of infection include chemokines (125, 126), extracellular matrix (127, 128) and soluble mediators (129, 130), and these will be discussed in the next section.

As with viral infection, animal models of lung cancer indicate that the timing of NK cell infiltration may be crucial for their anti-tumor activity. In a mouse model of lung cancer, depletion of NK cells at early stages of cancer — but not at later stages — accelerated tumor growth (131). Furthermore, as tumor growth progresses, both mouse and human NK cells exhibit decreased expression of the degranulation marker CD107a (29, 36, 131), as well as decreased IFNγ (29, 131) and perforin (36, 131) ( Figure 3 ). In mouse studies only, tumor NK cells have decreased expression of granzyme B and reduced proliferation (131) ( Figure 3 ). In lung adenocarcinomas, as well as in other solid tumors, intratumoral NK cells are predominantly CD16⁻, rather than the more cytotoxic CD16⁺ subset seen in the neighbouring healthy margin tissue (29, 36, 122–124). Furthermore, downregulation of activating receptors, including NKG2D [as is also seen in IPF (115)], and upregulation of the inhibitory receptor NKG2A, reduce anti-tumor potency (28, 29, 122) ( Figure 3 ). As mentioned previously, CD16 is enzymatically cleaved from NK cells after receptor engagement and activation (42, 43) and re-expression can be delayed (44, 45) meaning that NK cells in lung tumors may be ‘ex-CD16⁺’ rather than bona fide CD16⁻ NK cells. NK cells isolated from lung tumors (as well as matched blood and lung margin) have also been shown to produce tumor-promoting angiogenic factors, including vascular endothelial growth factor (VEGF), placental growth factor (PlGF) and IL-8 ( Figure 3 ) (123). Although there was no significant difference in the expression of these factors between tumor, lung margin and blood NK cells, it is interesting to note that CD16⁻ NK cells – which are present in greater abundance in lung tumors – produce far greater amounts than CD16⁺ NK cells (123).

This population of CD16⁻ NK cells in lung tumors may also comprise trNK cells, which, in the lung, are predominantly CD16⁻ (58). To support this idea, tumor NK cells express higher levels of CD69 compared to unaffected lung or blood NK cells (29, 36, 132); a marker highly expressed on trNK cells (38, 46, 58). Indeed, a population resembling trNKs (classified as ieILC1s) is apparent in human lung tumor tissue (2). Further characterization of tissue residency markers is required to determine what contribution trNK cells make to the intratumoral NK cell compartment.

cNK cells express a broad repertoire of chemokine receptors depending on their differentiation stage (25, 133) ( Table 2 ), providing one explanation for the variation in the ratio of CD16⁻ to CD16⁺ NK cells in different tissues. For example, CD16⁻ NK cells express CCR7 (and CD62L) which promote recruitment to secondary lymphoid tissues (25, 133, 134). Furthermore, organ-specific chemokine patterns exist in the steady state and in disease (135).

Although CXCR3 can be expressed by both CD16⁻ and CD16⁺ NK cells, it is predominantly expressed by CD16⁻ NK cells [~10% of CD16⁺ NK cells express CXCR3 vs ~90% of CD16⁻ NK cells (136)] and therefore is more important for mediating their recruitment (25, 136). At steady state, Cxcr3 ^−/− mice have decreased NK cells in the lungs, liver, blood and lymph nodes (118, 137) and fewer NK cells are recruited during lung infection (125, 138). In the peripheral blood of influenza and COVID-19 patients, fewer CXCR3⁺ CD16- NK cells are present (136), presumably because these cells have been recruited to the lung. In accordance with this, the expression of CXCL10 (a ligand for CXCR3) is increased in the BAL fluid of SARS-CoV-2 infected patients, and NK cells isolated from the BAL fluid of COVID-19 patients also express high levels of CXCR3 (82, 136). CCR5 is also implicated in NK cell recruitment to the lung, although to a lesser extent than CXCR3. Ccr5 ^−/− mice only show a small reduction in pulmonary NK cell numbers after influenza infection (125). CCL5 expression is also increased in the BAL fluid of COVID-19 patients (136).

CXCR3 and its ligands CXCL9-10 also mediate the recruitment of NK cells into tumors. In subcutaneous and pulmonary mouse tumor models (RMA-S, a lymphoma tumor cell line expressing low levels of class I MHC; B16 melanoma cell line transduced to express the NKG2D ligand RAE-1ϵ; and pulmonary tumor cell line 3LL-Luc2), decreased intra-tumoral NK cells are observed in Cxcr3^-/- mice and in mice treated with anti-CXCR3 antibodies; but despite this, no difference in survival is observed (65, 126). However, subcutaneous injection of tumor cells transduced to overexpress CXCL10, which results in increased NK cell tumor infiltration, does increase overall survival as compared to control cell injection (106). Tumor metastasis to the lung is facilitated by total NK cell depletion, but not by anti-CXCR3 antibodies, implying a role for other chemokine receptors (65). For example, IL-33 can drive the production of CCL5 from eosinophils and CD8⁺ T cells, which significantly increases survival and decreases lung metastases in an NK cell-dependent manner (139).

Chemokine production in infected or malignant tissue is also affected by other soluble factors. For example, IFNγ increases the production of the CXCR3 ligands CXCL9-11 by tumor cells, and correlates with an increase in NK cells in subcutaneous tumors and improved survival in mice (126). However, the immune modulator and tumor-promoting factor prostaglandin E2 (PGE2) inhibits IFNγ-induced secretion of CXCR3 ligands from breast cancer cell lines (140). Similarly, IFNγ-matured dendritic cells (DCs) exhibit reduced production of CXCL10, CCL5 and CCL19 in the presence of PGE2, which results in reduced NK cell migration in vitro (141).

In some instances, it is not NK cell recruitment that is the problem, but the recruitment of inappropriate NK cell subsets. For example, increased levels of CCL19, CXCL9 and CXCL10 in lung tumors may preferentially recruit CD16−NK cells, and reduced levels of CXCL2 may decrease recruitment of more cytotoxic CD16⁺ NK cells ( Figure 4A ) (25). These CD16⁻ NK cells may also be inappropriately retained via, for example, reduced expression of S1PR1 that is needed for tissue egress, and increased expression of CXCR6, promoting retention (132).

Clearly, therapeutic targeting of specific chemokine axes may facilitate the recruitment of cytotoxic CD16⁺ NK cells to tumors, but the potential for off-target effects remains a caveat. Genetic modulation of chemokine receptors is a current field of research for NK cell-based therapies, and provides a more targeted method of altering the chemokine axes of NK cells (142, 143). However, altered chemokine axes cannot explain all defects in the recruitment of NK cells, as in some cancers CX3CL1 (the ligand for CX3CR1) expression is high but CD16⁺ NK cells are not present, despite expressing CX3CR1 (144).

Alteration of the extracellular matrix (ECM) occurs in viral infections (145), and in cancer (78) and may affect NK cell recruitment, retention and activation.

In both animal tumor models and human cancers, tumors can be surrounded by ECM, with some encapsulated in laminin or collagen ( Figure 4B ) (146–148). Indeed, lung tumors are surrounded by type I collagen (149), and small cell lung cancers (SCLC) and their metastases are enveloped in a dense ECM composed of fibronectin, laminin, collagen IV and tenascin C (150). Additionally, tumor resistance to PD-1/PD-L1 blockade is associated with increased collagen deposition (151). It has also been observed in influenza mouse models that ECM alterations, which are observed in acute infection, can persist well after viral clearance (145). To migrate through tissues, particularly those with dense ECM, many cells (including NK cells) release proteases such as MMPs (152). In the Lewis lung carcinoma (LLC) mouse model, tumors with more altered structural properties have less NK cell infiltration compared to tumors with a high NKp46, IFNγ and, somewhat controversially, fibronectin signature (127). Furthermore, Ifnγ ^-/- mice have primary tumors with a more aggressive, ECM-enriched phenotype, and increased metastases (127). Fibronectin and other ECM proteins are generally associated with more mesenchymal, invasive tumors. However, the presence of fibronectin in tumors can be beneficial if accompanied by a decrease in other typical epithelial-to-mesenchymal transition (EMT) markers including keratin, vimentin and N-cadherin (127). It should also be noted, that cross-linking and density of tumor ECM influences NK cell recruitment with ‘loose’ facilitating and ‘compact’ inhibiting (128). The role of ECM in NK cell recruitment to lung tumors, and indeed in recruitment to the fibrotic lung, is an emerging area of research.

The soluble mediators TGFβ and PGE2 inhibit immunity in the healthy lung but also impair immunity in the post-viral lung and in lung cancer (37, 78).

TGFβ has a number of immunoregulatory effects on NK cells in vitro including a decrease of activating receptors NKG2D and NKp30; a reduction in IL-15- or IL-2-induced NK cell cytotoxicity, proliferation, degranulation and granzyme B expression; a reduction of CD16-induced production of IFNγ; a decrease in glycolysis; and enhanced expression of the pro-angiogenic factors VEGF and PlGF ( Figure 4 ) (10, 11, 123, 131, 153–155). With prolonged treatment, TGFβ also reduces NK cell-dependent ADCC responses (156). In mouse models of lung metastasis (B16 melanoma model, which spontaneously metastasizes to the lung), TGFβ promotes the expansion of an ‘intermediate ILC1’/’ex-NK cell’ population, associated with increased expression of inhibitory receptors, reduced cytotoxicity, and expression of the trNK-associated marker CD49a ( Figure 4C ). Interestingly, this population increases with tumor burden, as do the levels of TGFβ in the TME (157–159). In vitro studies with human blood-derived NK cells have also shown that TGFβ alone or in combination with IL-15 can promote the development of a trNK cell-like phenotype (53, 129, 153). However, unlike the studies described above in mice, these cells were highly cytotoxic (53). TGFβ (alone, or in combination with hypoxia and a demethylating agent) has also been shown to promote cNK cell acquisition of a dNK cell-like phenotype in vitro (160, 161). However in these studies, only the combination treatment resulted in a decrease in NK cell cytotoxicity (161). In the TME more broadly, TGFβ affects the chemokine milieu (78). For example, in a murine model of lung metastasis TGFβ suppressed the production of CXCL1 and CXCL5 by tumor cells (162) – chemokines associated with recruitment of CD16⁺ NK cells (36, 133).

PGE2 is a potent immune modulator produced during inflammation that, as mentioned previously, indirectly affects NK cells by reducing the levels of chemokines necessary for their recruitment (140, 141). PGE2 also directly suppresses NK cell proliferation, cytokine secretion, and cytotoxicity ( Figure 4C ) (141, 163, 164). In mouse tumor models (B16 melanoma and MC38 colorectal), a lack of the prostaglandin receptors EP2 and EP4 specifically on granzyme B⁺ cells – predominantly NK cells – causes tumor regression (165). PGE2 reduces the expression of the NK cell-derived chemokines CCL5 and XCL1 ( Figure 4C ), thereby decreasing the recruitment of conventional type 1 DCs to tumors (130, 166). In human immune cell datasets, XCL1 and XCL2 (homologues in humans for XCL1 in mice) are most highly expressed by CD16⁺ NK cells — the cells that are missing from the lung TME (130).

The ability of NK cells to become activated depends on the expression of cell-surface receptors, for example, NKG2D (167, 168). However, activating receptors can be internalized or cleaved leading to cancer and viral escape (169–171). For example, MMPs and A disintegrin and metalloproteinases (ADAMs), which are often present in the TME (172) and the post-viral lung (173), cleave NKG2D ligands (NKG2DLs) from the cell surface of tumor cells or infected cells, which leads to the evasion of NK cell-mediated killing (114, 174–176). Increased levels of soluble NKG2DLs are observed in the TME and serum of patients with cancer, including lung cancer (177), which correlate with decreased overall survival (177–181). Similarly in IPF, increased levels of soluble MICA are observed in the plasma, and polymorphisms in MICA are associated with increased risk for the development of IPF (115). Soluble NKG2DLs inhibit both CD8⁺ T cells and NK cells as they block ligation of membrane-bound NKG2DLs (180, 182, 183) and cause downregulation of NKG2D (182, 183). NK cells with reduced membrane expression of NKG2D exhibit impaired lytic ability (183, 184), and mice that lack NKG2D are more susceptible to tumors (185). In various mouse models, prevention of tumor cell shedding of NKG2DLs increases the anti-tumor activity of NK cells and decreases the number of tumor metastases (175, 186). Additionally, combining anti-soluble MIC treatment with PD-L1 blockade improves survival in mice (186), and therefore highlights a potential future therapeutic opportunity for human cancers – particularly those with high serum levels of soluble NKG2DLs.

Another feature that influences NK cell activity is hypoxia, which occurs when the growth of tumor cells outstrips their supply of oxygen. In addition, the dense nature of the ECM around and within tumors can act as a barrier to critical metabolites (148). Hypoxia is often accompanied by lower pH and glucose levels (172). In vitro, hypoxia alone (1% O₂) does not significantly affect NK cell cytotoxicity in comparison to normoxia (around 6% O₂) (187–189). However, anoxia (0%) decreases NK cell-specific lysis of target cell lines in vitro (189). Hypoxic NK cells lose the ability to upregulate various activating receptors and produce less inflammatory cytokines in response to stimuli (190, 191); however, they can produce VEGF to promote angiogenesis (129). Additionally, indirect NK cell inhibition also occurs in hypoxic conditions, as hypoxia is associated with an increase in the shedding of NKG2DLs from cancer cells (192). Decreased pH levels also suppress NK cell cytotoxic activity and cytokine production (189, 193). A profound decrease in NK cell cytotoxicity occurs with a combination of low O₂, pH and glucose in vitro, mirroring the conditions of the TME (189).

In tissues – whether they be trNK cells or recruited cNK cells – NK cells interact with many different immune cells (194), epithelial and endothelial cells (94, 117, 171). Lung trNK cells express different integrins to cNK cells (46, 58), therefore different NK cell subsets likely have different positions within the lung tissue and hence different cell types that they interact with.

NK cell interactions with DCs are well studied (39, 40, 46, 117). Some interactions enhance NK cell activity, such as the interaction of CD16⁻ NK cells with immature DCs (iDCs) during infection. These iDCs secrete a number of cytokines, including IL-12, IL-18, IL-15 and type I IFNs, which promote NK cell maturation, activation and memory-like responses ( Figure 5A ) (117, 194–197). There are also contact-dependent mechanisms of DC-mediated NK cell maturation ( Figure 5A ) (195, 197). Such interactions are often reciprocal: once primed by iDC interaction, the activated NK cell then either induces killing of the iDC or its maturation (196, 197). Factors affecting this outcome are shown in Figure 5B (34, 194, 198). Environmental factors change the interaction between NK cells and DCs, for example cigarette smoke exposure in mice enhances NK cell cytotoxicity in a DC-dependent manner (117). A similar priming effect of DCs on NK cells is observed in COPD (117) that culminates in enhanced NK cell-mediated destruction of lung epithelial cells (113, 117). However, much less is known about NK cell-DC interactions in the TME.

In addition to DCs, NK cells are known to interact with macrophages. In pulmonary infection, macrophages increase their surface expression of NKG2DLs, in some cases rendering them susceptible to NK cell-mediated lysis (94, 199) [reviewed by (Stojanovic, Correia and Cerwenka, 2018 92)]. Tumor macrophages can also display increased levels of NKG2DLs (92, 200, 201), which are postulated to act on NK cells in a similar way to soluble NKG2DLs – in that low-level engagement of NKG2D promotes receptor internalization and desensitization of NK cells (92, 170). As with soluble NKG2DLs in serum, elevated expression of NKG2DLs is also observed on circulating monocytes in some cancers (201). Interestingly, this expression decreases following tumor resection surgery (201). In vitro studies show that the hypoxia-related factor lactate dehydrogenase 5 (LDH5) is able to drive expression of NKG2DLs on monocytes (201); hypoxia may therefore affect NK cell–macrophage communication. Additionally, tumor- or metastasis-associated-macrophages can express membrane-bound TGFβ (66), which will likely have similar inhibitory effects on NK cells as those described earlier for soluble TGFβ.